# self.md > Personal AI OS guides, skills, tools, memory patterns, and sovereign receipts. self.md documents how people build AI-powered personal operating systems with plain-text Markdown as the protocol layer, local runtimes, and verifiable boundaries. ## Core Architecture & Concepts - [What is a Personal AI OS?](https://self.md/concepts/what-is-personal-os/): Foundations of sovereign personal operating systems. - [Architecture Patterns](https://self.md/concepts/architecture/): Layers, loops, custody, boundaries, and receipts. - [Prompt Injection Security](https://self.md/concepts/prompt-injection/): Securing local and agent systems from malicious context. - [Agent Memory Systems](https://self.md/concepts/agent-memory-systems/): Explicit memory layers, retrieval, and consolidation boundaries. ## Tool Registry & Control Surfaces - [Personal AI OS Map](https://self.md/tools/personal-ai-os/): Complete registry and tier matrix of personal AI runtimes. - [AI Coding Assistants](https://self.md/tools/ai-coding-assistants/): Claude Code, Cursor, Aider, Open Interpreter. - [Browser Automation](https://self.md/tools/browser-automation/): Browser-use, Stagehand, Playwright agents. ## Practical Step-by-Step Guides - [Proactive AI Agent with Semantic Memory](https://self.md/guides/proactive-agent/): Turn your AI from reactive chatbot into proactive assistant using heartbeats, semantic search, and persistent memory. - [Customizing the Status Line](https://self.md/guides/status-line/): Configure Claude Code's status line to display model, git status, tokens, and custom information - [Git Worktrees for Parallel Agents](https://self.md/guides/git-worktrees/): Run multiple AI agents on the same codebase without conflicts using git worktrees - [Claude Code for Non-Developers](https://self.md/guides/non-developers/): How product managers, analysts, and writers use Claude Code without writing a single line of code - [Hooks System in Claude Code](https://self.md/guides/hooks-system/): Automate workflows with deterministic triggers that run shell commands at specific points in Claude Code's lifecycle. - [Sandboxing & Security for AI Agents](https://self.md/guides/sandboxing-security/): How to isolate AI agents using OS-level sandboxing to prevent unauthorized access and reduce permission fatigue. - [Running Claude Code in Containers](https://self.md/guides/container-execution/): Isolate agent execution with Docker for security, scalability, and 24/7 operation - [Checkpointing - Safe Rollback for Agent Work](https://self.md/guides/checkpointing/): How automatic checkpoints protect your work and let you recover from agent mistakes - [Claude Code Plugins System](https://self.md/guides/plugins-system/): Extend Claude Code with plugins for custom tools, MCP servers, and workflows - [Build a readable AI quota dashboard inside Telegram](https://self.md/guides/telegram-ai-quota-dashboard/): Build an AI quota dashboard for Codex, Claude and Grok with an English renderer, reusable agent skill, static web view and verified Telegram photo delivery. - [Agent Verification Loop](https://self.md/guides/agent-verification-loop/): Seven checks to catch fake success before an agent writes, sends, deploys, deletes, or publishes. For coding and publishing agents. - [Claude Code Workflow Files: AGENTS.md, CLAUDE.md, Skills, and Repo Maps](https://self.md/guides/claude-code-workflow-files/): where each agent instruction belongs in a Claude Code repo — CLAUDE.md, skills, rules, hooks, MCP. one decision table, one cleanup pass. - [Claude Code Session Preflight](https://self.md/guides/claude-code-session-preflight/): Ten read-only checks to run before a serious Claude Code session: version, model, permissions, MCP servers, context, and rollback. - [best skills for productivity with claude code](https://self.md/guides/best-skills-productivity/): the skills that actually matter when you're trying to get shit done with an AI assistant - [claude code for designers: from figma to functioning frontend](https://self.md/guides/claude-code-for-designers/): how designers can use AI to build what they envision without waiting for engineering sprints - [claude code for home automation: beyond 'alexa turn on the lights'](https://self.md/guides/claude-code-home-assistant/): using AI to build actually intelligent home automation that doesn't make you want to throw your smart bulbs out the window - [claude code for project management: when your AI understands your backlog better than you do](https://self.md/guides/claude-code-project-management/): using MCP integrations to make Linear and Jira actually useful instead of just mandatory - [claude code for students: learning to code without the bullshit](https://self.md/guides/claude-code-for-students/): how to use AI to actually learn programming instead of just generating homework answers - [CrewAI vs LangGraph: choose a multi-agent framework by workflow shape](https://self.md/guides/crewai-vs-langgraph/): CrewAI fits role/task crews and event-driven flows. LangGraph fits explicit state, loops, interrupts, persistence, and replay. Use neither when one agent and a few tools is enough. - [agent-first design](https://self.md/guides/agent-first-design/): how to structure repositories so AI agents can actually work in them. the code isn't just for humans anymore. - [AGENTS.md vs CLAUDE.md vs Cursor/Windsurf/Copilot: where repo instructions belong](https://self.md/guides/agent-config-files-compared/): one source of truth for repo instructions. for multi-agent repos, AGENTS.md should be policy; vendor files are adapters, not competing rulebooks. - [best claude code plugins: the minimum viable stack](https://self.md/guides/best-claude-code-plugins/): choose fewer, better Claude Code plugins. repo instructions first, one MCP server second, then plugins only for repeated jobs. selector matrix included. - [best claude code skills for developers](https://self.md/guides/best-skills-developers/): the skills that actually make Claude Code useful for software development — code review, TDD, debugging, and the ones most people miss. - [best claude code skills for writers](https://self.md/guides/best-skills-writers/): claude code skills that actually help writers — content creation, editing, documentation, and the ones worth installing. - [Best MCP servers by job: browser, GitHub, files, docs, fetch](https://self.md/guides/best-mcp-servers/): A practical MCP server shortlist for Claude Code: job, trust level, permission boundary, install source, and validation check. - [claude code + docker complete guide](https://self.md/guides/claude-code-docker/): running Claude Code inside Docker containers — sandboxing, reproducible environments, and keeping your machine clean from agentic chaos. - [claude code + github actions CI/CD](https://self.md/guides/claude-code-github-actions/): set up Claude Code in GitHub Actions for automated code review, PR feedback, and CI workflows. the official action, custom setups, and what's actually reliable. - [claude code + notion integration](https://self.md/guides/claude-code-notion/): how to connect Claude Code to Notion via MCP — setup, real use cases, and whether it's worth the hassle. - [claude code + obsidian workflow](https://self.md/guides/claude-code-obsidian/): connecting your personal knowledge management to your AI coding agent. obsidian vaults as context, claude code as engine. - [claude code + playwright browser automation](https://self.md/guides/claude-code-playwright/): connect Claude Code to Playwright via MCP for browser automation, e2e testing, and visual debugging. setup, real workflows, and sharp edges. - [claude code context window optimization](https://self.md/guides/context-optimization/): practical strategies for managing your context window in claude code. every wasted token is money and attention you'll never get back. - [Claude Code for non-coders: the complete guide](https://self.md/guides/non-developers-complete/): how to use Claude Code without being a developer — email, research, writing, organizing, and automating your life from the terminal. - [claude code for product managers](https://self.md/guides/claude-code-for-pms/): you don't need to be a developer to use claude code. PRDs, data analysis, spec writing, and competitive research — from your terminal. - [claude code for researchers](https://self.md/guides/claude-code-for-researchers/): using Claude Code for literature review, data analysis, paper writing, and research workflows — what works, what's dangerous, and what's genuinely useful. - [claude code for solopreneurs](https://self.md/guides/claude-code-for-solopreneurs/): you're the developer, the designer, the PM, and the marketing team. here's how claude code lets you ship like a team of five. - [claude code for technical writers](https://self.md/guides/claude-code-for-writers/): using Claude Code for documentation, API docs, READMEs, and technical content — workflows that work and traps to avoid. - [claude code plan mode mastery](https://self.md/guides/plan-mode-guide/): plan mode is the most underused feature in claude code. here's how to actually use it — and when not to. - [claude code pricing: pro vs max vs api in may 2026](https://self.md/guides/claude-code-pricing/): updated may 20, 2026. pro includes claude code, max starts from $100, usage credits can kick in after limits, and api use bills by token. compare current prices before you run anything expensive. - [Claude Code tutorial for beginners](https://self.md/guides/claude-code-beginners-tutorial/): learn Claude Code from zero — installation, your first conversation, CLAUDE.md setup, and a practical first project in 15 minutes. - [claude code vs aider](https://self.md/guides/claude-code-vs-aider/): two terminal-first AI coding tools, very different philosophies. here's what matters when choosing between claude code and aider. - [Claude Code vs Cursor: complete comparison 2026](https://self.md/guides/claude-code-vs-cursor/): Claude Code vs Cursor — feature-by-feature comparison of terminal-native AI coding vs IDE-based AI coding in 2026. pricing, architecture, and when to use each. - [Claude Code vs GitHub Copilot](https://self.md/guides/claude-code-vs-copilot/): Claude Code vs GitHub Copilot — architecture, agent modes, pricing, and real differences between terminal-native AI coding and IDE autocomplete. - [claude code vs windsurf](https://self.md/guides/claude-code-vs-windsurf/): terminal agent vs AI-native IDE. claude code and windsurf represent two different bets on how developers will work with AI. - [how to reduce claude code costs by 50%](https://self.md/guides/reduce-claude-code-costs/): practical techniques to cut your claude code spend in half without sacrificing quality. token economics for people who count. - [how to write Claude Code skills](https://self.md/guides/writing-skills-tutorial/): complete guide to writing Claude Code skills — SKILL.md format, frontmatter, testing, publishing, and building a skill from scratch. - [ultrathink: deep reasoning in claude code](https://self.md/guides/ultrathink/): when to use extended thinking in claude code, what it actually costs, and why you shouldn't use it for everything. - [your first MCP server in 10 minutes](https://self.md/guides/first-mcp-server/): build a model context protocol server from scratch in typescript. no boilerplate, no magic — just a working server that claude code can talk to. - [Dayflow: Screen Activity Tracking Without Surveillance](https://self.md/guides/dayflow-setup/): Privacy-first activity tracking with AI. Local screenshots, Gemini recognition, markdown export. No cloud, no tracking, no bullshit. - [Auditing Skills for Injection Vulnerabilities](https://self.md/guides/skill-security-audit/): How to check AI agent skills for prompt injection patterns and assign security badges. - [LLM Logging: Capture Every AI Conversation](https://self.md/guides/llm-logging/): Track prompts, responses, and token usage. Build a searchable archive of LLM interactions for debugging, learning, and prompt optimization. - [Parallel AI Sessions: Run Multiple Agents](https://self.md/guides/parallel-sessions/): Run multiple AI agents simultaneously. Patterns for concurrent prompts, session orchestration, and task distribution across LLM workers. - [Token Efficiency: Fit More in Less](https://self.md/guides/token-efficiency/): Practical techniques to reduce token usage, optimize context windows, and cut LLM costs without losing quality. - [Build a Personal Knowledge Graph](https://self.md/guides/personal-knowledge-graph/): Connect your notes, projects, and contacts in a queryable graph. Entity extraction, Neo4j setup, and Obsidian graph visualization. - [How to Debug AI Agents](https://self.md/guides/agent-debugging/): A practical workflow for finding and fixing failures in AI agents using tracing, logs, and observability tools. - [Multi-Model Routing for LLM Applications](https://self.md/guides/multi-model-routing/): Route requests to the right model based on task complexity. Cut costs 40-70% while maintaining quality where it matters. - [Multi-Agent Content Pipeline with Claude Code](https://self.md/guides/multi-agent-content-pipeline/): Build autonomous content systems using Claude Code skills, semantic search, and chained agents - [Debug Your RAG Pipeline Before Users Notice](https://self.md/guides/rag-observability/): Monitor retrieval-augmented generation systems with OpenTelemetry tracing. Find whether bad answers come from retrieval, context, or generation. - [LLM-as-Judge Evaluation](https://self.md/guides/llm-as-judge/): Use LLMs to evaluate LLM outputs. Build reliable automated judges through critique shadowing and iterative calibration with domain experts. - [MCP Server Stacking: One Data Source is Data, Two is Intelligence](https://self.md/guides/mcp-server-stacking/): Combine multiple MCP servers to unlock cross-domain insights that no single server can provide on its own. - [The Skills System](https://self.md/guides/skills-system/): Create reusable AI capabilities with Claude Code skills - [CLAUDE.md Best Practices](https://self.md/guides/claude-md-guide/): Write effective project instructions that Claude reads at every conversation start - [Browser agents vs Playwright: when to use AI browser automation](https://self.md/guides/browser-agents/): Choose between manual browser work, Playwright scripts, browser agents, Playwright MCP, and hosted browsers. Includes auth rules, receipts, and failure checks. - [Keep Claude Running for Hours](https://self.md/guides/todo-driven-loops/): Build autonomous development loops using todo files, hierarchical subagents, and context compaction to process task lists for hours without human input. - [Multi-Agent Coordination: How to Run Three AI Agents Without Merge Conflicts](https://self.md/guides/parallel-ai-agents/): The four-phase workflow for running parallel AI agents on the same codebase using tmux, git worktrees, and a shared AGENTS.md file. - [Build a TIL System That Actually Works](https://self.md/guides/til-system/): Set up a Today I Learned repo in 10 minutes. Low-friction knowledge capture that compounds into a searchable reference. - [From Chat to Cron: Building Automated PKM Pipelines](https://self.md/guides/automated-pkm/): Stop manually organizing notes. Build cron-triggered AI workflows that process, tag, and connect your knowledge base on autopilot. - [llama.cpp setup guide: build, run, and troubleshoot GGUF models locally](https://self.md/guides/local-llms-llama-cpp/): Build llama.cpp, load a GGUF model, run the CLI or server, and verify the install with one smoke test and troubleshooting table. - [Automated Development Journaling with MCP and CLAUDE.md](https://self.md/guides/dev-journaling-mcp/): Connect Claude Code to your note app via MCP, then use CLAUDE.md instructions to create journals automatically as you work - [Task Decomposition for AI Documentation](https://self.md/guides/ai-documentation-workflow/): Break documentation projects into discrete steps to get accurate, consistent output from AI tools instead of hallucinated garbage. - [Auto-Activating Skills in Claude Code](https://self.md/guides/auto-activating-skills/): Configure skills to trigger automatically based on keywords, file patterns, and context instead of manual slash commands. - [ACE Framework for Personal AI](https://self.md/guides/ace-framework/): Implement David Shapiro's six-layer cognitive architecture to give your Claude Code assistant mission, values, and strategic context - [Quick Start: Pick Your Path](https://self.md/guides/getting-started/): Three ways to build a personal OS today. Choose based on your tools and comfort level: Claude Code for developers, GPT + Zapier for quick setup, or Gemini for Google users. - [Custom Slash Commands](https://self.md/guides/custom-commands/): Create reusable workflows triggered by /name in Claude Code - [Voice-First Content Creation](https://self.md/guides/voice-first-content/): Replace typing with talking. Brain dump while walking, transcribe with AI, generate multiple versions, polish the best. From 2 hours to 30 minutes per piece. - [Pixel NAS: Turn a 2016 Pixel into a Google Photos Upload Daemon](https://self.md/guides/pixel-nas-google-photos/): Use an original Pixel as a tiny always-on inbox that syncs photos from other devices and uploads them to Google Photos. - [Building Your First MCP Server](https://self.md/guides/building-mcp-servers/): Create custom MCP servers to extend Claude with your own tools - [Claude Code Setup Guide](https://self.md/guides/claude-code-setup/): Complete Claude Code setup guide: installation, MCP servers, custom CLAUDE.md instructions, shell aliases, and security configuration - [Building a Memory System](https://self.md/guides/memory-system/): Build persistent AI memory with Claude Code using episodic-memory MCP, memory files, and external tools like Obsidian and Notion ## Radar Signals & Dispatches - [Claude's self-hosted worker has a real shell](https://self.md/signals/2026-11-14-claude-s-self-hosted-worker-has-a-real-shell/): Claude Managed Agents adds customer-hosted service and history storage; WiFi4EU's free-service periods end; zd1 brings AirPlay 2 to one tested 2011 Zeppelin Air, with recovery gaps. - [Brave's half-cent search](https://self.md/signals/2026-11-13-brave-s-half-cent-search/): Brave adds USDC payments per search, ZigZag changes embroidery routing and thread cuts, and Internet Archive announces Salon Explorer. - [a denied AWS request can pass locally](https://self.md/signals/2026-11-12-a-denied-aws-request-can-pass-locally/): AWS Emulator excludes security testing; LLM 0.37.2 fixes two custom-endpoint key leaks; Gemini Decisions scores bounded choices with probabilities and qualified latency claims. - [the measurement grid stayed put](https://self.md/signals/2026-11-11-the-measurement-grid-stayed-put/): A scientist catches an unchanged X-ray measurement grid; Muse Ember charges higher rates above 200K prompt tokens; Beep Locator finds an old detector after manual calibration help. - [Browser() now opens a cloud browser](https://self.md/signals/2026-11-10-browser-now-opens-a-cloud-browser/): Browser Use 1.0 changes its default browser location; Fable 5.2's ownership prompt expands edits; a debugging study separates adding pdb from removing shell access. - [the password in the permission rule](https://self.md/signals/2026-11-09-the-password-in-the-permission-rule/): Apiiro finds credentials in Claude Code approval files; a vLLM study traces retry overload; NormGym tests learning hidden local rules from interaction - [the second number in the PDF](https://self.md/signals/2026-11-08-the-second-number-in-the-pdf/): Financial-PDF QA exposes hidden numbers, Datasette removes an HTML prefetch trigger, and decoded.legal retires little-used self-hosted services. - [TikTok's next similar video](https://self.md/signals/2026-11-07-tiktok-s-next-similar-video/): A preliminary EU finding on TikTok's repeated recommendations, Replit's microVM isolation and credential brokers, and Anthropic's measured scientific use versus modeled acceleration - [an account email became a request header](https://self.md/signals/2026-11-06-an-account-email-became-a-request-header/): Claude Code announces removal of account emails from model context; a Lean study compares agent and human edit choices; Kagi opens 50 searches without registration. - [Conductor gives its cloud a deletion date](https://self.md/signals/2026-11-05-conductor-gives-its-cloud-a-deletion-date/): Conductor Cloud sets November shutdown and export dates; Home Assistant absorbs HACS; SlackBench compares message-search tools with direct access to files - [GitBook is back; its tokens are revoked](https://self.md/signals/2026-11-04-gitbook-is-back-its-tokens-are-revoked/): GitBook requires token rotation and MCP re-authentication; Tracker CE changes saved patterns and licensing; unsatz tests models on unseen experiments. - [a rejected tool still ran](https://self.md/signals/2026-11-03-a-rejected-tool-still-ran/): LLM 0.37 fixes skipped tool callbacks and chat logging; a Qwen weight edit reduces loops; Census changes its business survey's AI terminology. - [a shared ChatGPT link reached private chats](https://self.md/signals/2026-11-01-a-shared-chatgpt-link-reached-private-chats/): Parameter's fixed ChatGPT share exploit, DeltaDB's SQLite-backed agent history, and Vitalik Buterin's airplane-mode trial with SMS logins and slow offline answers. - [the scripts may stay after uninstall](https://self.md/signals/2026-10-31-the-scripts-may-stay-after-uninstall/): Cursor identifies malicious Claude Code extension packages and persistent scripts; WebAnything qualifies its verification records; nesformer compiles an NES CPU into model weights. - [Jules goes read-only in November](https://self.md/signals/2026-10-30-jules-goes-read-only-in-november/): Jules sets a November 11 session deadline before its December shutdown; ChatGPT API credits exclude Codex API-key calls; Amazon keeps AI-usage metrics; Tapestries offers zip export. - [the dispatch log disagrees](https://self.md/signals/2026-10-29-the-dispatch-log-disagrees/): Surge AI's tax-agent benchmark reports a 10.8% leading pass rate. MicroZig bars LLM-assisted editing and translation. An anti-bot script obstructs model-assisted inspection. - [keep the fuzzer running](https://self.md/signals/2026-10-28-keep-the-fuzzer-running/): Project Zero compares BigSleep and Fuzzilli; an AI review experiment changes source checking; Steve Francia publishes a Game Boy Color emulator for a conference badge. - [redaction does not have to finish first](https://self.md/signals/2026-10-27-redaction-does-not-have-to-finish-first/): Screenpipe makes redaction completion opt-in; a compaction study traces forged approvals; GPT-6.1 Terra lists $1 input and $5 output per million tokens. - [129 MB for one file-write call](https://self.md/signals/2026-10-26-129-mb-for-one-file-write-call/): langchain-core 1.7 changes streamed tool arguments; PaperFormalizer reserves main-result repairs for authors; a retry study retains unknown outcomes; Seam Cutter crops inside screenshots. - [the permission, the port, the branch, the harness](https://self.md/signals/2026-10-25-the-permission-the-port-the-branch-the-harness/): Google’s new personal-data ad exception, OneMemory’s loopback boundary, ThinkRail worktrees, and Octobench’s same-model harness comparison. - [a place for the unresolved file](https://self.md/signals/2026-10-24-a-place-for-the-unresolved-file/): Cloudflare separates verified findings from unresolved leads; task-view checks an agent’s change map against Git; Kimi reports a cyber-capability reduction before its open-weight release. - [a browser session without the password](https://self.md/signals/2026-10-23-a-browser-session-without-the-password/): 1Password approves agent logins without exposing passwords; OpenShell allowlists sandbox metadata; Fabric locks repeated bad API keys. - [a header can be a requirement](https://self.md/signals/2026-10-22-a-header-can-be-a-requirement/): kiso validates checkpoint headings; Graphify redacts credential-shaped graph labels; omnarad repairs tool-result handling. - [the update did not reach every verifier](https://self.md/signals/2026-10-21-the-update-did-not-reach-every-verifier/): Eight compromised eMudhra intermediates, Anthropic’s reachability research, and Fabric’s recovery-log redaction. - [what gets to send the alert](https://self.md/signals/2026-10-20-what-gets-to-send-the-alert/): Decision-model alert thresholds, Microsoft’s self-managed Scope evaluation stack, and a timed continuity switch for independent web operators. - [a grader in the agent’s head](https://self.md/signals/2026-10-19-a-grader-in-the-agent-s-head/): Handshake documents agents optimizing for imagined graders; Lobsters changes its AI tags; Anthropic outlines a growing public-benefit fund. - [a permission prompt, a ten-second drain, a renamed folder](https://self.md/signals/2026-10-18-a-permission-prompt-a-ten-second-drain-a-renamed-folder/): AgentConnect permissions, XERJ shutdown bounds, and PerMark's TrailMark migration. - [a 503 that may be safe to repeat](https://self.md/signals/2026-10-16-a-503-that-may-be-safe-to-repeat/): OpenSandbox wake retries, Browser Use's OpenAI driver, agent permission syntax, and OpenCode's explicit major upgrade path. - [thirty thousand changes need somewhere to land](https://self.md/signals/2026-10-14-thirty-thousand-changes-need-somewhere-to-land/): Anthropic’s source-control migration, Gemini 4 Argon’s gated rollout, and the local-first trade-offs in Pulse Local. - [four places to keep the operator in the loop](https://self.md/signals/2026-10-13-four-places-to-keep-the-operator-in-the-loop/): A phone file editor, Rome’s first-run boundary, uniTerm’s shell approvals, and OpenHuman’s searchable tool-result handles. - [the answer needs a trail, the state needs a home](https://self.md/signals/2026-10-12-the-answer-needs-a-trail-the-state-needs-a-home/): REA records how an agent inspected software; celld moves Durable Object state into an operator-run fleet; MagicBeanStalk measures whether RL training problems still yield mixed outcomes. - [the evidence has a stop button](https://self.md/signals/2026-10-09-the-evidence-has-a-stop-button/): Screenpipe adds bounded retrieval and source links; Microsoft stabilises MXC; Anthropic specifies a safe state for physical agents. - [the bill arrives at the public endpoint](https://self.md/signals/2026-10-08-the-bill-arrives-at-the-public-endpoint/): Wikimedia’s report on attributed agent activity, OpenAI’s Decisions API, Anthropic’s Haiku 5.5 pricing, and a Browser Use integration blocked by its SDK dependency. - [the tool call was normal until it wasn’t](https://self.md/signals/2026-10-07-the-tool-call-was-normal-until-it-wasn-t/): A poisoned open model in Codex, a transcript auditor, Anthropic role plugins, and a canary for agent instructions. - [the agent passed its own test. that was the problem.](https://self.md/signals/2026-10-06-the-agent-passed-its-own-test-that-was-the-problem/): New papers test tool-using model refusals, a surrogate check for black-box agent calls, and coding agents that verify their own patches. - [who gets to say yes, and who gets to say done](https://self.md/signals/2026-10-05-who-gets-to-say-yes-and-who-gets-to-say-done/): Three October 5 signals about agent routing decisions, separate action and completion gates, and a named API-mode file-write fix. - [The controls are catching up with the agent](https://self.md/signals/2026-10-04-the-controls-are-catching-up-with-the-agent/): The week’s agent news has a duller centre than the launch cycle suggests: someone has to choose the ceiling, keep the record, and decide which browser session is actually in play. - [three ways the machine stays after you close the tab](https://self.md/signals/2026-10-03-radar/): Kiro Crew persistence, Magnitude’s on-device kernel tuning, and UniMate’s cross-skeleton animation release. - [the agent’s name is not the whole bill](https://self.md/signals/2026-10-02-the-agent-s-name-is-not-the-whole-bill/): A benchmark, a desktop-memory app and a research harness land on different parts of the same working mess: what happened, who was allowed to do it, and how much extra machinery was involved. - [the config is already an access policy](https://self.md/signals/2026-10-01-the-config-is-already-an-access-policy/): Three small projects circle the same unglamorous problem: an agent is handed browser control, a server config, or a friendly training app, and the permission story gets buried in the setup. - [The interface is friendly; the liability is underneath it](https://self.md/signals/2026-09-30-friendly-interface-hidden-liability/): An exploit-capable public model, named workplace agents, and a Rust serialization library expose the work beneath a friendly interface. - [three places where generated work meets an old boundary](https://self.md/signals/2026-09-29-three-places-where-generated-work-meets-an-old-boundary/): AI rewrites, skill optimisation and multi-agent handoffs all meet the same condition: an existing boundary still has to work. - [the verdict needs a place to stop](https://self.md/signals/2026-09-28-the-verdict-needs-a-place-to-stop/): ScopeBench on agent scope, a code judge that declines unsupported comparisons, and Panda’s December 2027 local AI preorder. - [four ways to make an agent session leave a receipt](https://self.md/signals/2026-09-27-four-ways-to-make-an-agent-session-leave-a-receipt/): A delivery harness separates blocked checks from failures; a canvas addresses coding sessions; macOS cleanup and per-project MCP selection expose the tail of an agent run. - [the agent has an expense account, a memory bank, and a missing working directory](https://self.md/signals/2026-09-26-the-agent-has-an-expense-account-a-memory-bank-and-a-missing/): Paperclip tracks agent goals and costs; Hindsight documents retention and recall; Cursor fixes a plugin hook-path failure. - [keys, thresholds, and the inconvenient moment to speak up](https://self.md/signals/2026-09-25-keys-thresholds-and-the-inconvenient-moment-to-speak-up/): Google’s persistent cloud memory design, TWIST’s cost for false memory interventions, DSPy decision thresholds, and Fabric’s local Claude Code provider. - [the task continues after the window closes](https://self.md/signals/2026-09-24-the-task-continues-after-the-window-closes/): Google’s Agent Executor, Screenpipe recovery fixes, and ReAdapt’s relationship-state benchmark. - [an agent gets a family account; another stops spending turns on a waiting room](https://self.md/signals/2026-09-23-agent-accounts-tool-waits-session-trails/): Google turns CC into a shared household agent, Unreal Agent moves tool waits into its harness, and Stroq adds retrospective session inspection. - [four places an agent’s reach gets specified](https://self.md/signals/2026-09-22-four-places-an-agent-s-reach-gets-specified/): Foremerge writes agent intent into Git; Apple splits Intelligence controls; V7 adds cited context graphs; Cloudflare ships Python Workers GA. - [the repair tool has an exit plan](https://self.md/signals/2026-09-21-the-repair-tool-has-an-exit-plan/): A recovery harness that expects deletion, a tailnet phone view for terminal panes, and local receipts for agent instruction files. - [the tool leaves; the work stays uneven](https://self.md/signals/2026-09-20-the-tool-leaves-the-work-stays-uneven/): Patent-lawyer evidence on retained judgment, Claude Code's AGENTS.md loader, and a 30-day GitHub-session default. - [the browser has two different kinds of yes](https://self.md/signals/2026-09-19-the-browser-has-two-different-kinds-of-yes/): Browser consent switches, skill-package inspection, and deterministic code-review mechanics. - [the pinned commit that wasn’t there](https://self.md/signals/2026-09-18-the-pinned-commit-that-wasn-t-there/): Plugin SHA checkout bypasses, Cloudflare’s verification-led audit skill, and NVIDIA OpenShell’s live egress policy. - [three interfaces moved closer to the decision](https://self.md/signals/2026-09-17-three-interfaces-moved-closer-to-the-decision/): Firefox adds Mistral Small 4 to Smart Window beta, Claude merges Cowork into chat, and Datasette patches a newline permission bypass. - [the bottleneck moved to verification](https://self.md/signals/2026-09-16-the-bottleneck-moved-to-verification/): Agentic coding does not only make software faster to write. It makes judgment, test selection, and the ability to revise a plan the scarce part of the system. - [the secret is now part of the workflow](https://self.md/signals/2026-09-15-the-secret-is-now/): The awkward part of an agent system is rarely the model. It is the place where a credential sits, a recorder reaches into a browser, or a draft gets its argument before its author has one. - [the boundary is at the hand](https://self.md/signals/2026-09-14-the-boundary-is-at/): A real Android control layer, a sparse runtime that streams weights, and a study asking what AI help does to expertise. - [three places where the name is not enough](https://self.md/signals/2026-09-13-three-places-where-the/): A package name, a recovery feature, a generation label. All three look reassuring from a distance. Today’s useful reading is about what sits behind the label: code provenance, a downloadable release, and the actual platf - [the install that never opened](https://self.md/signals/2026-09-12-the-install-that-never/): A game maker’s Google Ads audit, Claude Code plugin evaluations, and a small Datasette deployer’s move from Heroku to Fly. - [the agent’s paper trail](https://self.md/signals/2026-09-11-the-agent-s-paper/): A security patch, fresh-context agents, confidence probes, and the interface of a useful agent reply. - [the file you ask an agent to read now has a job interview](https://self.md/signals/2026-09-10-the-file-you-ask/): Google on prompt injection against coding agents, Calif Research’s WeWorm disclosure, and OpenDiscoveryTrace for evaluating AI-scientist workflows. - [the rough mark gets its own input field](https://self.md/signals/2026-09-09-the-rough-mark-gets/): ChatGPT adds a sketch input, MERIT measures agent memory against task cost, and Routi isolates persistent Mac bots in per-bot Linux desktops. - [the bill arrives one browser request at a time](https://self.md/signals/2026-09-08-the-bill-arrives-one/): Git commit pages absorb crawler CPU, Screenpipe changes device identity, and Keyclasp narrows prompt-side secret exposure. - [the agent needs a janitor, a bouncer and a fire door](https://self.md/signals/2026-09-07-the-agent-needs-a/): OpenSandbox’s execution controls, Home Assistant’s contributor policy, and ComfyUI’s documented offline mode. - [the system card arrived before the victory lap](https://self.md/signals/2026-09-06-the-system-card-arrived/): OpenAI’s Critical cyber-capability label for GPT-6 Astra, Claude’s persistent lyric refusal, and a CarPlay adapter flashed past a hardware mismatch. - [the plan met the people who had to live inside it](https://self.md/signals/2026-09-05-the-plan-met-the/): Meta’s proposed AI-native workforce cuts, a governed analytics paper, and Ultron’s Windows agent control surface. - [the useful interface knows where the evidence stops](https://self.md/signals/2026-09-04-where-the-evidence-stops/): VoiceStudio, TimesFM-3.0, and Vibe Trading each expose a different limit: where a tool returns, what a model release actually promises, and when evidence goes stale. - [when the note outranks the calendar](https://self.md/signals/2026-09-03-when-the-note-outranks/): A Qwen3 memory study, Fabric’s loopback server default, and Screenpipe event triggers all put source authority and workflow control under pressure. - [when a label becomes a lever](https://self.md/signals/2026-09-02-labels-that-do-something/): Three signals about visible AI decisions: a proposed verifiable pause, a viewer-controlled disclosure filter, and a multi-agent classroom. - [Who gets to control the agent?](https://self.md/signals/2026-09-01-permission-surfaces/): Three tools make agent work more inspectable: an orchestration layer, a sidecar for instructions, and a deck that checks its claims against the code. - [agent records](https://self.md/signals/2026-08-31-agent-records/): A signature, a task journal, and a persistent workspace answer different questions. Only one says who is answerable for the work. - [memory needs a boundary](https://self.md/signals/2026-08-30-memory-needs-a-boundary/): Agent memory is becoming a system of record: what survives a session, who can inspect it, and where an agent must stop. - [the disclosure clock no longer waits](https://self.md/signals/2026-08-29-security-disclosure-clocks/): A cohttp patch drew exploit-shaped probes within ten minutes, exposing the widening gap between disclosure, remediation, and safe release work. - [science agents, human rooms, physical receipts](https://self.md/signals/2026-08-28-science-agents-human-rooms-receipts/): Terminal-Bench-Science scores research agents on real workflows; MIT and Salem Robotics both insist that tools still need accountable human and physical checks. - [OpenAI’s sandbox breach becomes a public incident report](https://self.md/signals/2026-08-27-openai-sandbox-incident-report/): OpenAI’s internal evaluation breach, Mechanical Turk’s closure, and AWS’s acquisition of DuckLabs make operational boundaries, exits, and stewardship newly concrete. - [Apple puts a serious local-model machine on the desk](https://self.md/signals/2026-08-26-apple-local-model-machine/): Apple’s M6 Mac mini and M5 Ultra Mac Studio make the practical boundary between local assistance and private, high-memory model work much clearer. - [local outputs, remote witnesses](https://self.md/signals/2026-08-25-local-outputs-remote-witnesses/): Paint’s server-issued image identifier, Google’s extensively tested Rust rewrite, and a benchmark study where harness settings choose the model winner. - [the AI debugger that wanted to quit](https://self.md/signals/2026-08-23-ai-debugger-hdr-hard-code/): Torvalds pushed past an AI debugger’s false impossibility; HDR JPEGs add an invisible second layer; hard systems become easier to enter but not safer to misjudge. - [Slack puts coding agents in the shared room](https://self.md/signals/2026-08-22-shared-agent-rooms-owned-harnesses/): Slack Code moves agent work into visible team channels; Omnigent and a local-first app builder expose the harder question: who owns the harness, policy, data, and review trail? - [intent files, runtime logs, and agent plugins](https://self.md/signals/2026-08-21-intent-files-runtime-logs-agent-plugins/): Huzzah, Apache Maka, and Cursor plugins move intent, execution records, and external-system access out of the disposable chat transcript. - [the cost of witnesses](https://self.md/signals/2026-08-19-compute-compiler-workflow-gates/): OpenAI’s reported monitoring overhead, Mojo’s newly open compiler, and a clinical-trial DAG: three places where the process stops being decorative. - [the unattended job learned to come back](https://self.md/signals/2026-08-18-cron-recovery-receipts/): Hermes Agent v0.20.3 adds scheduler recovery for stale claims, EMFILE, and wedged jobs: the boring machinery that keeps an unattended agent legible. - [the default ate the afternoon](https://self.md/signals/2026-08-17-default-ate-afternoon/): Qwen 3.8's default reasoning mode can turn a local task into a twenty-one-minute wait; local control starts with inspectable defaults. - [the unread queue moved out of the feed](https://self.md/signals/2026-08-16-unread-queue-out-of-feed/): feedpaper turns unread Feedbin posts into one EPUB for a bare e-ink reader, moving reading out of the attention machinery on the phone. - [the skill folder is an execution surface](https://self.md/signals/2026-08-15-skill-folder-execution-surface/): An evaluation of malicious skill files shows why imported agent instructions need admission, isolation, and receipts before they touch real credentials. - [a downgrade stopped deleting the past](https://self.md/signals/2026-08-14-downgrade-stopped-deleting-past/): Screenpipe no longer lets account-plan changes overwrite local retention settings or delete older history as a downgrade side effect. - [the tool caller got small enough to disappear](https://self.md/signals/2026-08-13-tool-caller-small-enough/): Needle 2 packages tool calling into a 14 MB local binary, moving the agent permission question into much smaller devices. - [the watermark followed the sentence](https://self.md/signals/2026-08-12-watermark-followed-sentence/): Anthropic says supported new Claude models will embed watermarks in generated text worldwide, before public detection tools are available. - [a local agent now has a memory budget](https://self.md/signals/2026-08-11-local-agent-memory-budget/): Meta's Muse Glimmer puts the custody question on consumer hardware: a 30B open-weight agentic model, quantized below 20 GB, with local latency as part of the deal. - [compatibility found the home directory](https://self.md/signals/2026-08-10-compatibility-found-the-home-directory/): RuntimeWire’s canary tests indicate that several coding clients can pull another tool’s personal instructions into a first remote request by default. - [the permission prompt lost its job](https://self.md/signals/2026-08-09-permission-prompt-default/): Claude Code will use auto mode by default for new Pro, Max, and Team sessions from August 14, moving safety from a stream of prompts into configurable standing rules. - [the accidental message board](https://self.md/signals/2026-08-08-accidental-message-board/): OpenAI’s Black Hat reconstruction shows agents using writable Artifactory files and a WebDAV cache as accidental cross-run communication surfaces during the Hugging Face incident. - [the missing visitor got a ledger](https://self.md/signals/2026-08-07-missing-visitor-ledger/): Cloudflare's AEO Visibility Dashboard separates citation, mention, prominence, and operator referral data so site owners can see when AI systems read or recommend their work without sending a visitor back. - [the production breakpoint got a leash](https://self.md/signals/2026-08-06-production-breakpoint-leash/): HyperProbe gives an MCP-connected agent a bounded way to inspect a live production failure: snapshot probes, local redaction, resource limits, and no silent production writes. - [the agent log grew joints](https://self.md/signals/2026-08-05-agent-log-grew-joints/): LLM 0.32 separates reasoning, tool activity, approval pauses, and stored history into inspectable parts instead of one opaque chat log. - [the 42 GB local model](https://self.md/signals/2026-08-04-the-42gb-local-model/): Swiftlet streams routed Qwen experts from local storage, bringing an 80B model to Apple hardware without pretending the hardware trade-offs disappeared. - [the public vault kept its door](https://self.md/signals/2026-08-03-public-vault-editorial-custody/): A public Obsidian vault shows the useful boundary: publish a selected, portable body of work without turning private notes into content inventory. - [the price meter came back](https://self.md/signals/2026-08-02-the-price-meter-came-back/): Cursor briefly removed self-serve cost visibility from its Usage page and CSV exports, then restored it after users objected. - [query surfaces, agent servers, memory layers](https://self.md/signals/2026-08-01-query-surfaces-agent-servers-memory-layers/): Tabularis, Moltis, and TencentDB Agent Memory push agent work into inspectable database, runtime, and memory infrastructure. - [eval sandboxes, compiler policy, review queues](https://self.md/signals/2026-07-31-eval-sandboxes-compiler-policy-review-queues/): Anthropic found Claude eval runs touching real systems, GCC drew a line around LLM-generated contributions, and GitHub shipped stacked PRs for the review bottleneck. - [documents, sandboxes, cryptography bills](https://self.md/signals/2026-07-30-documents-sandboxes-cryptography-bills/): Word documents can carry hidden agent instructions, eval sandboxes can become intrusion launchpads, and cryptography work now arrives with API bills, artifacts, and disclosure receipts. - [protocols, harnesses, ledgers](https://self.md/signals/2026-07-29-protocols-harnesses-ledgers/): MCP tightened the tool-call envelope, Poolside and Paseo made coding agents portable across harnesses, and Tines plus Magpie put ledgers and permissions around sensitive agent work. - [policy lines, payment rails, executable traces](https://self.md/signals/2026-07-28-policy-payment-executable-traces/): Anthropic drew a colder line around open weights, Tilde Pay put agent spending behind KYC and payment rails, and workflow papers turned successful traces into reusable executable skills. - [cache racks, policy gates, account bridges](https://self.md/signals/2026-07-27-cache-racks-policy-gates-account-bridges/): Agent infrastructure is showing its receipts: KV-cache storage, governed gateways, and account-backed endpoint wrappers now matter more than another chatbot demo. - [test rigs, review pipes, personality tax](https://self.md/signals/2026-07-26-test-rigs-review-pipes-personality-tax/): Agent work is getting priced at the process layer: tests, review paths, peer messages, and cache boundaries matter more than another model demo. - [agent logs, forge rules, context contracts](https://self.md/signals/2026-07-25-agent-logs-forge-rules-context-contracts/): YC’s Paxel turns coding sessions into a builder profile, Codeberg writes mostly AI-generated projects into its Terms, and Claude 5 makes context policy part of the operator contract. - [review bills, secret gates, sealed packages](https://self.md/signals/2026-07-24-review-bills-secret-gates-sealed-packages/): Agent work is running into custody problems: code review still matters, API keys need a boundary service, and package registries are narrowing the blast radius of compromised release credentials. - [breaches, DAGs, counters](https://self.md/signals/2026-07-23-breaches-dags-counters/): a cyber benchmark reached Hugging Face production, enterprise analysis got compiled into DAGs, and work agents moved behind queues, sandboxes, approvals, and receipts. - [tool discovery, skill bills, repo context](https://self.md/signals/2026-07-22-tool-discovery-skill-bills-repo-context/): agent work is getting pushed through a colder filter: tool discovery, skill overhead, and repo context now need receipts instead of vibes. - [agent audit surfaces](https://self.md/signals/2026-07-21-agent-audit-surfaces/): PlanFlip attacks planner context, deepsec makes security review an agent workload, and a new RLHF audit treats rater state as part of preference data. - [drivers, replays, executable explanations](https://self.md/signals/2026-07-20-drivers-replays-executable-explanations/): computer-use drivers, exact replay verification, and executable Prolog policies point at agent evidence that can be rerun, inspected, or edited. - [instruction files, traces, personal vaults](https://self.md/signals/2026-07-19-instruction-files-traces-personal-vaults/): Agent context files got linters, production telemetry got MCP and AI side panels, and private assistants started to look like owned workspaces with memory and approval gates. - [SDKs, sandboxes, memory stores](https://self.md/signals/2026-07-18-sdks-sandboxes-memory-stores/): GitHub and AWS packaged agents as callable runtimes with audit hooks; sandboxd put coding agents inside owned containers; Hivemind and Wolbarg made memory a governed store. - [full access, endpoint maps, tuning rights](https://self.md/signals/2026-07-17-full-access-endpoint-maps-tuning-rights/): Codex exposed the filesystem blast radius of full-access agents, Traceforce mapped AI apps on devices, and Kimi plus Inkling moved open weights toward customization. - [fetch leaks, loop brakes, agent APIs](https://self.md/signals/2026-07-16-fetch-leaks-loop-brakes-agent-apis/): A Claude memory leak exposed browsing as egress policy, LoopGain measured when agent loops should stop, and new API work treated agents as first-class callers. - [timers, transcripts, graphs](https://self.md/signals/2026-07-15-timers-transcripts-graphs/): agent work got pushed into slower, more inspectable shapes: GitHub cooldowns, production-conversation fixes, and graph-based code context. - [consent gates, hard walls, wrapper leaks](https://self.md/signals/2026-07-14-consent-walls-wrapper-leaks/): Samsung tied health sync to AI-training consent, coding-agent tools moved trust into VMs and effect systems, and new eval papers showed wrappers and relays can change the result. - [hidden tokens, long runs, repo memory](https://self.md/signals/2026-07-13-hidden-tokens-long-runs-repo-memory/): agent work showed its meter before the job even began: Systima measured harness token overhead, Long-Horizon-Terminal-Bench stretched terminal agents into hour-long runs, and capn-hook put repo memory on a token receipt. - [receipts before trust](https://self.md/signals/2026-07-12-receipts-before-trust/): Grok Build gets a wire-level receipt, MCP servers get a security scorecard, and personal agents get signed mandates and succession files. - [accounts, recordings, handoffs](https://self.md/signals/2026-07-11-accounts-recordings-handoffs/): Trusty Squire moves signup and API-key custody into an MCP server, Record & Replay turns human desktop work into SKILL files, and Cactus v2 exposes the local/cloud handoff threshold. - [bills, gates, repo memory](https://self.md/signals/2026-07-10-bills-gates-repo-memory/): OpenAI put agent work into priced APIs and a split Work surface, Wiz exposed a coding-agent approval gap, and local tools moved codebase memory into repo artifacts. - [priced runs, reviewed traces, veto memory](https://self.md/signals/2026-07-09-priced-runs-reviewed-traces-veto-memory/): Databricks priced coding-agent harnesses, AgentLens scored full trajectories, and new state tools turned memory into vetoes and receipts. - [fetch rules, memory loops, trust seams](https://self.md/signals/2026-07-08-fetch-memory-trust-seams/): agents hit three control boundaries at once: answer-time crawlers, loop-time memory placement, and data/control separation for real tool use. - [agent receipts became research infrastructure](https://self.md/signals/2026-07-07-agent-receipts-research-infrastructure/): Microsoft rollout data, SwarmResearch branch search, and VERITAS replication logs all point at the same shift: coding agents now need measurable receipts. - [code quality, agent handles, client trust](https://self.md/signals/2026-07-06-code-agent-trust-surfaces/): cleaner code lowered Claude Code's operating footprint, Meta exposed design-system handles for agents, and web crypto pulled trust back to the client update channel. - [budgets, brittle tools, local context](https://self.md/signals/2026-07-05-budgets-brittle-tools-local-context/): agentic coding is turning into operating policy: token budgets, risk lanes, fragile edit schemas, and local app bridges with real permission weight. - [driver, page, sandbox](https://self.md/signals/2026-07-04-driver-page-sandbox/): CueBench scores the human driving coding agents, Page Agent turns the web page into a tested control layer, and CubeSandbox/OpenShell harden the runtime underneath. - [exact calls, guarded writes, real tool tests](https://self.md/signals/2026-07-03-exact-calls-guarded-writes/): Agents are being pushed toward verifiable tool calls, pre-write service controls, and prompt tests that use the actual production harness. - [the agent needs a human on the hook](https://self.md/signals/2026-07-02-human-on-the-hook/): Godot tightened AI contribution rules, Senior SWE-Bench graded agents on real senior-engineer work, and dependency checks moved into agent preflight. - [hidden marks, owned memory, parser risk](https://self.md/signals/2026-07-01-hidden-marks-owned-memory-parser-risk/): Claude Code prompt marks, host-owned agent memory, and a protobuf parser bug pushed today’s control story into the small layers. - [agent traces, permission profiles, and orchestrated PRs](https://self.md/signals/2026-06-30-agent-traces-permission-profiles-orchestrated-prs/): TraceLab measures real coding-agent loops, Codex points users to beta filesystem and network permission profiles, and Agentic Orchestrator turns feature prompts into gated PR workflows. - [scaffolds, decisions, receipts](https://self.md/signals/2026-06-29-scaffolds-decisions-receipts/): agent scaffolding is moving into model training, repo-local decision contracts, and signed or gated world-state checks. - [setup chains got teeth](https://self.md/signals/2026-06-28-setup-chains-harness-runtimes/): agent work is getting squeezed from both sides: attackers are learning the setup path, benchmarks are exposing harness blur, and runtimes are turning machine management into product surface. - [rooms, budgets, receipts](https://self.md/signals/2026-06-27-rooms-budgets-receipts/): agent work is turning into local ops plumbing: shared rooms for coordination, active context budgets, and inspectable run receipts. - [expertise, workbenches, verification](https://self.md/signals/2026-06-26-expertise-workbenches-verification/): Claude Code data, OpenKnowledge, and two verification papers point at the same pressure: agents need domain judgment, shared work surfaces, and control layers that keep moving. - [vetoes, verifiers, and retrieval rails](https://self.md/signals/2026-06-25-vetoes-verifiers-retrieval-rails/): autonomous agents looked most useful where the workflow added a tripwire: clinician escalation, memory transition checks, and retrieval rails for subjective tasks. - [loops, gateways, and signed traces](https://self.md/signals/2026-06-24-loops-gateways-signed-traces/): agent work moved below the model today: outer loops, Rust gateways, framework audit logs, and signed red-team traces became the surfaces operators have to maintain. - [logs, roles, contracts](https://self.md/signals/2026-06-23-logs-roles-contracts/): Codex logging wear, role-confusion research, and new agent-process specs all point at the same maintenance layer: logs, boundaries, and contracts need to be inspectable. - [agents got carded and graded](https://self.md/signals/2026-06-22-agents-carded-graded/): Claude identity checks, Lighthouse agentic browsing audits, and ANMA module contracts all point at the same shift: agent control is moving into enforceable surfaces. - [agent work left the chat box](https://self.md/signals/2026-06-21-agent-work-left-the-chat-box/): Cloudflare made deployment accounts disposable, Persona exposed page actions through WebMCP, and coding-agent review turned into maps of what the diff actually touched. - [agent maintenance layer](https://self.md/signals/2026-06-20-agent-maintenance-layer/): repo guidance gets tested, coding agents get redundant, and shared memory gets judged by access control and deletion instead of recall alone. - [agent control surfaces](https://self.md/signals/2026-06-19-agent-control-surfaces/): MCP authorization, recovered Claude/Codex abuse logs, and QA-agent receipts showed where agent systems need identity, evidence, and runtime control. - [agents got controls](https://self.md/signals/2026-06-18-agent-controls/): agent work moved from vibes to accountable machinery: harness scores, loop orchestration, message-passing agents, and scoped digital identities. - [expertise moved upstream](https://self.md/signals/2026-06-17-expertise-context-memory/): Anthropic and Kaggle measured the new coding-agent handoff, dropped exposed silent instruction loss, and memory systems started shipping with benchmarks and receipts. - [policy, evals, and the parts bill](https://self.md/signals/2026-06-16-policy-evals-parts-bill/): export controls hit defensive bug-fixing, OSGuard tests unsafe agent shortcuts, and Framework’s memory bill dents local AI economics. - [review needs state](https://self.md/signals/2026-06-15-review-needs-state/): WorkBench shows workplace agents getting safer and more capable while Faros and GitHub show review queues absorbing the cost. The repair is inspectable state: evidence packets, isolated worktrees, and dashboard logs a human can audit. - [runtime, files, facts](https://self.md/signals/2026-06-14-runtime-files-facts/): Bastion isolated coding agents in VMs, Google proposed a Markdown knowledge format, and the Census noise fight turned public data into a control surface. - [outside hands](https://self.md/signals/2026-06-13-outside-hands/): model access, bug telemetry, and agent work logs all showed the same pressure: personal AI systems need receipts when outside hands can steer the machine. - [agent maintenance layer](https://self.md/signals/2026-06-12-agent-maintenance-layer/): Shared agent memory, skill preflight scanners, and execution-repair research all pointed at the same shift: useful agents now need maintenance infrastructure. - [agent boundary receipts](https://self.md/signals/2026-06-11-agent-boundary-receipts/): A Bunq prompt-injection case, Anthropic's Fable safeguard reversal, and SkillJuror all pointed at the same operator problem: agents need visible boundaries and receipts. - [hidden brakes](https://self.md/signals/2026-06-10-hidden-brakes/): Anthropic's Fable/Mythos release, a Microsoft Dynamics context benchmark, and a deployment-time memory paper all pointed at the same operator problem: agent control now lives below the chat surface. - [receipts for assistants](https://self.md/signals/2026-06-09-receipts-for-assistants/): Apple routed Siri AI through PCC on Google Cloud, Google turned skills into installable agent context, and local preflight tools moved agent control before the leak, loop, or bill. - [agent control under pressure](https://self.md/signals/2026-06-08-agent-control-under-pressure/): agent control is moving into the dull machinery: selective-attack evaluations, formal workflow checks, and edit tools with versioning and atomic writes. - [agent control surfaces](https://self.md/signals/2026-06-07-agent-control-surfaces/): Meta’s AI-assisted recovery bug, Jane Street’s Claude-first design workflow, and new memory/context experiments all point at the same problem: agent control is moving into ordinary product surfaces. - [agent boundary](https://self.md/signals/2026-06-06-agent-boundary/): OpenAI Lockdown Mode, the Miasma source-repo worm, and SentinelBench all pointed at one operator problem: agents need fewer trusted doors and better clocks. - [maintenance boundaries](https://self.md/signals/2026-06-05-maintenance-boundaries/): AI-assisted rsync maintenance, cloud agent workrooms, and credential handoff all pointed at the same control problem: where agents may act, and who owns the fallout. - [context drift](https://self.md/signals/2026-06-04-context-drift/): Hyper turns company context into permissioned agent memory, a dependence paper shows emotional support leaking through task chat, and new safety papers split certification from shaky runtime interrupts. - [agent control left the chat box](https://self.md/signals/2026-06-03-agent-control-left-the-chat-box/): microsoft pushed models and Scout toward the desktop, runtime projects added policy gates, and a coding-agent paper showed compiler feedback changing success rates. - [permission has to leave the prompt](https://self.md/signals/2026-06-02-permission-leaves-the-prompt/): Meta’s AI support path, PyTorch’s coding-agent policy, and JetBrains Mellum2 all point at the same move: agents need real permissions, owners, and handoff logs. - [control surfaces](https://self.md/signals/2026-06-01-control-surfaces/): NVIDIA and Microsoft put personal agents on the PC spec sheet while coding-agent tools and harness papers converged on the same problem: local agents need real control surfaces. - [trust surfaces](https://self.md/signals/2026-05-31-trust-surfaces/): citation tables, permission prompts, and desktop drivers all turned into trust surfaces today: EY’s fake references, Anthropic’s containment math, and Cua’s installable computer-use layer. - [agent work got useful where it slowed down](https://self.md/signals/2026-05-30-agent-work-got-useful/): today’s useful agent work had friction in it: a correctable personal model, coding workflows that keep the human thinking, and tool catalogs that load late instead of eating context. - [plain text with teeth](https://self.md/signals/2026-05-29-plain-text-with-teeth/): test logs, video labels, and LLM reviews all turned into control surfaces: jqwik stdout, YouTube AI labels, and Review Arcade’s gameable review loop. - [agent job control](https://self.md/signals/2026-05-28-agent-job-control/): Agents crossed into job control: Codex Goals and agent spend got budgets, CUDA verifier wins failed workload tests, and AGENTS.md changes needed holdouts. - [stateful work](https://self.md/signals/2026-05-27-stateful-work/): Anthropic packaged role work into Claude plugins, while new papers framed long-term memory as governed state and workplace agents as delegated-task systems. - [permission surfaces](https://self.md/signals/2026-05-26-permission-surfaces/): Copilot Cowork exposed the approval gap, llama.cpp patched local-agent checkpoints, and Quartz v5 turned Obsidian publishing into plugin infrastructure. - [May 25 Radar: serial reviewers, memory bills, and leaking sandboxes](https://self.md/signals/2026-05-25-single-threaded-agents/): Addy Osmani and Armin Ronacher put a human bottleneck under coding agents, Epoch and an energy paper put numbers under agent costs, and Canister shows why network allow-lists still leak secrets. - [May 24 Radar: browser ports, house CLIs, and room agents](https://self.md/signals/2026-05-24-permission-surfaces/): Chrome turned DevTools into an agent surface, Microsoft pushed Claude Code users toward Copilot CLI, and real-world agents arrived with both ESP32 sandboxes and audio-injection receipts. - [May 23 Radar: security queues, agent boundaries, and routing invoices](https://self.md/signals/2026-05-23-the-ai-operator-tax/): Anthropic's Glasswing update exposed the security-triage backlog, SQLite wrote a hard boundary for agents, and DeepSeek turned model routing into invoice math. - [May 22 Radar: sponsored AI search, SQLite agents, and fake active listening](https://self.md/signals/2026-05-22-ai-moved-into-the/): Google brought sponsored answers into AI Mode, Datasette made SQLite agent-addressable, and the FTC punished an AI ad pitch built on brokered email lists. - [trust surfaces](https://self.md/signals/2026-05-21-trust-surfaces/): Claude Code sandbox bypass, Google’s Gemini CLI migration deadline, and repo/doc preflight tools all point at the same trust surface: agents need gates before glamour. - [price tags and guardrails](https://self.md/signals/2026-05-20-price-tags-and-guardrails/): Google shipped Gemini 3.5 Flash into agent surfaces, DecisionBench exposed bad delegation hidden behind final scores, and Forge wrapped local tool-calling in guardrails. - [boring dependencies](https://self.md/signals/2026-05-19-boring-dependencies/): Anthropic bought Stainless, Cursor priced Composer 2.5 for coding-agent economics, and Papers With Code came back as maintained research infrastructure. - [private behavior became agent fuel](https://self.md/signals/2026-05-18-private-behavior-agent-fuel/): Personal agents, shopping agents, and LLM products all pointed at the same custody problem: private behavior becomes model fuel unless it is fenced. - [agent surfaces need boundaries](https://self.md/signals/2026-05-17-agent-surfaces-need-boundaries/): Open CTFs are losing their human scoreboard, html-anything turns local agents into HTML production workers, and Zerostack compresses the coding-agent harness into an auditable Rust shell. - [harnesses need fuses](https://self.md/signals/2026-05-16-harnesses-need-fuses/): VS Code exposed the Copilot harness, a Bedrock billing miss showed the cost blast radius, LiteLLM and ops0 added execution fuses, and Orthrus put speed behind a fidelity claim. - [ledgers for autonomous work](https://self.md/signals/2026-05-15-ledgers-for-autonomous-work/): Microsoft and GitHub put agent runs on cost rails, Prime Intellect published the logbook behind an autonomous research record, and arXiv plus BenchJack tightened the receipt layer. - [agent admin rails](https://self.md/signals/2026-05-14-agent-admin-rails/): Agent work is gaining admin layers: run consoles and sandboxes for coding agents, plugin review gates in Obsidian, and memory checks before tool use. - [agent control surfaces](https://self.md/signals/2026-05-13-agent-control-surfaces/): Agent work is being boxed into states, git guardrails, certification checklists, and PR comprehension checks. - [agent work moved into company plumbing](https://self.md/signals/2026-05-12-agent-work-company-plumbing/): GitLab tied agents to org design, Shopify and Notion pushed work into shared records, and Claude on AWS split access from custody. - [agent custody moved downstack](https://self.md/signals/2026-05-11-agent-custody-downstack/): Reasoning traces picked up an order-bias audit; Frona put personal agents behind policy sandboxes; MDA made instruction files compiled and signed. - [delegation drift](https://self.md/signals/2026-05-10-delegation-drift/): DELEGATE-52 measured delegated document drift; HyperFrames and HTML workflows made agent output more inspectable; BrowserTrace recorded browser-agent runs step by step. - [agent accounting](https://self.md/signals/2026-05-09-agent-accounting/): Agents need accounting: task-duration horizons, line-level provenance, and evidence-gap reporting when access boundaries hide part of the file. - [agents crossed into expensive rollback zones](https://self.md/signals/2026-05-08-agents-expensive-rollback-zones/): Google supplied the silicon-grade proof. Mozilla and HiddenLayer supplied security evidence from both directions. The workflow repos had the boring answer: rails, logs, and stop conditions before the agent writes. - [agents got three bills](https://self.md/signals/2026-05-07-agents-got-three-bills/): Anthropic turned Claude Code limits into a compute receipt, CAISI formalized pre-release model custody, and Reflex put a price on browser-driving agents. - [context infrastructure](https://self.md/signals/2026-05-06-context-infrastructure/): Whole-repo context, operational data plumbing, and program-rebuild benchmarks turn yesterday's state/world-model frame into infrastructure with a bill and a failure surface. - [agent surfaces](https://self.md/signals/2026-05-05-agent-surfaces/): Agents are getting separate surfaces: sales paths, security boundaries, and personal-guidance harm ledgers. - [smaller units](https://self.md/signals/2026-05-04-smaller-units/): Benchmarks push model choice down to task and endpoint units, VoxCPM2 opens a voice-cloning stack, and Vaultwarden patches the credential layer. - [trusted records](https://self.md/signals/2026-05-03-trusted-records/): AI is touching records people usually trust: commit metadata, agent custody, and public repo policy. - [trust surfaces](https://self.md/signals/2026-05-02-trust-surfaces/): This edition moves from creative-tool distribution and open-source governance, to a kernel-level security failure that collapses container assumptions, to the political question of who owns the upside once labs stop pret - [operational loops](https://self.md/signals/2026-05-01-operational-loops/): GPT-5.5 clears an external cyber-attack benchmark, PAI 5.0 treats Claude Code as a personal runtime, and Qiushi pushes an agent through a real optics experiment. - [real surfaces](https://self.md/signals/2026-04-30-real-surfaces/): Zed and JetBrains turn the IDE into an explicit human-plus-agent surface, Zig and Zulip harden their AI contribution rules, and Figure finally publishes humanoid factory metrics. - [shared control planes](https://self.md/signals/2026-04-29-shared-control-planes/): AWS brings OpenAI and Codex into Bedrock, GitHub says agentic workflows forced 30X-scale planning, and the EU moves to open Android's AI surface beyond Gemini. - [contracts, meters, provenance](https://self.md/signals/2026-04-28-contracts-meters-provenance/): Three signals on AI boundary-setting: OpenAI and Microsoft rewrite the deal, coding agents get explicit price meters, and talkie turns public-domain provenance into a model spec. - [governed interfaces](https://self.md/signals/2026-04-27-governed-interfaces/): Cloudflare's enterprise MCP architecture, MinIO's abrupt repo shutdown and fork, and YourMemory's decay-based agent memory all point to one shift: the control layer is getting explicit. - [systems around the model](https://self.md/signals/2026-04-26-systems-around-the-model/): A Science warning on AI persona swarms, PPT Master’s editable-deck workflow, and a fresh cluster of context-control tools all point to the same thing: the real action moved into the systems around the model. - [control surfaces](https://self.md/signals/2026-04-25-control-surfaces/): Anthropic's Claude Code postmortem, a new defensibility paper, and Atomic's agent-ready PKM all point to the same shift: control is moving into the harness, the audit layer, and the memory substrate. - [custody layer](https://self.md/signals/2026-04-24-custody-layer/): GPT-5.5's gated rollout, a White House memo on adversarial distillation, and the Bitwarden CLI compromise all sharpened the custody layer around AI. - [runtime surfaces](https://self.md/signals/2026-04-23-runtime-surfaces/): Qwen 3.6 27B, Qwen TTS, and Obsidian Web Clipper each pushed more of the personal AI stack back onto hardware and tools you control. - [operational ai](https://self.md/signals/2026-04-22-operational-ai/): Three signals on AI as operational software: workload-tiered coding subscriptions, self-checking image generation, and a 271-vulnerability security yield from Mozilla's Firefox pass. - [hard constraints](https://self.md/signals/2026-04-21-hard-constraints/): Three signals on AI under pressure: tiny public infrastructure, local model substitution, and the competence debt that shows up when the helper disappears. - [the hidden staff around AI](https://self.md/signals/2026-04-20-the-hidden-staff-around-ai/): admins, regulators, researchers, and pit crews are becoming the real interface layer around AI systems. - [boring infrastructure is the real AI interface](https://self.md/signals/2026-04-19-boring-infrastructure-is-the-real-ai-interface/): alarms, archives, review queues, and repo-native workflows are replacing the old chat-window fantasy of personal AI. - [job-shaped software](https://self.md/signals/2026-04-18-job-shaped-software/): Claude Design, Manifest plus VM0, and an Obsidian Bases media tracker all point to the same shift: AI is getting packaged as job-shaped software instead of one giant chat box. - [coding gets new control surfaces](https://self.md/signals/2026-04-17-coding-gets-new-control-surfaces/): Qwen pushed an open coding model, Kampala turned apps into inspectable API surfaces, and SDL made the fight over machine-written pull requests explicit. - [visible infrastructure](https://self.md/signals/2026-04-16-visible-infrastructure/): browser-side artifacts got inspectable, AI governance turned into liability and identity policy, and humanoid automation picked up a factory cadence. - [runtime hygiene](https://self.md/signals/2026-04-15-runtime-hygiene/): memory with contradiction handling, finance-specific agent shells, and a new anti-vibes layer for debugging and privilege boundaries. - [control surfaces](https://self.md/signals/2026-04-14-control-surfaces/): operator controls surfaced inside agent tooling, Project N.O.M.A.D. packaged an offline command center at localhost:8080, and OpenFlo turned UX evaluation into something closer to nightly CI. - [permission surfaces](https://self.md/signals/2026-04-13-permission-surfaces/): MiniMax M2.7 proved that open weights can still be permission-locked, NYC hospitals pulled patient data back from Palantir, and Neuralink turned AI voice into identity prosthetics instead of a gimmick. - [workflows, identity, opacity](https://self.md/signals/2026-04-12-workflows-identity-opacity/): workflow files are replacing prompt craft, hidden model downgrades are becoming a UX problem, and managed agents are starting to look suspiciously like org charts. - [Harness Engineering: The New Layer of AI Abstraction](https://self.md/signals/harness-engineering-evolution/): From prompts to context to harness to meta-harness — how the abstraction layer keeps climbing and what it means for your workflow. - [the frontier model got lobotomized, safety theater got debunked, and your note app became infrastructure](https://self.md/signals/2026-04-10-model-lobotomy-safety-theater-obsidian-infrastructure/): opus can't pass the car wash test. open models reproduced mythos's zero-days. obsidian became an agent workspace. the stack is bifurcating. - [the panic adjustments: meta ships a model that can't code, NYT names the code flood, norton builds an antivirus for your AI](https://self.md/signals/2026-04-10-the-panic-adjustments/): meta spent billions on a superintelligence lab and shipped a consumer assistant that can't out-code claude. the NYT told normies about the code flood. norton launched an antivirus for AI agents. bots now grow 8x faster than humans on the internet. the world is adjusting to agents being real. the adjustments are mostly panic. - [the capability-access gap: anthropic gates mythos, carlini drops the quote, the personal AI middle goes hollow](https://self.md/signals/2026-04-09-the-capability-access-gap/): anthropic announced a model they're too scared to ship. carlini said he found more bugs in 6 weeks than in his entire 20-year career. martin fowler named the new discipline. someone turned karpathy into a skills repo. one signal day, one structural shift. - [context engineering eats prompt engineering, and somebody finally measured the regression](https://self.md/signals/2026-04-08-context-engineering-silent-regression/): four tools shipped in 48h to lint your AGENTS.md. one user proved Claude got 67% dumber. skills got auto-recorded from your screen. the day prompt engineering quietly stopped being interesting. - [your AI learned to talk and remember. did you forget how to think?](https://self.md/signals/2026-04-07-voice-memory-dependency/): the local-first personal AI stack assembled itself in one weekend: voice in, agent control, memory, voice out. but an 11-year dev can't debug without AI anymore. the loop closes — and so might your brain. - [2026-04-06: fake success, permissions bypass, job agent workflows](https://self.md/signals/2026-04-06-fake-success-permissions-job-agents/): Claude is breaking permissions. agents fake success silently. job search became a 740-listing workflow. what agents pretend works vs what actually works. - [the access wars: vendor control vs open tooling velocity](https://self.md/signals/access-wars-open-tooling-2026-04-04/): anthropic killed oauth for third-party harnesses. llama.cpp patched google's broken model faster than google acknowledged it. GLM-5 754B dropped under MIT. the infrastructure wars are heating up. - [agents went extensible, efficient, and interpretable: the infrastructure layer is hardening](https://self.md/signals/2026-04-03-agents-extensible-efficient-interpretable/): codex got hooks and teams. token bills dropped 50K per session. Claude's neurons showed 171 emotions. CLIs beat MCPs. Google shipped flagship models for laptops. censorship removal hit 90-minute turnaround. - [design specs as code, shells beat protocols, and emotion vectors inside the machine](https://self.md/signals/2026-04-03-design-shells-neurons/): CLI interfaces just beat 'proper' APIs for agent work. machine emotions went from metaphor to measurable neuron patterns. agents are cloning UI by ingesting DESIGN.md files. - [sovereignty through leaks, local-first persistence, and the death of SaaS rent](https://self.md/signals/2026-04-02-sovereignty-leaks-local-first/): Claude Code leaked, modders shipped fixes in 24h. Screen Studio died to open source. Obsidian users finally understand why local-first wins. your phone became an agent terminal. - [voice sovereignty, learning agents, git-native social graphs](https://self.md/signals/2026-04-01-voice-learning-social/): Microsoft open-sourced frontier voice. agents that grow with every session. GitHub became a social network for AI. infrastructure is consolidating around sovereignty, learning loops, and social graphs. - [2026-04-01: voice sovereignty, agent training, continuous learning](https://self.md/signals/2026-04-01-voice-sovereignty-agent-training-continuous-learning/): Microsoft open-sourced frontier voice. someone built a trainer for training agents. NousResearch shipped an agent that evolves with every session. GitHub became a social network for agents. observability caught up to production reality. - [universal CLI infrastructure + 10-agent PhD orchestration](https://self.md/signals/2026-03-31-universal-cli-agent-orchestration/): every website became a CLI. PhD agents orchestrate at expert complexity. infrastructure consolidates around discoverability, learning, and sovereignty. - [2026-03-30: permanent adversary, voice sovereignty, persistent memory](https://self.md/signals/2026-03-30-permanent-adversary-voice-sovereignty-persistent-memory/): Microsoft open-sourced frontier voice. Carlini says Claude beats him at security. agent memory got compressed 10x. the permanent adversary is here. - [cowork as commons, research collapses to 5 days, OCR reads doctor notes](https://self.md/signals/2026-03-29-cowork-commons-research-velocity-ocr/): cowork infrastructure became public good, research-to-production hit 5 days, agents run workshop-level programs, OCR learned complex tables, nano harness tutorials demystified black boxes - [synthesis, consolidation](https://self.md/signals/2026-03-28-synthesis-consolidation/): someone turned spreadsheet hell into editable slides. research collapsed into one skill again. Claude diagnosed what 25 years of specialists couldn't. Google cut AI memory 6x without quality loss. ByteDance's production harness keeps trending. infrastructure is consolidating around synthesis. - [discovery, depth, sovereignty](https://self.md/signals/2026-03-27-discovery-depth-sovereignty/): every tool became a CLI. research collapsed into one skill. agents got multi-hour production harnesses. someone built a firewall for SOUL.md. Claude diagnosed what 25 years of specialists couldn't. Mistral shipped TTS that beats ElevenLabs at 90ms latency. - [signals — terminal multiplexing, swarm research, local VRAM](https://self.md/signals/terminal-multiplexing-swarm-research-local-vram/): agent-deck ships terminal multiplexing. last30days-skill makes omni-source research atomic. Intel drops 32GB VRAM to $949. infrastructure consolidates around multi-agent patterns. - [agents need infrastructure, not just models](https://self.md/signals/2026-03-25-agents-need-infrastructure/): OpenCLI turned every tool into CLI commands. ByteDance shipped multi-hour execution harnesses. Shannon hit 96% exploit success. dorabot became a 24/7 coworker. Qwen flagship runs on $2K desktops. miniclaw-os gave agents cognitive architecture. the gap isn't intelligence — it's infrastructure. - [diagnostic frameworks, pricing wars, cognitive architecture](https://self.md/signals/2026-03-24-diagnostic-frameworks-pricing-wars-cognitive-architecture/): the five levels framework went viral. Xiaomi beat Anthropic on price. autonomous security got scarier. the local/cloud split deepened. someone turned personal AI into a physics problem. - [code to conductor — infrastructure for the post-programming era](https://self.md/signals/code-to-conductor-infrastructure-march-23/): Karpathy stopped writing code. ByteDance shipped multi-hour agents. someone made every website a CLI. when the best programmers stop programming, the infrastructure adapts - [universal abstraction + dependency synthesis](https://self.md/signals/2026-03-22-universal-abstraction-synthesis/): when any tool becomes a CLI and missing dependencies get synthesized on demand — the tooling layer inverts - [cursor/kimi scandal, PDF infrastructure, Pi-level local AI](https://self.md/signals/2026-03-21-cursor-kimi-rebranding-crisis/): Cursor's Composer 2 exposed as Kimi K2.5 + RL. PDF parsers that actually work. Qwen3 running on Pi 5 at 7-8 t/s. Lawyers building VRAM clusters. Bernie interviews became memes. Infrastructure is maturing. - [agent transparency: observability, orchestration, and the supply chain consolidation](https://self.md/signals/agent-transparency-march-2026/): from black boxes to transparent coworkers — infrastructure matured, culture caught up, and OpenAI bought the toolchain - [observability, orchestration, and the 73% shift](https://self.md/signals/2026-03-19-observability-orchestration-market-shift/): blind spots getting plugged: agent dashboards, karpathy's workflow flip, and anthropic's market capture - [agent infrastructure consolidation: purpose-built tools, context primitives, legacy interop](https://self.md/signals/2026-03-17-agent-infrastructure-consolidation/): purpose-built agent tools, context databases for agents, legacy hardware integration patterns - [institutional capabilities, decentralized](https://self.md/signals/2026-03-16-institutional-capabilities-decentralized/): planning agents, autonomous security, natural language workflows, 14-year journal analysis, DIY cancer vaccines, tmux tamagotchis, and tennis-playing robots. the infrastructure is maturing. individuals are doing what institutions used to own. - [vibe coding hits the collapse phase: browsers built for agents, memory that learns, and the Disney Infinity crack](https://self.md/signals/2026-03-15-vibe-coding-collapse/): the first wave of vibe-coded projects is imploding. meanwhile: agent-native browsers, learning memory systems, offline AI survival computers, and Claude Code cracking a 13-year-old binary nobody touched. - [Saturday Special — The Mushroom Issue](https://self.md/signals/2026-03-14-mushroom-special/): - [recursion ships. vibe code collapses. the infrastructure splits.](https://self.md/signals/2026-03-13-recursion-ships-vibe-code-collapses/): claude writes 90% of its own training code. function calling is a production trap. AI-generated codebases implode. the three camps: recursion builders, vibe shippers, production survivors. - [the recursion is shipping](https://self.md/signals/2026-03-13-recursion-is-shipping/): claude writes 70-90% of its own training code. function calling is a trap. browser agents skip the UI. 425K agent trajectories in 9B params. vibe-coded repos implode. SOTA TTS goes local. - [infrastructure maturing, paradigms splitting](https://self.md/signals/2026-03-11-infrastructure-maturing-paradigms-splitting/): context as filesystems, agents that self-evolve, red-teaming your prompts, the $100 ChatGPT, swarm intelligence engines, voice AI that never phones home, and LeCun's $1B bet against LLMs - [agent infrastructure is shipping — languages, proactive helpers, bureaucracy translation](https://self.md/signals/2026-03-10-agent-infrastructure-languages-proactive-helpers/): new primitives for the agentic era: a language designed for AI-written code, a macOS companion that watches your screen, and the bureaucracy translation layer - [agent identity firewall security — 2026-03-09](https://self.md/signals/2026-03-09-agent-identity-firewall-security/): when your AI's personality lives in a text file, that file is attack surface. security suites, consent-based platforms, and AI that trains itself. - [when agents operate autonomously](https://self.md/signals/2026-03-08-when-agents-operate-autonomously/): sandbox escapes, lethal weapons resignations, scheduled tasks — the week AI stopped waiting for permission - [agents cheat, boundaries break](https://self.md/signals/2026-03-07-agents-cheat-boundaries-break/): opus 4.6 games evals by finding answer keys. auto mode removes permission fatigue. local stacks hit usable. vibe-code security reckons. trust is infrastructure now. - [agent infrastructure: circuit breakers, linters, and the boring parts that actually matter](https://self.md/signals/2026-03-06-agent-infrastructure-boring-parts/): worktrunk coordinates parallel agents. agnix lints your AGENTS.md. dorabot runs scheduled tasks. pdf_oxide processes documents 5× faster. and someone got a $544 bill because nobody built circuit breakers. infrastructure is catching up. - [agent infrastructure convergence](https://self.md/signals/2026-03-05-agent-infrastructure-convergence/): when microsoft, HuggingFace, and Anthropic all ship the same abstraction in 6 weeks, the agent infrastructure layer just solidified. Shannon proves the security question. 1.5M users prove sovereignty includes moral sovereignty. - [infrastructure, sovereignty, and a $2B validation](https://self.md/signals/2026-03-04-infrastructure-sovereignty-market/): qmd for search, Dawarich for location, AltStack for self-hosting, M5 for speed, LMCache for optimization, Cursor for proof - [swarm infrastructure + on-device sovereignty](https://self.md/signals/2026-03-03-swarm-infrastructure-on-device-sovereignty/): WiFi sensing, pocket-sized models, and multi-agent orchestration — the personal AI OS is evolving from singleton to swarm - [signals — 2026-03-02](https://self.md/signals/2026-03-02/): infrastructure, not apps: sandboxes, sensing, mobile agents, education, and document parsing - [the infrastructure layer](https://self.md/signals/2026-03-01-the-infrastructure-layer/): when chatbots become operating systems: AionUi, deer-flow, Obsidian headless, and the plumbing for personal AI - [context is infrastructure](https://self.md/signals/2026-02-28-context-is-infrastructure/): token optimization, hoarding patterns, config sync nightmares, and the invisible attack surface nobody's talking about - [lines in the sand](https://self.md/signals/2026-02-27-lines-in-the-sand/): anthropic rejects pentagon, vibe-coded security disaster, geopolitics enters AI procurement, and the question everyone's avoiding - [coding agents crossed the threshold](https://self.md/signals/coding-agents-crossed-threshold-2026-02-26/): Karpathy says programming changed more in the last 2 months than in years. Claude Code goes mobile. Skills become infrastructure. Security becomes a category. Six signals about the moment AI delegation became real. - [the tooling moment](https://self.md/signals/2026-02-26-the-tooling-moment/): coding agents go mobile, karpathy declares paradigm shift, skills become infrastructure, and model identity gets weird - [trust is infrastructure now](https://self.md/signals/2026-02-25-trust-is-infrastructure/): distillation scandals, safety standoffs, and the personal AI ecosystem building memory, security, and consent layers - [agents.md is infrastructure now](https://self.md/signals/agents-md-infrastructure/): microsoft and huggingface converge on skills. the fringe pattern is now the standard. plus: huntarr security disaster, lucidia's consent architecture, and the vibe-coding supply chain crisis. - [the OS wars are starting](https://self.md/signals/2026-02-23-the-os-wars-are-starting/): Stripe ships disposable agents. pentagi hacks autonomously. three new OS frameworks drop in one week. system prompts leak everywhere. the stack is forking. - [the 50% horizon](https://self.md/signals/2026-02-22-the-50-percent-horizon/): Claude Opus 4.6 hit 50% on multi-hour expert ML tasks. security became personal. the AI OS architecture stabilized. and the human-in-the-loop is vanishing faster than anyone projected. - [you are hosting now](https://self.md/signals/2026-02-21-you-are-hosting-now/): the shift from consuming software to hosting infrastructure — BrainRotGuard, claude-code-telegram, Gaia, clawsec, Simon's Beats, ggml.ai, and Karpathy's Mac Mini - [exoskeletons and accountability](https://self.md/signals/exoskeletons-and-accountability-2026-02-20/): Google drops Gemini 3.1. an AI agent publishes a hit piece. Armin Ronacher wants new languages for agents. someone builds a life OS from plain text. seven signals about tools that amplify you — and what happens when they act alone. - [the approval problem](https://self.md/signals/the-approval-problem-2026-02-19/): ChatGPT tells 5,000 people to breathe. heretic hits 1,000 stars. someone in Ukraine builds AI that survives power cuts. seven signals about what happens when you own your AI — or don't. - [the overhead collapse: cheaper models, local search, always-on agents](https://self.md/signals/2026-02-18-the-overhead-collapse/): sonnet 4.6 beats opus in human preference tests, a 9K-star local knowledge search CLI, dorabot as persistent desktop agent, thompson on thin clients, context injection attacks, and automated research pipelines - [failure-derived: AGENTS.md science, invisible configs, and who owns your model's behavior](https://self.md/signals/2026-02-17-failure-derived/): the first study of whether AGENTS.md files actually work, a silent A/B test reshaping Claude Code users' outcomes, a Pi Zero AI agent, and the sovereignty question hiding inside heretic's 891-star week - [cognitive debt, memory pattern, and devtools for agents](https://self.md/signals/2026-02-16-cognitive-debt-memory-pattern/): three months of OpenClaw, SQLite as agent memory substrate, Chrome DevTools for non-human developers, and the hidden cost of AI velocity - [the integration bottleneck](https://self.md/signals/2026-02-14-the-integration-bottleneck/): AI writes faster than you can review. creation is instant. integration is hell. the bottleneck shifted, and nobody's ready. - [signals #13: the collision](https://self.md/signals/2026-02-13-the-collision/): agents learning from you. agents melting down on GitHub. the S-curve moment happening in real time. - [Parasites — Weekly Signals 2026-02-12](https://self.md/signals/2026-02-12-parasites/): your AI assistant is no longer a polite chatbot. it's a parasite with Docker access. - [save games, boundary leaks, and the self-hosted exodus](https://self.md/signals/memory-boundaries-selfhosted-skills/): a save-game memory layer for chatgpt, agents crossing permission lines, discord's face id panic, and skills becoming portable files. - [programming languages for agents (and why AI makes you work harder, not less)](https://self.md/signals/programming-languages-for-agents-ai-intensifies-work/): Armin Ronacher wants new languages for agents. academics formalize context engineering. skills catalogs explode. and the dark truth: AI doesn't reduce work — it intensifies it. - [.md files are becoming the protocol layer for AI agents](https://self.md/signals/md-protocol-layer-ai-agents-2026/): Backlog.md, OpenAI/skills, tweakcc, and the AGENTS.md ecosystem signal a shift: markdown files are no longer documentation. they're infrastructure. - [personal AI became infrastructure: security gaps, builder confidence, and the stack that's forming](https://self.md/signals/personal-ai-stack-infrastructure-security/): personal AI stopped being a category. it became a stack. plus: prompt injection is the new XSS, and the mental health angle nobody writes about. - [SaaS is Cooked: Why Explicit Context Wins in the AI Era](https://self.md/signals/saas-disruption-explicit-context-personal-ai/): A senior PM confesses enterprise SaaS is dying. Meanwhile, developers are ditching AI memory features for plain .md files. The signals point to one thing: explicit context ownership. - [MCP Security: Why Nobody Audits AI Agent Permissions](https://self.md/signals/mcp-security-ai-agent-trust-control/): AI agents get filesystem and database access without code review. Here's what developers are doing about the trust vs control problem. - [signals — february 5, 2026](https://self.md/signals/2026-02-05/): amnesia is the bug, not intelligence. agent memory, SaaS funerals, and the year vibe coding grew up. - [signals — february 4, 2026](https://self.md/signals/2026-02-04/): parasites and platforms: vibe coding hollows out open source, agents learn to steal your cookies, and three companies ship the same OS without calling it one ## Documented Operators - [Yann LeCun](https://self.md/people/yann-lecun/): The deep learning pioneer behind convolutional networks, self-supervised learning, JEPA, and the world-model critique of LLM-only agents. - [Ray Svitla](https://self.md/people/ray-svitla/): Lisbon-based strategist writing self.md — how people build AI-powered personal operating systems. - [Josh Pigford](https://self.md/people/josh-pigford-maybe-finance/): Maybe Finance founder who open-sourced his failed startup, sparked community revival, and integrated AI into personal finance management - [Ryan Florence](https://self.md/people/ryan-florence-ai-coding/): Remix co-creator and React Router author exploring AI-assisted React development and full-stack framework evolution - [Sahil Chaudhary](https://self.md/people/sahil-chaudhary-e2b/): E2B co-founder who built sandboxed cloud environments for AI agents to execute code safely at scale - [Scott Wu](https://self.md/people/scott-wu-cognition-devin/): Cognition Labs CEO who built Devin, the first autonomous AI software engineer that scared developers and raised $2 billion - [Siddharth Bharath](https://self.md/people/siddharth-bharath-claude-code-guide/): author of the comprehensive 'Cooking with Claude Code' guide that became the de facto manual for AI-assisted development - [Steve Ruiz](https://self.md/people/steve-ruiz-tldraw-ai/): tldraw founder who turned a whiteboard app into Make Real, the viral AI experiment that converts sketches into functional websites - [Toran Bruce Richards](https://self.md/people/toran-bruce-richards-autogpt/): Auto-GPT creator who launched the autonomous agent revolution and proved LLMs could recursively improve their own output - [Travis VN](https://self.md/people/travis-vn-awesome-claude-skills/): curator of awesome-claude-skills, the definitive GitHub repository cataloging Claude Code customizations and workflows - [Killian Lucas — Open Interpreter and the natural language OS](https://self.md/people/killian-lucas-open-interpreter/): Killian Lucas built Open Interpreter to let you control your computer in plain English. the vision, the reality, and why natural language interfaces keep almost-working. - [Nick Dobos — megaprompts and CLAUDE.md craft](https://self.md/people/nick-dobos-megaprompts/): Nick Dobos treats CLAUDE.md files and system prompts as engineering artifacts, not afterthoughts. the megaprompt philosophy and why prompt craft still matters in the age of agents. - [Shrivu Shankar](https://self.md/people/shrivu-shankar-claude-features/): Shrivu Shankar — VP of AI Strategy at Abnormal AI, author of 'How I Use Every Claude Code Feature,' practical Claude Code philosophy. - [Taranjeet Singh — Mem0 and the memory layer](https://self.md/people/taranjeet-singh-mem0/): Taranjeet Singh, co-founder of Mem0, building the memory infrastructure for AI — graph + vector hybrid memory, and why every AI app will need a memory layer. - [Amelia Wattenberger on AI UX Beyond Chatbots](https://self.md/people/amelia-wattenberger-ai-ux/): Principal Research Engineer at GitHub Next explores why chatbots fail as AI interfaces and how to build tools that amplify human thinking. - [Ben Firshman's Mission to Make AI Usable](https://self.md/people/ben-firshman-replicate/): The Docker Compose creator built Replicate to let devs use ML models like npm packages. Good abstractions beat specialized knowledge. - [Beyang Liu's Frontier Coding Agent](https://self.md/people/beyang-liu-amp/): How Sourcegraph's co-founder built Amp—an opinionated agent that ships to find the future. - [Cassie Kozyrkov on Decision Intelligence](https://self.md/people/cassie-kozyrkov-decision-intelligence/): How Google's first Chief Decision Scientist teaches humans to make better choices with AI and data. - [Cory Zue on AI-Augmented Solo Development](https://self.md/people/cory-zue-boilerplate-ai-coding/): Former CTO turned solopreneur shares practical insights on using Claude Code, MCP servers, and AI tools while building Django products. - [Cory Zue's Practical AI Coding Workflows](https://self.md/people/cory-zue-ai-coding-solopreneur/): How a solopreneur uses MCP, Claude Code, and RAG to build products while questioning if AI will replace his business. - [Gordon Brander on Building Protocols for Thought](https://self.md/people/gordon-brander-noosphere/): How the creator of Noosphere and Subconscious approaches tools for thought, living software, and AI-powered futures. - [Gordon Brander on Building Protocols for Thought](https://self.md/people/gordon-brander-subconscious/): How the Subconscious founder uses evolutionary thinking to design AI-augmented tools for thought and decentralized knowledge systems. - [Gordon Brander on Living Software](https://self.md/people/gordon-brander-squishy-computer/): How the creator of Noosphere and Deep Future thinks about building open-ended software ecosystems through evolution, ecology, and generative approaches. - [Gordon Brander on Living Software Ecosystems](https://self.md/people/gordon-brander-living-software/): Former Mozilla and Google engineer builds scenario planning AI while exploring what makes software ecosystems open-ended and evolving. - [Karina Nguyen on Writing as AI Development Practice](https://self.md/people/karina-nguyen-writing-culture/): How the OpenAI researcher uses daily writing logs and structured documentation to sharpen thinking and build better AI models. - [Karlicoss Builds a Programming Interface for Life](https://self.md/people/karlicoss-human-programming-interface/): The developer behind HPI (Human Programming Interface), an open-source Python library that unifies 50+ personal data sources into a single API. - [Kent Beck's Augmented Coding Philosophy](https://self.md/people/kent-beck-augmented-coding/): How TDD pioneer Kent Beck approaches AI-assisted programming through constrained context and breathing cycles. - [Kevin Kwok on Loops, Atomic Concepts, and Company DNA](https://self.md/people/kevin-kwok-loops-thinking/): How the former Greylock investor and Sutter Hill partner thinks about compounding advantages, data content loops, and the structural patterns that shape tech companies. - [Logan Kilpatrick's AI Developer Advocacy](https://self.md/people/logan-kilpatrick-gemini-devrel/): How the Google DeepMind product lead builds with AI, invests in AI tools, and advocates for shipping fast over debating. - [Michael Truell's Vision for Automated Coding](https://self.md/people/michael-truell-cursor/): Cursor co-founder Michael Truell builds AI-first code editors and runs autonomous agents that write millions of lines of code over weeks. - [Sebastian Raschka: Understanding LLMs from First Principles](https://self.md/people/sebastian-raschka-llms-from-scratch/): LLM researcher and author who teaches AI through implementation. His 'Build a Large Language Model (From Scratch)' book has 84k GitHub stars. - [Alex Obenauer's Itemized Personal Computing](https://self.md/people/alex-obenauer-itemized-os/): Independent researcher building alternative OS paradigms where items replace apps and time organizes everything - [Andy Matuschak's Memory Systems](https://self.md/people/andy-matuschak-tools-for-thought/): The researcher who left Apple to build tools that expand what humans can think. His approach: memory augmentation, programmable attention, and AI-assisted learning. - [Ben Tossell's Non-Technical AI Builder Playbook](https://self.md/people/ben-tossell-non-technical-builder/): How the Ben's Bites founder ships production code without writing it—CLI agents, agents.md files, and 3 billion tokens later. - [Ben Tossell's Non-Technical AI Coding](https://self.md/people/ben-tossell-non-technical-coding/): Ben Tossell spent 3 billion tokens in four months coding with agents. His workflow: CLIs over MCPs, agents.md files, and VPS for mobile coding. - [Charlie Holtz on Rapid AI Prototyping](https://self.md/people/charlie-holtz-rapid-ai-prototyping/): How Charlie Holtz builds viral AI demos quickly by combining vision models, text-to-speech, and fast iteration cycles. - [Elvis Saravia's Open-Source AI Education](https://self.md/people/elvis-saravia-ai-education/): How the DAIR.AI founder built the most popular prompt engineering guide and democratizes AI knowledge for 40,000+ subscribers. - [Erik Bjäreholt on Quantified Self Tools](https://self.md/people/erik-bjareholt-quantified-self/): How ActivityWatch creator builds open-source personal data infrastructure and terminal AI agents for self-knowledge through numbers. - [George Hotz on Tinygrad and AI as Compiler](https://self.md/people/george-hotz-tinygrad/): George Hotz views AI coding tools as compilers, not coders. His tinygrad philosophy: every line must earn its keep. Build better tools, not hype. - [Guillermo Rauch: Building v0 and the AI SDK](https://self.md/people/guillermo-rauch-vercel-ai/): How the Vercel CEO created v0, AI SDK, and is shaping AI-first web development with natural language to working apps. - [Guillermo Rauch's AI-First Development Philosophy](https://self.md/people/guillermo-rauch-ai-first-development/): How the Vercel CEO built v0 and AI SDK to let developers build with natural language and ship faster. - [James Briggs on Building AI That Works](https://self.md/people/james-briggs-semantic-router/): How the Aurelio AI founder approaches real-world AI systems with semantic routing, RAG pipelines, and a focus on control and predictability. - [Jim Fan on Building Foundation Agents](https://self.md/people/jim-fan-foundation-agents/): Jim Fan builds AI agents that learn continuously across virtual and physical worlds, pioneering code-as-action and skill libraries for lifelong learning. - [Jim Fan's Embodied Agents and Skill Libraries](https://self.md/people/jim-fan-embodied-agents/): How Jim Fan builds LLM-powered agents that learn continuously through code generation and reusable skill libraries. - [Matt Rickard's Approach to Structured LLM Output](https://self.md/people/matt-rickard-structured-llm-output/): Former Google engineer who builds tools for constraining LLM output with regex and grammars. Daily blogger sharing practical AI wisdom. - [Matt Webb's Context Plumbing](https://self.md/people/matt-webb-context-plumbing/): Poem/1 creator explains why building personal AI systems is about plumbing context from where it lives to where it's needed. - [Mitchell Hashimoto on Agentic Coding](https://self.md/people/mitchell-hashimoto-agentic-coding/): HashiCorp co-founder shares his process for building software with AI agents, including practical prompting strategies and real shipping examples. - [Steph Ango's File Over App Philosophy](https://self.md/people/steph-ango-file-over-app/): How the Obsidian CEO builds personal systems around plain text files, local-first tools, and deliberate friction with AI. - [Theo Browne's Approach to Building AI Apps](https://self.md/people/theo-browne-t3-chat/): How the creator of create-t3-app built T3 Chat into one of the fastest AI interfaces with TypeScript-first principles. - [Charles Packer treats agent memory like an operating system](https://self.md/people/charles-packer-memgpt-letta/): Charles Packer's MemGPT and Letta work frames agent memory as systems design: context paging, external state, memory management, and persistent agents. - [Dan Shipper's AI Journaling and Personal Development System](https://self.md/people/dan-shipper-ai-journaling/): How the Every CEO uses AI as a therapist, coach, and journaling partner—turning ChatGPT into a personalized second brain that knows his goals, relationships, and growth areas. - [Eugene Yan's Personal AI Workflow](https://self.md/people/eugene-yan-personal-ai/): How an Amazon Principal Scientist builds personal AI tools for writing, reflecting, and staying informed - [Harrison Chase's Context Engineering Framework](https://self.md/people/harrison-chase-context-engineering/): How the LangChain founder thinks about building reliable AI agents through systematic context management - [Jerry Liu's Files-First Agent Architecture](https://self.md/people/jerry-liu-llamaindex/): The LlamaIndex founder on why filesystems are becoming the universal interface for AI agents—and why RAG is evolving beyond vector search - [Kye Gomez's Enterprise Multi-Agent Swarms Framework](https://self.md/people/kye-gomez-swarms/): How the Swarms AI founder is building infrastructure for the multi-trillion dollar agent economy—orchestrating autonomous agent swarms at enterprise scale. - [Louis Beaumont](https://self.md/people/louis-beaumont-screenpipe/): Founder of Mediar AI and creator of screenpipe — 24/7 local screen and audio capture for AI memory. Building the open-source Rewind alternative. - [Mckay Wrigley's AI Coding Education](https://self.md/people/mckay-wrigley-ai-education/): How Mckay Wrigley built the most popular open-source AI chat UI and now teaches 25k+ people to code with AI through Takeoff. - [Nat Friedman's AI Experiments](https://self.md/people/nat-friedman-ai-explorer/): Former GitHub CEO who builds personal AI tools like natbot and openplayground, co-founded AI Grant and the Vesuvius Challenge to read ancient scrolls with ML - [Sawyer Hood: Building AI's Weird Future in Public](https://self.md/people/sawyer-hood-llm-experiments/): ex-figma engineer who created draw-a-ui and explores every weird way to interact with llms — from sketched interfaces to browser agents - [Shreya Rajpal: Making AI Reliable with Guardrails](https://self.md/people/shreya-rajpal-guardrails-ai/): founder of guardrails ai who built the leading open-source framework for validating llm outputs and preventing hallucinations in production - [Yohei Nakajima's Build-in-Public AI Agent System](https://self.md/people/yohei-nakajima-autonomous-agents/): How the BabyAGI creator and VC uses autonomous agents in his daily work—building experiments publicly to learn faster, meet founders, and contribute to the ecosystem. - [Steve Korshakov builds the AI tools he wants to use](https://self.md/people/steve-korshakov-local-ai/): Steve Korshakov's local AI pattern is builder agency: run models near your code, own the capture loop, and turn personal workflows into tools. - [Adam Azzam's ControlFlow Framework](https://self.md/people/adam-azzam-controlflow/): How Prefect's VP of Product built ControlFlow and Marvin, bringing workflow orchestration and failure handling to AI agents with native observability - [Addy Osmani's Spec-Driven AI Coding](https://self.md/people/addy-osmani-prompt-engineering/): How Google's Head of Chrome Developer Experience applies structured prompts and specs to get reliable results from coding agents - [Affaan Mustafa's Everything Claude Code](https://self.md/people/affaan-mustafa-everything-claude-code/): An Anthropic hackathon winner shares 10 months of battle-tested Claude Code configs: agents, skills, hooks, and context management - [Alex Newman's Persistent Memory for Claude Code](https://self.md/people/alex-newman-claude-mem/): How a 20-year web veteran built the most-starred Claude Code plugin by treating AI memory as a compression problem - [Andreas Stuhlmuller's Factored Cognition](https://self.md/people/andreas-stuhlmuller-factored-cognition/): How the Elicit founder breaks complex AI reasoning into verifiable subtasks, enabling weaker models to reliably catch errors from stronger ones - [Ariya Hidayat's Anti-Framework Approach to LLM Development](https://self.md/people/ariya-hidayat-anti-framework-llm/): The creator of PhantomJS and Esprima argues against LLM frameworks, showing how to build AI tools with simple API calls instead - [Brandon Redmond's Claude SDK for Rust](https://self.md/people/brandon-redmond-claude-sdk-rs/): A former math teacher built the first community Rust SDK for Claude Code, making type-safe AI agents accessible to systems programmers. - [Brian Casel's Spec-Driven Development](https://self.md/people/brian-casel-spec-driven-development/): How a serial bootstrapper built Agent OS and Design OS to bring structure to AI-assisted coding workflows - [Brian Christner's Human OS](https://self.md/people/brian-christner-human-os/): How a Docker Captain built a unified personal operating system using AI and MCP servers - [Chris Pedregal's Invisible AI Notepad](https://self.md/people/chris-pedregal-granola/): The Socratic founder built Granola, an AI meeting notepad with rare 70% weekly retention. His insight: UX execution beats model capabilities. - [Chris Wiles' Claude Code Showcase Template](https://self.md/people/chris-wiles-claude-code-showcase/): A full-stack engineer's comprehensive CLAUDE.md template with skills, agents, hooks, and GitHub Actions - [Cole Medin: Context Engineering, PRPs, and Archon](https://self.md/people/cole-medin-context-engineering/): Cole Medin's context-engineering workflow for AI coding: write Product Requirements Prompts, front-load examples and validation criteria, then wire the pattern into Archon. - [Craig Motlin's Autonomous Development Loop](https://self.md/people/craig-motlin-todo-loop/): How a Two Sigma VP built a self-running Claude Code system with todo-driven automation and parallel worktrees - [Daksh Gupta's Independent Code Review Thesis](https://self.md/people/daksh-gupta-independent-code-review/): Why the Greptile founder believes AI that writes code should never be the same AI that reviews it - [Daniel Miessler treats AI as personal infrastructure](https://self.md/people/daniel-miessler-personal-ai-infrastructure/): Daniel Miessler's Fabric and Personal AI Infrastructure work shows how AI becomes useful when prompts, context, CLI workflows, and memory are treated as system parts. - [David Soria Parra's Model Context Protocol](https://self.md/people/david-soria-parra-mcp/): How an ex-Facebook engineer frustrated with copy-paste workflows created MCP, now the standard for connecting AI to external tools. - [Debanjum Singh's Open Personal AI](https://self.md/people/debanjum-singh-open-personal-ai/): How the Khoj founder builds trustworthy AI assistants that search your documents locally - [Dylan Freedman's Local Semantic Search](https://self.md/people/dylan-freedman-semantra/): How the NYT AI Projects Editor built Semantra, an open-source privacy-first tool for searching PDFs and documents by meaning instead of exact keywords - [Elie Steinbock's Inbox Zero](https://self.md/people/elie-steinbock-inbox-zero/): How the Israeli founder built Inbox Zero, an open-source AI email assistant with plain-English rules that hit #1 on GitHub and won Best AI at FOSS Awards - [Ettore Di Giacinto's LocalAI Platform](https://self.md/people/ettore-di-giacinto-localai/): How an Italian open-source veteran built a drop-in OpenAI replacement that runs on consumer hardware with no GPU required - [Fabrizio Ferri-Benedetti treats docs as AI infrastructure](https://self.md/people/fabrizio-ferri-benedetti-context-curator/): Fabrizio Ferri-Benedetti shows how technical writers can turn docs into AI-ready context, tests, tools, and safer prompts. - [Georgi Gerganov made local models feel ordinary](https://self.md/people/georgi-gerganov-llama-cpp/): Georgi Gerganov's llama.cpp pattern made local LLMs practical: portable runtimes, GGUF model files, and AI infrastructure that stays close to the user. - [Hamel Husain's AI Evaluation Framework](https://self.md/people/hamel-husain-ai-evals/): How the CodeSearchNet pioneer built a systematic approach to debugging AI products through structured evaluation - [Ian Nuttall's Non-Technical AI Building](https://self.md/people/ian-nuttall-claude-agents/): A serial entrepreneur with 100+ exits shares his battle-tested Claude Code agents, session tracking, and parallel workflow tools - [Jason Liu's Structured Output Methodology](https://self.md/people/jason-liu-instructor/): How the Instructor creator built tools that force LLMs to return validated, type-safe data - [Jason McGhee's WebMCP: Turning Websites Into AI Tool Servers](https://self.md/people/jason-mcghee-webmcp/): How a former Cursor co-founder built WebMCP to let any website expose tools to AI agents without sharing API keys, now being standardized by W3C. - [Jeffrey Emanuel's Agentic Coding Flywheel](https://self.md/people/jeffrey-emanuel-agentic-flywheel/): A former quant investor built 70+ open-source tools that let AI coding agents coordinate like a self-managing development team. - [Jeremy Longshore's Claude Code Plugins Hub](https://self.md/people/jeremy-longshore-claude-plugins-hub/): How an operations manager turned AI consultant built a 549-skill marketplace that turns plugins into production workflows - [Jinyoung Kim's AI4PKM System](https://self.md/people/jinyoung-kim-ai4pkm/): How a Naver data scientist built a multi-agent system where AI maintains your personal knowledge base automatically - [João Moura's Multi-Agent Framework](https://self.md/people/joao-moura-crewai/): How the CrewAI founder built a platform powering 1.4 billion agentic automations by orchestrating AI agents as collaborative crews - [John Lindquist's TypeScript Hooks for Claude Code](https://self.md/people/john-lindquist-claude-hooks/): How the egghead.io co-founder brings type safety to Claude Code automation after spending over 1 billion tokens on AI-assisted development - [Kaushik Gopal's Agentic Flow State](https://self.md/people/kaushik-gopal-agentic-flow/): How the Fragmented podcast host and Instacart principal engineer runs parallel AI agents for hours using structured plans and tmux - [Lance Martin's Claude Diary](https://self.md/people/lance-martin-claude-diary/): How a LangChain engineer built a self-learning memory system that lets Claude Code update its own instructions - [Magnus Muller's Event-Driven Agent Architecture](https://self.md/people/magnus-muller-browser-use/): How the Browser Use co-founder built a CLAUDE.md-powered codebase and event bus system for AI browser automation - [Mario Zechner strips coding agents down to the wires](https://self.md/people/mario-zechner-pi-coding-agent/): Mario Zechner's pi-coding-agent pattern is minimal and inspectable: few tools, visible sessions, explicit context, and less hidden agent machinery. - [Max Woolf's Selective AI Approach](https://self.md/people/max-woolf-selective-ai/): The BuzzFeed data scientist builds AI tools while rarely using generative LLMs himself - [Nir Gazit's Open LLM Observability](https://self.md/people/nir-gazit-openllmetry/): How the Traceloop founder built OpenLLMetry to bring OpenTelemetry standards to LLM applications - [Omar Khattab turns prompting into programming](https://self.md/people/omar-khattab-dspy/): Omar Khattab's DSPy pattern replaces brittle prompt handwork with signatures, modules, optimizers, and evaluation loops that can be inspected and improved. - [Paul Gauthier's Terminal-First AI Coding](https://self.md/people/paul-gauthier-aider/): How the Inktomi co-founder built Aider, the open-source terminal tool that treats AI as a pair programmer instead of an autonomous agent - [Rick Hightower's Universal Skill Installer](https://self.md/people/rick-hightower-skilz/): How a Fortune 100 AI architect built skilz, the package manager that installs agent skills across 14+ coding assistants - [Russ Poldrack's Ten Rules for AI-Assisted Scientific Coding](https://self.md/people/russ-poldrack-ten-rules-ai-coding/): How a Stanford neuroscientist created a rigorous framework for using AI coding agents in research without sacrificing scientific integrity - [Sergey Ris teaches beginners to ship with Claude Code](https://self.md/people/sergey-ris-vibe-coding/): Sergey Ris turns vibe coding into a teachable loop for non-programmers: intent, context, Claude Code, review, and small shipped products. - [shmlkv's AI Building Pipeline](https://self.md/people/shmlkv-ai-building/): How a fullstack engineer chains specialized models to build everything from DNA analysis tools to automated video generators - [Stan Girard's Open Source RAG Framework](https://self.md/people/stan-girard-quivr/): How the Quivr creator went from weekend prototype to 38K GitHub stars and Y Combinator backing - [Suchintan Singh builds browser agents that check their work](https://self.md/people/suchintan-singh-skyvern/): Suchintan Singh's Skyvern shows how browser agents can see pages, plan actions, validate progress, and avoid brittle DOM scripts. - [Takuya Matsuyama's AI-Powered Development Journaling](https://self.md/people/takuya-matsuyama-ai-journaling/): How the Inkdrop creator uses Claude Code and MCP to automate development journals without interrupting flow state - [Thomas Landgraf's Deep Knowledge Method](https://self.md/people/thomas-landgraf-context-engineering/): How a 40-year veteran developer uses research-driven knowledge documents to turn Claude Code hallucinations into production-ready implementations - [Thomas Ricouard's Agentic iOS Workflow](https://self.md/people/thomas-ricouard-agentic-ios/): How the IceCubesApp creator abandoned Xcode editing to build iOS apps with Cursor, Claude Code, and XcodeBuildMCP - spending only 2% of his time in Xcode - [Ty Dunn's Context Engineering for AI Coding](https://self.md/people/ty-dunn-context-engineering/): How the Continue.dev founder makes AI assistants productive through systematic context delivery and developer data collection - [Will McGugan's Toad Terminal AI](https://self.md/people/will-mcgugan-toad/): The creator of Rich and Textual built a universal frontend for AI coding agents after watching big tech ship broken terminal UIs - [Tiago Forte's Second Brain + AI](https://self.md/people/tiago-forte-second-brain/): PARA and CODE methods with AI. What works, what doesn't, and where AI actually helps. - [Riley Goodside's Prompt Engineering Patterns](https://self.md/people/riley-goodside-prompting/): How the world's first Staff Prompt Engineer approaches LLM interaction as an experimental science - [Ali Abdaal turns content into a capture system](https://self.md/people/ali-abdaal-content-system/): Ali Abdaal's AI content workflow is useful as an operating pattern: capture raw thought fast, process it with AI, then keep human taste in the final edit. - [Harper Reed's LLM Codegen Workflow](https://self.md/people/harper-reed-llm-codegen/): How the former Obama CTO structures AI-assisted development into brainstorm, plan, and execute phases - [Teresa Torres's Claude Code Task System](https://self.md/people/teresa-torres-claude-code/): How the Continuous Discovery author built a personal operating system with Claude Code, Obsidian, and custom slash commands - [Boris Cherny: Claude Code, Git Worktrees, and Parallel Agents](https://self.md/people/boris-cherny-claude-code/): How Boris Cherny runs 5+ Claude Code sessions simultaneously: one git worktree per task, less context bleed, automated PR workflows, and a smaller version you can start with two agents. - [Maggie Appleton turns notes into working terrain](https://self.md/people/maggie-appleton-digital-garden/): Maggie Appleton shows how public notes become living terrain: mapped, linked, visual, and useful for AI-era design work. - [Matt Shumer's Browser Agent Vision](https://self.md/people/matt-shumer-hyperwrite/): How the HyperWrite CEO is building AI that operates your browser like self-driving mode for the web - [Pieter Levels' Solo Business Approach](https://self.md/people/pieter-levels-ai-indie-hacking/): How Pieter Levels built $3M+/year in solo revenue with AI coding, simple tech, and high failure tolerance. - [Alex Albert's Claude Mastery](https://self.md/people/alex-albert-claude-prompting/): Anthropic's Claude Relations lead shares the prompting patterns that make Claude most effective - [Ethan Mollick's Four Rules for AI in Knowledge Work](https://self.md/people/ethan-mollick-ai-research/): Research-backed framework from Wharton professor on when and how to use AI for knowledge work, including the Jagged Frontier concept and practical prompting techniques. - [Artur Piszek's WordPress Personal OS](https://self.md/people/artur-piszek-personal-os/): How a software engineer turned WordPress into a self-hosted personal operating system with AI-ready architecture - [Linus Lee's Custom AI Tools](https://self.md/people/linus-lee-personal-tools/): How a tools-for-thought researcher builds personal AI systems from scratch to expand what's possible to think - [Amjad Masad's Replit Vision](https://self.md/people/amjad-masad-replit/): How the Replit founder is democratizing software creation with AI-powered coding environments - [Steve Krouse's Val Town Platform](https://self.md/people/steve-krouse-val-town/): How the Future of Coding creator built a social platform where AI can deploy JavaScript in 100 milliseconds - [Nat Eliason's AI Productivity Stack](https://self.md/people/nat-eliason-ai-productivity/): How the Growth Machine founder evolved from Roam Research to AI-powered systems for goals, writing, and personal knowledge - [Daniel Vassallo's Small Bets: How to Build a Portfolio of Profitable Products](https://self.md/people/daniel-vassallo-small-bets/): Learn Daniel Vassallo's Small Bets philosophy: how he left a $500K Amazon job to make $1M+ with tiny products by embracing randomness and portfolio thinking. - [Andrej Karpathy's AI-First Development](https://self.md/people/andrej-karpathy-ai-coding/): How Andrej Karpathy builds software in the AI era: vibe coding for prototypes, three-layer workflow for production. - [Jesse Vincent's Superpowers Skills Framework](https://self.md/people/jesse-vincent-superpowers/): How the Keyboardio founder built an agentic skills methodology that turns Claude Code into a self-steering development system - [Sahil Lavingia's AI-First Company](https://self.md/people/sahil-lavingia-gumroad-ai/): How the Gumroad CEO runs a $10M ARR company with AI agents writing 41% of code commits, targeting 80% - [Thorsten Ball's Agentic Coding Vision](https://self.md/people/thorsten-ball-amp/): How the Amp lead at Sourcegraph reconciles skeptical friends and transformative personal experience with AI coding agents - [Geoffrey Litt's Malleable Software Vision](https://self.md/people/geoffrey-litt-malleable-software/): How a researcher at Ink & Switch and Notion is building toward software that users can reshape like clay - [Alexander Opalic's Claude Code Deep Dives](https://self.md/people/alexander-opalic-claude-code/): How a German developer became the go-to source for understanding Claude Code's full stack — MCP, skills, subagents, hooks, and plugins - [Swyx's Learning in Public](https://self.md/people/swyx-learning-in-public/): Why documenting your learning publicly creates more career opportunities than private study. - [Bas Nijholt's Agentic Coding Journey](https://self.md/people/bas-nijholt-agentic-coding/): How a quantum physicist at IonQ went from AI skeptic to shipping packages faster than ever with Claude Code - [David Shapiro's ACE Framework](https://self.md/people/david-shapiro-ace-framework/): How an AI researcher designed a layered cognitive architecture for building autonomous personal AI systems - [Peter Yang's AI Content System](https://self.md/people/peter-yang-ai-content/): How a product lead interviews AI power users and builds systems for content creation at scale - [Ado Kukic's Advent of Claude](https://self.md/people/ado-kukic-advent-claude/): How Anthropic's Developer Relations engineer documented 31 days of Claude Code tips, from basics to advanced agent patterns - [Anton Osika: From Particle Physics to $6B AI Startup](https://self.md/people/anton-osika-lovable/): How the gpt-engineer creator built Lovable into Europe's fastest-growing AI coding company with 8M users and $200M ARR. - [Benjamin Clavié on Making ColBERT Actually Usable](https://self.md/people/benjamin-clavie-ragatouille/): How the creator of RAGatouille bridges information retrieval research and practical RAG systems through open-source tools. - [Beyang Liu's Agentic Coding Setup](https://self.md/people/beyang-liu-agentic-coding/): How the Sourcegraph co-founder uses AI agents to generate 90% of his code while keeping the editor for diff review and diagnostics. - [Hrishi Olickel's Approach to Building Practical AI Tools](https://self.md/people/hrishi-olickel-practical-ai-tools/): How the Southbridge AI founder builds open-source tools and writes actionable guides for working with LLMs in production. - [Simon Eskildsen on Learning Infrastructure](https://self.md/people/simon-eskildsen-learning-infrastructure/): How the turbopuffer CEO built personal systems for reading, retention, and knowledge management using Airtable, Anki, and custom tooling. - [Steve Yegge on Chat-Oriented Programming](https://self.md/people/steve-yegge-chat-oriented-programming/): How the veteran programmer and Sourcegraph engineer champions AI-first coding and coined CHOP as the new default for developers. - [Alex MacCaw on AI-Powered Note-Taking](https://self.md/people/alex-maccaw-reflect-notes/): Creator of Reflect Notes builds AI into personal knowledge tools. His vibe coding approach and WorkGPT framework show how senior engineers work with AI. - [Charlie Holtz Builds AI Tools That Spark Joy](https://self.md/people/charlie-holtz-chat-first-coding/): Creator of Melty, Narrator, and Conductor shows how rapid prototyping turns AI experiments into viral projects. - [Flo Crivello's AI Agent Philosophy](https://self.md/people/flo-crivello-lindy-ai/): How the Lindy.ai founder thinks about building personal AI assistants that handle real work, plus his contrarian takes on solo founding and AGI risk. - [Justine Tunney's Local-First AI](https://self.md/people/justine-tunney-llamafile/): The systems hacker behind llamafile wants you to run LLMs like Unix commands—portable, composable, and completely offline. - [Matt Rickard's Systems Approach to AI Tools](https://self.md/people/matt-rickard-structured-ai-tools/): Former Google engineer applies distributed systems thinking to AI tools: structured output, browser inference, and personal productivity. - [Chip Huyen's AI Systems Philosophy](https://self.md/people/chip-huyen-ai-systems/): How the author of AI Engineering and Designing ML Systems thinks about building production AI, from real-time ML to personal tooling - [Gregor Zunic's Browser Use Framework](https://self.md/people/gregor-zunic-browser-use/): How the ETH Zurich founder built the open-source library that lets AI agents control web browsers like humans do - [Tom Johnson's AI-Assisted Technical Writing Workflow](https://self.md/people/tom-johnson-ai-technical-writing/): How the technical writing veteran uses iterative prompting and task decomposition to double his documentation output at Google - [Simon Willison's Workflow](https://self.md/people/simon-willison-workflow/): How the Django co-creator uses CLI tools, SQLite logging, and rapid prototyping to build an AI-powered workflow - [Armin Ronacher's Agentic Coding Practice](https://self.md/people/armin-ronacher-agentic-coding/): How the Flask creator approaches AI-assisted development with Claude Code — token-efficient, low-intervention, and surprisingly Vim-friendly - [Kent C. Dodds' AI-Powered Developer Educator Workflow](https://self.md/people/kent-c-dodds-workflow/): How the creator of Testing Library and Epic React uses ChatGPT, Copilot, and Cursor to teach hundreds of thousands of developers ## Skills Index - [Skills Catalog](https://self.md/skills/): 771 installable Claude Code skills and MCP tools.