self.md radar — 2026-04-30
The useful AI story today is not another chat trick; it is AI claiming a real surface — the IDE, the contribution policy, and the factory line.
Two editor bets turn human-plus-agent work into a first-class product mode, two maintainer communities tighten the rules on AI-mediated contribution, and Figure backs its humanoid story with throughput and yield numbers instead of another kitchen demo.
1. IDEs now assume you switch between typing and delegating
sources:
what happened: Zed declared 1.0 on April 29, with users able to run multiple agents in parallel and ACP opening the editor to Claude Agent, Codex, OpenCode, and Cursor. Zed is also launching Zed for Business with centralized billing, role-based access controls, and team management. JetBrains, on the same week, explicitly framed two valid ways to write code — classic line-by-line work and delegated AI work — and said agents should run in opt-in modes that can work for hours, while a human stays responsible for shipped code and the core coding experience is not compromised.
why this matters: The IDE is being redesigned as a shared control room where humans can type, delegate, inspect, and take ownership without leaving the same product. “Editor adds AI” is over; the editor itself is the agent surface.
2. Maintainers are putting AI terms in writing
sources:
what happened: Zig’s code of conduct now forbids LLMs in issues, pull requests, and bug-tracker comments. Loris Cro frames the rationale as “contributor poker” — maintainers are betting on future trusted contributors, not just accepting code-shaped output — and points to hallucinated drive-by PRs and giant low-trust first submissions as the immediate trigger. Zulip takes the softer line: contributors may use AI tools, but must personally understand and explain the change, PRs that violate this are closed without review, AI-generated messages are banned in the development community, and contributors are told to prefer primary sources over pasted LLM answers.
why this matters: The new governance fight is not abstract safety rhetoric. Maintainers are directly pricing reviewer time, trust formation, and communication quality into project policy — and writing it into the rulebook.
3. Figure starts publishing factory metrics
sources:
what happened: Figure says BotQ has delivered more than 350 third-generation robots and that production has gone from 1 Figure 03 per day to 1 per hour — a 24x throughput jump in under 120 days — across more than 150 networked workstations driven by custom manufacturing execution software. End-of-line first-pass yield is now above 80%, battery-line first-pass yield is 99.3%, and the company has produced over 9,000 actuators across more than 10 SKUs. Figure ties the scale-up to Helix 02, whose whole-body controller replaces 109,504 lines of hand-engineered C++ and trains on more than 1,000 hours of human motion data.
why this matters: The company is finally talking like a manufacturer and fleet operator, not a lab showing off a clip. Yield, throughput, and SKU counts are the right vocabulary for taking a humanoid pitch seriously.
supporting links
- Amazon Quick — native desktop assistant reaching into local files, calendar, communications, and deliverable generation.
- portless v0.12 — stable named local URLs aimed at both humans and agents.
- Structured Output Benchmark — a concrete benchmark for deterministic model outputs.
- Figure BotQ — deeper manufacturing context behind today’s production numbers.
left on the table
- Talkie — already shipped on 2026-04-28; no repeat allowed.
- Warp — real traction, but the same-day delta is too thin for a main slot.
- Obsidian 3D graph — visual novelty without enough workflow change to anchor a signal.
- RobotEra L7 deployments — same robotics bucket as Figure and weaker sourcing.
- Hound — off-thesis for an AI-surface edition.