Matt Webb's Context Plumbing

Matt Webb has been blogging at Interconnected since February 2000. Twenty-five years. That’s not a typo. He’s maintained a 304-week posting streak (and counting). When someone’s been thinking out loud in public for that long, you pay attention when they start building AI systems.

Webb isn’t your typical tech founder. He co-founded BERG, the design studio that shipped products into MoMA. He spent 2020-2022 working with Google’s AI research group. He co-wrote Mind Hacks, translated into seven languages. And somewhere along the way, he shipped an AI poetry clock called Poem/1 that got into the New York Times.

But the interesting stuff? It’s happening right now.

The core insight: Context plumbing

Webb’s framework for thinking about AI systems comes down to two words: intent and context.

Intent is what you want. The big revelation of LLMs is that computers can finally understand human intent without forcing you through menus and forms. You can just… say what you mean.

Context is everything the AI needs to actually help you. The world knowledge baked into the model, sure. But also: documentation about available tools. What you’ve done before. Time of day. The document you’re working on. Whether this task is part of something bigger.

Here’s where Webb gets specific. He calls it context plumbing:

The job of making an agent run really well is to move the context to where it needs to be. Essentially copying data out of one database and putting it into another one — but as a continuous process.

This isn’t abstract philosophy. He’s been deep in code building systems on Cloudflare with “context flowing between all kinds of entities and AI agents and sub-agents running where they need to run.”

The framing matters. Web 2.0 had CRUD apps. AI systems need plumbing diagrams.

Agents belong in your Reminders app

Webb’s most provocative take: the natural home for AI agents isn’t some fancy new interface. It’s your to-do list.

His reasoning is practical. Agents sequence tasks into steps. They pause when they need clarification. They create plans (Claude Code already does this — structured lists of tasks stored in text files). They run in parallel while you’re doing other stuff.

Sound familiar? That’s a task manager.

What we’re talking about is a multiplayer to-do list which AI agents can use too. Really this is just the Reminders app on my iPhone?

He points to Linear for Agents as the model — agents appearing as teammates with avatars you can tag. Apple, he argues, is sitting on a goldmine. Extend Reminders to let agents pick up tasks, show when something’s blocked, and you’ve got the coordination surface for personal AI.

Team Augmented Environments

Webb frames the AI future as a choice between cyborgs (AI in your head via glasses and earbuds) and rooms (AI in your environment). He’s firmly on Team Rooms:

So much of what I care about happens in small groups in physical space: family time, team collaboration, all the rest.

That’s what his new startup Inanimate is exploring. Room-scale AI. What’s the OS for physical spaces when interfaces become intent-first?

Poem/1 was a prototype of this thinking — an e-ink clock that displays a new AI-generated poem every minute. Not a screen you stare at. An object in your space that happens to be intelligent.

The design studio mindset

What separates Webb from most AI builders is the design lens. He’s not starting with “what can the technology do?” He’s starting with “what should computing feel like?”

His writing is full of these reframes:

• Progress bars, not notifications (AI brings back long-running processes)
• Agents are teammates (give them avatars, let people tag them)
• DWIM interfaces — Do What I Mean as a design principle

He traces the smartphone beating the desktop to this kind of thinking. Touching what you see beats moving a mouse that moves a pointer. Closer to intent wins.

The resources

Blog: interconnected.org/home — 25 years of archives, search for AI-tagged posts

Key posts:

• Context plumbing — the core framework
• The natural home for AI agents is your Reminders app — coordination surfaces
• The future of interfaces is Do What I Mean — intent-first design

Products:

• Poem/1 — AI poetry clock (117% funded on Kickstarter)
• Galactic Compass — iOS app pointing to the center of the galaxy
• Inanimate — ambient AI startup (early stage)

Book: Mind Hacks — cognitive science tips and tricks

GitHub: github.com/genmon — includes aboutfeeds, his RSS getting-started guide

What to steal

1. Think in plumbing diagrams. When building AI systems, map where context originates and where it needs to flow. That’s your architecture.

2. Use existing coordination surfaces. Before building a new interface for your agents, ask: could this just be a to-do list? A shared document? Something people already check?

3. Design for physical space. Not every AI interaction needs a screen. What would an AI object be in the room where the thing actually happens?

4. Maintain long-running context. Webb’s 304-week blogging streak is itself a form of context engineering — public memory that compounds.

5. Start with intent. What does the person actually want? How do you get closer to that moment of desire? Everything else is bureaucracy.

Webb’s been thinking about human-computer interaction since before most AI engineers were born. Now he’s building systems that put those decades of design thinking into practice. Worth watching.