Inspiration

Its lesson is that the browser is already basically you, and that impersonation is the missing primitive. The next thing I am building copies its shape.

It was the kernel I had been trying to build myself, so I stopped building a runtime and became a userland on top of it.

It convinced me in early 2026 that local agents are the way. It runs sixteen of my agents, and the fleet view it lacks became the definition of what I am building instead.

A coding agent plus a folder plus built-up scripts and data is what a prototype is now. It changed the architecture of my personal system: the agent is the shell, the tools are system calls.

A repository defines an agent and the agent lives on its own machine as a persistent process. That is the shape I had wanted to build.

It made me want a language that represents the same thing in a hundred lines, and it showed that git is where convergence happens.

A primitive that deserves first-class support. My system's daily cycle is literally it.

It gave me the reframe that markdown is the high-level language for AI systems.

Launching my own AI commands from a keystroke in 2024 is where I first used the word cyborg about myself. Two years later it was a project name.

Feeding arbitrary text through a model into it gave me spaced repetition on anything. That pattern, a tool's killer feature plus AI equals a primitive you own, is what my personal system's organs generalize.

The shape is right even where the app was not: collect, enrich, queue, work through it. The lanes on my board are built on it.

It is the one tracker that actually gets used, and the goals layer I built is the thing it lacks.

One whole version of my product compiled classes to them. The current one is a folder, and the sites still run on Cloudflare.

It limited how much I could schematize, which was good. An old runtime was built on it because realtime felt complex, and the elegant-looking solution turned out to be the complicated one.

Its Dataview plugin gave me the idea of a queryable document structure. I wanted plain files rather than the app.

It is object-based notes done well. It needs a server, and I wanted an object store that holds code and data.

Unix-like composable prompts, and the first version of my own command-line tools started as its patterns.

It inspired a lighter connected notebook, the second life of an idea I had first called microlog.

The problem is not too many abstractions but weak ones. My intelligence-design thesis, explicit context manipulation and no hidden prompts, was defined against it.

I want its distribution machine, made for developers: open-source, hackable software that looks crafted.

Two versions of my product were Notion-like block editors. In 2026 I decided not to compete with Notion, Slack, Discord, or Linear, all of which are adding agents.

They help people build apps, not AI apps, because AI apps need intelligence design and there is no framework for that yet.

In 2019 the goal was to design your own workflow through code. In 2026 I described the same idea as computable history.

A background process that invokes the cognitive abilities of different agents over a JSON-RPC API. Two years later I used a standard protocol for exactly that.

Kernel and user contexts, and processes that talk to each other. The name on this domain comes from it.

A smart-contract language I did not finish. In 2026 I decided my old ideas were early rather than bad, and that it was time to reintegrate them.

The question I keep asking: what is the modern HyperCard, and why are useful AI apps still stuck in prompt wrappers?