Dispatcher Assistant

I type the idea, move the card, and a local agent runs. Daily personal system, not a product.

Overview

This one isn't a product. It's how I work, made explicit. I type the idea. I move the card. The bridge does the rest — locally, on my machine, every day. Constraint: a system of one. It fits me perfectly, which is also why it isn't a product.

How building software changed

Then: IDE + autocomplete
Recently: terminal-native AI coding
Now: agent orchestration

The fork in the road

The industry's orchestration is going cloud, repo connected, worktrees spun up, PRs raised, all remote.

Mine is local. The fleet already lives on my machine: Claude Code · Codex · opencode · Antigravity · Grok. They're already there, already authenticated, already holding my context. The missing piece was never compute. It was that the idea and the executor lived in two different places, and I was the transport layer between them.

The design

Dispatcher Assistant is a Notion custom agent with exactly one job: take the idea off a card, and hand it to my terminal.

The architecture: I drive everything from a Notion board; a Notion-blind MCP bridge relays only directory, agent, mode and prompt to whichever CLI runs locally, and reports state back, so the executor never has to know about Notion.

The interaction is two things I was already doing anyway, I type the idea into Notion, and I move the card. That's it. The trigger isn't a new button or a new habit; it's the thing I do to track the work regardless. The best automation is invisible because it's attached to a gesture you already make. (Same principle as DesignMode: don't teach a new behaviour, capture the one that's already there.)

One loop, Draft → Planning: I move a card to Todo, the dispatcher builds the prompt from the card's title and body, the bridge runs the chosen CLI in read-only plan mode, and the plan comes back into the card for me to review.

Why this

Execution is no longer the hard part. The scarce skill is choosing what to build. I spend most of my time in ideation and prioritisation. Building is moving a card from left to right.

That sentence is either a boast or a thesis depending on how you read it, so let me be precise about which. It's not "AI writes my code." It's that the bottleneck moved, and if the bottleneck moved, the tooling around it should move too. My tools should be optimised for deciding, because deciding is where the hours now go. Most tooling is still optimised for typing.

Over the horizon

A year on Ray-Ban Meta taught me voice-first works, but it's closed.

Next: Halo glasses, custom firmware, talking straight to my Notion AI. Not for photos or notifications, a pure AI inference layer over my personal knowledge.

The honest bit

This is a system of one, built for many. It's tuned to my machine, my fleet, my Notion. It works because it fits exactly one person perfectly, which is also the reason it isn't a product.