What four days of building looks like

I'm starting a four-day sprint today. Not a hackathon, not a launch crunch. Just four days where the only thing I'm allowed to work on is making the homepage of getkommit.ai accurate.

Right now the homepage promises that Kommit captures your thinking and keeps it present every time your AI picks up the work. About 70% of that is true today. The remaining 30% is the gap between what's shipping in the alpha and what the copy claims. I'm closing that gap by Friday.

The gap I'm closing

This is harder than it sounds, because I have ten feature branches in flight that I'm not allowed to touch this week.

There's a governance UI I'm three weeks into. There's a billing integration I started last Tuesday. There's an Azure runtime branch blocking a deploy fix. There's a dashboard redesign I keep itching to merge. Each of these is real work. None of them advance the homepage promise. So they all stay parked.

The discipline is the strategy. The single biggest risk to this sprint isn't technical. It's that something on the parked list pulls me back in mid-week. I've watched myself do this on every project I've ever shipped. Some feature feels productive but isn't on the critical path, and three days disappear before I notice.

The plan

Here's what I'm actually doing this week.

Monday

Merge a feature branch I've been sitting on for two weeks that adds an event-sourced decision graph. Build a UI that lets users see, in plain English, the decisions they made during a spec session. The data structure already exists. The view that makes it visible doesn't.

Tuesday

Expose that decision rationale through MCP, the protocol Claude Code and Cursor use to query Kommit. Right now the agents can read the what of your captured thinking. After Tuesday they can read the why.

Wednesday

Make the loop observable. When an agent queries your decisions in real time, you should be able to see it happening. Which agent, what it asked for, when. The mechanics are already working. What's missing is the surface where you can watch the agent picking up where you left off.

Thursday

No new features. Polish, demo recording, smoke-testing every claim on the homepage. By end of day I record a 20-second loop showing the full cycle: capture decisions, agent queries them live, activity reflects back. That's the demo that goes on the homepage.

Friday

Ship.

Why I'm writing it down

I'm writing this on Monday because I want a record of what I committed to at the start of the week. If by Wednesday I'm working on something that isn't in the plan above, the plan failed and I should call it. The version of me writing this on Monday wants to remind the version of me on Wednesday that the constraint is the strategy.

The interesting thing about working with AI tools is that the bottleneck isn't typing anymore. I can ship more code in a day than I used to ship in a week. Which means the bottleneck is what to build. And what to build is decided by who has the discipline to ignore everything that isn't load-bearing.

This week I'm running an experiment on myself. How much can a small team ship when the only constraint is don't drift. I'll tell you what happens.