Funding Skelcrew with client work
Update: the offer has changed since I wrote this. I now set up software factories for small teams and can run them for you. See Work with me for the current details.
I've been experimenting with AI coding tools since April, and in September I started building Skelcrew, my own agent harness. It runs coding agents through tasks, from a short description to working software, with me acting as the team lead rather than the person typing the code.
Starting today, I'm also taking on clients. I'll build software for teams that know what they want but have no developers to build it, and I'll build it with Skelcrew.
Why client work
The honest reason is that Skelcrew needs funding, and client work pays the bills.
The better reason is that tools built in a vacuum tend to solve imaginary problems. Until now I've designed Skelcrew around my own projects, and that only gets me so far. My projects have my habits baked in. A design studio with a stack of mockups and no developers will push the tool in directions I would never think of on my own.
This isn't a new idea. Plenty of good software started as the thing a small consultancy built to do its own client work better. The work keeps the tool honest, and the tool makes the work possible.
What I'm actually building
Most AI coding today happens in a chat window. You describe something, read a long reply, correct it, and do it again. It works, but it's draining, and it doesn't scale to several agents running at once.
Skelcrew makes tasks the unit of work instead of chat. You hand over a task, agents plan and build it, and you get back something you can use. Chat is still there for when you need it, but it's the fallback, not the default.
I've also come to think the goal isn't productivity. Building is already fast. The goal is a way of working with agents that a developer can sustain without burning out: fewer interruptions, less code review as a full time job, more time on the decisions that matter.
What clients get
You send tasks the way you would to a developer: a description, a mockup, a screen recording. I scope them, the agent crew builds them, and I make sure the result holds up. Everything runs async, with written updates and no standing meetings.
We start with a two week pilot on one clearly scoped thing. If it works, we continue month to month. The details are on the work with me page.
Building in public
I'll write here about what I learn: what clients actually ask for, where the agents do well, where they need a firmer hand, and how Skelcrew changes because of it. Some of it will be about the tool, some about the craft, and some about what it means to lead a team that is mostly not human.
If you have something you want built, get in touch.