SFML  ·  30 Sep 2026

Meet SFML

Developers keep telling me working with AI is joyless or isolating. I don't think that future holds up. Here is my bet on what replaces it.

A millworker carrying a crate labeled SFML v0.1 out of a brick mill beside a waterwheel

Developers keep telling me the same thing: working with AI is either joyless or isolating. For a lot of them, it's both.

If that's what it feels like to work closely with AI, and we're all headed there, the future looks bleak.

I don't think that future holds up.

Why the Bleak Future Fails

People who can leave joyless jobs will. The best and brightest are exactly those people. They'll find or make jobs that work differently. Those ways of working will turn out to be better, and companies looking for an edge will follow. We already see this with solopreneurs, who have taken the isolation and turned it into an advantage.

So the bleak future doesn't make sense, but the future also won't look like the present. The real question is: how will we work differently?

That's the question I've been working on since going full-time on Craftmill, by talking to potential customers and rebuilding prototypes into real products.

My current bet is that company-specific process will be what separates good places to work from the rest. Process and culture are deeply intertwined. For a long time culture was where the leverage was. Today I think process customization has more upside. To be clear, both must be good.

That's why I'm exploring software factories: defined, shareable flows for how code gets made with agents. I think they could change our relationship with agentic coding, and with each other.

So, Meet SFML

SFML is an open specification for how agents and humans collaborate to build software. It breaks an agentic coding flow into composable steps, so a team can see each part and improve it on its own.

SFML is designed to be a shared artifact, just like Terraform. Today developers tune their own private harness configs. With SFML the team can tune the flow together. That means:

  • People who love optimizing agents can focus on improving them.
  • People who love building product can focus on writing good prompts for what gets built.
  • Because the work is split into discrete steps, people can work together from their own areas of expertise, not just chime in at the end.

Help Wanted

Right now SFML is only a spec, and it needs work. v0.1 is out and v0.2 is coming. I'd love your help:

For me, the goal isn't to make software factories happen. It's to help the people working closest with AI find joy in their work and connection with the people they work alongside. Without those two things, companies don't make much sense. To do that, I need perspectives from every side of this. So, let me know what you think.