Dan Pugsley

Building Pitwhiz (AI for motorsports). Solo-built Javu. I ship AI products end-to-end — systems architecture + design craft.

[Dan Pugsley @dan_pugsley]

@naval Thanks for the tip 🙏

[Dan Pugsley @dan_pugsley]

@nivi @naval I think you did that really well with this one. Really enjoyed it, despite everybody talking over each other 😄

[Dan Pugsley @dan_pugsley]

@shaneparrish Where can I get one of these AI kids?

[Dan Pugsley @dan_pugsley]

@AnthropicAI I love Opus 4.6. Running side-by-side (same task) in my codebase with Opus 4.8, the difference is clear: Half the context usage, clear, unbloated replies vs. Opus 4.8's confusing over-engineering. Fable was nice while we had it, but Opus 4.6 just feels so clean.

[Dan Pugsley @dan_pugsley]

Repo: https://t.co/1EB7G8K7SD

[Dan Pugsley @dan_pugsley]

1/ Generic AI gives generic answers. Not because the model is weak, but because it doesn't know you.

Ask "how should I train this week?" and it can't know about your bad knee, or that last week wrecked you. So it hedges into uselessness.

[Dan Pugsley @dan_pugsley]

5/ Open-sourced the framework today: no personal data, just the architecture + a worked demo branch. Fractal Markdown, point any agent at it.

Still refining it, building in public.

Curious what you'd put in yours. Repo in the replies.