Been working on a quantitative trading platform for about 2.5 years, off and on, then full-time for the last eight months. I started using agents seriously around 30,000 lines of code and found an agentic workflow I could trust, and I turned that into Pullboard.
Pullboard builds Pullboard itself. (MIT):
[https://github.com/pullboard-dev/pullboard](https://github.com/pullboard-dev/pullboard)
Essential Loop: claim -> build -> verify (different agent reviews, see website.)
Primitive Concepts:
https://pullboard.dev/tutorials/ I absolutely needed to trust my agents to write proper code and worked out an Agentic Item Lifecycle (AILC) that works for me, within an Agentic Development Lifecycle (ADLC). My flavor of all this may not be "correct" but it does work, and it helped me build out a quant trading system. That system is now roughly 500K LOC and runs across more than a dozen Mac minis. It trades autonomously and runs dozens of strategies, and it must work at market open.
That aside, what makes this workflow valuable is the "Agentic Institutional Knowledge" that gets built up over time outside the codebase in the Pullboard history; e.g., items, shouts and doctrine all get referenced, discussed and lessons remain available to future agents instead of disappearing into old chats. The best part is I hardly re-explain myself.
You do have to steer agents and make decisions, but this is quite nearly autonomous development. Anyways it is early and open source. See: Pullboard.dev or try,
npm install -g pullboard pullboard tour
Feedback welcome. Thank you!!
Comments URL: [https://news.ycombinator.com/item?id=50025023](https://news.ycombinator.com/item?id=50025023)
Points: 1