How BPOS builds software
BPOS is the platform we build our software with. Scroll to follow one goal through it: planned, built, checked, and learned from. Every step links to the article behind it.
The steps, in words
-
AI that builds software, step by step.
You give BPOS a goal. It writes down, in order, the steps every way to that goal has to go through, and each step breaks down the same way.
Read: A goal has necessary stages -
Every step has written rules.
Each job gets the rules it has to meet before it counts as done.
Read: Every job is a contract -
Workers made for one step.
Each step is done by AI workers made for that step only. When the step is done, they are gone.
Read: Nodes, not agents -
Read only what you need.
Before starting, each worker reads only what this job needs.
Read: What an AI agent reads before it acts -
Stop instead of looping.
A job that stops getting closer is stopped and flagged, not retried forever.
Read: When to stop an agent loop -
Every result is tested.
Each rule is checked, then the build and the tests run. What the rules don’t name is not checked.
Read: The quality gate -
A person has the last word.
A person looks once, at the finished step as a whole.
Read: Where a person decides -
Everything is written down.
Every move every worker makes is recorded as it happens, so that after each job BPOS can work out why it went the way it did.
Read: The Gettier problem for AI agents -
The path is what stays.
What stays is the path the workers took: which instructions they followed, and whether it worked. The instructions are pages anyone can read.
Read: Beyond SKILL.md -
No retraining. Any good model.
Paths that led to good work get stronger, paths that led to a stopped job get weaker, like an ant trail. The knowledge is in the paths, not in the model, so switching to a better model is a swap, not a rebuild.
Read: Procedural memory for AI agents -
See it all in one place.
One view of what is done, what is waiting for a person, and what was decided.