Slipway
A scheduled pipeline that suggests software worth building, then runs a team of AI agents to build it behind a verification gate. Nothing ships until the work is proven and a person signs off.
- Claude Agent SDK
- Claude API
- QA / Verification
- Orchestration
The idea
Slipway runs on a schedule. It looks at what’s new in tech, combines that with my own context, and suggests software that might be worth building. It rates each idea, and for the ones I approve it hands the work to a team of AI agents.
The part I care most about: verification
This comes straight out of my day job in QA. AI coding agents do real work, but they’re overconfident and can’t be trusted to say when they’re done. So Slipway has one hard rule: it never ships on its own. Every task has to pass a gate before a person looks at it.
The gate produces a Proof of Work bundle that answers one question: did the agent do what it said, and can I check that without redoing it? The bundle holds the test output, the exact diff, before and after screenshots, the commands the agent ran with their exit codes, independent check results, and where it all came from (model, prompt hash, git SHA). A reviewer can approve quickly, and the bundle stays around as an audit trail.
Seeing the whole team
You can follow the work at every level, from a board and backlog across several projects down to the proof for a single task and whether it passed its gate.
Where it is now
In progress. The verification gate and the Proof of Work format are the core of it, and I keep building on them.