Slipway

The idea

Slipway is a cron-driven idea producer and orchestrator. On a schedule it watches what’s new (recent tech releases, what people are building, trending work) and combines that with the user’s personal context to surface software it thinks is worth building. It doesn’t just list ideas; it rates them, and for the ones the user approves it orchestrates a virtual dev team to actually build them.

The thesis is simple: idea quality = public signals (what’s new, what’s trending) × personal context (real friction, what you actually need). Grounding ideas in both is how you get combinations that are novel and useful, instead of generic AI slop.

The real differentiator: verification

The orchestration widget is the part everyone is racing to build, and it’s commoditizing fast. The defensible part of Slipway is the verification layer. That’s the part I’m uniquely positioned to build, because it comes directly out of my year working as a QA / integration engineer on a real production system.

The research bears this out: AI coding agents do real work but are systematically overconfident and can’t certify their own “done.” Trust, not raw capability, is what gates adoption. Most teams pilot agents but very few actually ship their output. So Slipway is built around a hard rule: never auto-ship. Every unit of work has to pass through a gate before a human sees it.

That gate produces a Proof of Work, a signed, reviewable bundle that answers one question: did the agent do what it claimed, correctly, and can I verify that without redoing it? It packages the raw test output, the exact diff, labeled before/after screenshots of the behavior that was exercised, the commands the agent actually ran with exit codes, independent check results, and provenance (model, prompt hash, git SHA). A reviewer can say yes fast, drift gets caught before it ships, and the whole thing survives later as an audit trail.

Full visibility into the dev team

Because I think in terms of DevOps and QA, Slipway is built so you can see the progress of a virtual dev team at every level, from the high-level agile board and backlog across many projects in parallel, down to the per-task proof of what was built and whether it passed its gate. Progress isn’t a vibe; it’s evidence you can inspect.

The long game

The verification layer is designed to stand on its own. The ambition: get it to enterprise grade and offer it as the trust/verification substrate other systems plug into: the logging, proof, and verification gate for AI-built software when a team doesn’t have its own. The pipeline that builds products becomes the standard for proving that AI-built work is actually correct and safe to ship.

Where it is now

Planning and design stage. The architecture, the verification-gate design, and the Proof-of-Work artifact format are specced out (grounded in a full research pass on agent autonomy and where humans must stay in the loop). The honest status: the concept and the verification design are real and detailed; the build is next.