Indie builder shares multi-model agent pipeline: Opus manages, Astra builds, Grok reviews until 2-3K tests pass
PawelHuryn · x · 2026-11-01
Responding to skepticism about reviewing an entire SaaS build in 2-3 hours, Pawel Huryn breaks down his autonomous build pipeline:
- Default roles: Opus 5.5 manages, Astra builds, Grok 4.7 reviews
- Review loop: after each phase, review runs 10-15 times, often for 24 hours; the fix-test-review cycle repeats until no findings above Low severity remain
- Test volume: a product like staygreen carries at least 2-3K tests spanning unit, integration, and e2e — with agents impersonating real users in the browser
- Visual checks: agents screenshot every affected screen to verify it "looks right," not just that tests pass
He credits the tiny human review time to extensive agentic "suffering" upstream.
More from coding & agent
- Dev observes AI has made systems-heavy games like RTS and strategy easier to build — IanArawjo · 2026-11-01
- Cursor ships Agent SDK, the framework behind Bugbot, with one-command deploy — dean_rie · 2026-11-01
- Selling AI agents to paper-bound Balkan SMBs: how do you close non-tech traditional businesses? — Then-Occasion18 · 2026-11-01
- Watch an AI agent run fully autonomous in YOLO mode: session recording — umeshonai · 2026-11-01
- A 25-bot Grok fleet in 7 neighborhoods on one shared Linux box, charted by its own bot — bfrench · 2026-11-01
- Claude autonomously directs an AI music video: 80 renders, ~12 GPU-hours, just 8% of a $20 plan — Cheap_Credit_3957 · 2026-11-01