One factory, four products: why we split the SDLC into a chain
PFactory plans, AIFactory builds, TFactory verifies, CFactory watches. Here's why that's a family, not a monolith.
The industry narrative for 2026 is the “autonomous software factory” — agents that take an idea to production with humans in the loop only where it matters. Most attempts build that as one big monolith. We bet on the opposite: a chain of focused, composable products, each excellent on its own, each handing off to the next.
The PARR loop, as products
- PFactory — Prepare. Governed planning, grounded in your real infrastructure (cloud, Backstage, wikis), with review gates that cite their evidence and a human-approval gate before anything ships.
- AIFactory — Act. Spec-first execution in isolated worktrees, multi-provider, enterprise-grade, able to delegate the coding phase to other agents.
- TFactory — Reflect. Autonomous tests graded on a five-signal verdict — coverage delta, stability, mutation, lint, semantic relevance — so you get tests you can trust.
- CFactory — Review. The control tower: one cockpit over all three, with an agentic copilot that explains and steers.
Why a chain beats a monolith
Three reasons. Adoption: you can take TFactory for QA, or PFactory for governed planning in front of any coding agent, without buying the whole stack. Focus: each product competes on its own merits — and TFactory’s verdict rigor or PFactory’s governance is far sharper than a generalist monolith. Trust: the boundaries between products are exactly where human-in-the-loop gates belong.
The connective tissue
A chain is only as good as its handoffs. The piece we’re building now is the
spine: a shared correlation key (the humble GitHub issue number) threading
plan → code → branch/PR → tests, a normalized completion-event schema, and a
clean port map. Once that’s in place, CFactory can show — and steer — the whole
pipeline from one place.
That’s the work. Follow it on the roadmap and across the repos.