YATFA and Factory — two ways past the single agent
Every claim about Factory below links to Factory’s own site, docs or engineering blog. Factory is YATFA’s closest neighbour: both reject the one-agent-does-everything model. They put the structure in different places.
Factory’s bet
Multi-agent structure inside Missions — Missions architecture
A Missionlaunched by you
A well-specified goal
A programmatic runner decomposes it
Feature 1 — its own worker, fresh context
Feature 2 — its own worker, fresh context
Feature 3 — its own worker, fresh context
Validators judge — they don’t fix
Scoped to the Mission you launch
A programmatic runner decomposes a well-specified goal; a worker per feature with a fresh context; validators that judge without fixing.
YATFA’s bet
One continuous rule layer across everything — the category
An organizationcontinuous
Rules in between: your autonomy settings · orchestration without an LLM
Found in your codebase, filed with reasons
Written by one agent
Reviewed by another, then QA
Your gate — or approved under your rules
Lessons folded into knowledge · v1
Then the next one — between tasks too
Nothing to launch: it picks stalled work up again, finds new work in your codebase under your rules, and keeps knowledge that corrects itself.
Factory’s design
A programmatic runner spawns a worker per feature, each with a fresh context.Missions architecture →
YATFA’s design
One continuous rule layer across everything — and work that starts without being launched.The category →
Dimension by dimension
Every claim about another product links to its primary source. Filter to see where they are stronger — said plainly, so the rest is believable.
stronger elsewhere todaya different bet