Understand the factory
before you choose one.
“AI software factory” is an umbrella term. This atlas separates the room a system occupies, the work it repeats, the boundary where it stops, and the humans still responsible for the outcome.
Find the stopping point.Then ask who proves the result and who owns the failure.
Map the department.
Then compare the factories.
The two useful surfaces have different jobs. System Atlas shows where a capability belongs. Compare factories holds the mission still so the hard edges become visible.
System Atlas
Four rooms make the department legible: Make, Run, Know and Govern. Open a system passport to see its job, evidence, boundary and stance.
Compare factories
Choose a work family, compare candidates doing the same job, and read the red edge: artifact, proof, Day 2, human residue and primary evidence.
Four questions from
logo soup to a shortlist.
Follow this sequence once. After that, jump directly to the surface you need.
Day 1 makes an artifact.
Day 2 owns a living system.
A green agent run, passing build or merged PR is useful evidence—but it is not operational closure. The Day‑2 page explains exactly what earns that label.
See the six-part definition →