Most AI coding workflows share the same failure mode: every session reconstructs the whole task from scratch.
You explain the architecture once. You’ll explain it again next session. You made a decision that wasn’t recorded — the AI doesn’t know. You established a code style convention — you’ll need to restate it next time.
This isn’t an AI capability problem. It’s a continuity infrastructure problem.
Every Conversation Is a New Worker
Think about how a human engineer works: design documents, requirement records, code review history, deployment logs. These artifacts mean they don’t have to mentally reconstruct the entire project state every time they sit down.
AI coding tools, by contrast, are like contractors with no memory. Every engagement, you re-deliver the entire context from scratch.
Codex Factory Kit’s design conviction: make memory concrete through durable in-repo artifacts.
Eight Stages, Eight Artifacts
Codex Factory Kit structures a complete task across eight stages:
- Bootstrap — establish
PRODUCT.md, describing goals and constraints - Office Hours — clarify requirements with AI, update
PRODUCT.md - Sprint Conductor — generate
PLAN.mdandTESTPLAN.md - Implementation — execute the plan, AI as subagent
- Review Gate — generate
REVIEW.jsonl, structured review record - QA Runtime — verify, update
TESTPLAN.md - Document Release — update
RELEASE.md - Retro — generate
RETRO.md, record what was learned
Each stage’s output is the next stage’s input. This loop doesn’t rely on AI memory — it relies on facts written to disk.
Why a Lightweight Mode Exists
Not every task needs all eight stages. Running a small one-off fix through the full workflow adds more overhead than it saves.
Codex Factory Kit has a lightweight mode that keeps simple tasks simple. The design principle: larger tasks get more reliable outcomes; smaller tasks stay fast. You shouldn’t have to choose between the two.
What the Factory Metaphor Means
“Factory” isn’t a random comparison. The defining property of a factory is that standardized processes make outputs predictable.
A factory doesn’t depend on a specific worker’s individual capabilities — it depends on the process itself. Codex Factory Kit applies this logic to AI coding: standard artifacts and stages make the outcomes of AI conversations more predictable, letting work across sessions genuinely compound instead of restart.
Who This Is For
This isn’t for users whose workflow is a series of small, isolated edits. It’s for:
- Developers using Codex for serious engineering work on real repositories
- Tasks that need to span multiple sessions coherently
- Teams that want planning, review, and QA to have an explicit, queryable record
If your workflow is “ask a question, get an answer, move on,” Codex Factory Kit may be over-engineered for you. If your workflow is “use AI to advance a complex project with many moving parts,” it may be the layer you’ve been missing.