Develop like every

Develop like Every

Profiled as of 2026-07-11 · consent tier: self-published · full bibliography in references/sources.md. Cultures drift — if this is more than ~6 months old, re-run /dev-like every to refresh.

Core principle

"Each unit of engineering work should make subsequent units easier — not harder" [CEP README]. Roughly 80% of effort goes to planning and review, 20% to execution [guide].

Principles

1. Plan-first, 80/20 — most effort before code [guide]
2. Compound every unit of work into reusable repo knowledge [CEP]
3. Simplify as an explicit pipeline step, not an afterthought [CEP]
4. One person + agents ≈ a team; design workflows for that ratio [podcast transcript]
5. Boring stack, radical process [SF Ruby keynote]
6. Agent-native parity — if a human can run tests, read logs, or open a PR, the agent should be able to as well [guide]
7. Safety nets over gatekeeping — trust comes from tests, automated review, monitoring, and rollback, not line-by-line babysitting [guide]
8. Make institutional knowledge discoverable — solved problems live in docs/solutions/, and instruction files must surface that store so future agents find it [ce-compound-refresh]

Workflow

Execute these checkpoints before and during the task. Treat them as required actions, not background description:

Run the six-step loop: brainstorm → plan → work → simplify → review → compound [CEP]. Use the four-beat core cycle — plan → work → review → compound → repeat — and keep simplify as the extra gate the plugin inserts [guide]. At the compound step (/ce-compound), write learnings to docs/solutions/ and use them to ground the next loop — make knowledge accrete in the repo, not in heads [ce-compound]. For the fully autonomous pipeline (/lfg), plan → work → review → PR → watch CI until green. Keep reviewer personas as skill-local prompt assets (29 skills, 0 standalone agents post-migration) [CEP].

See references/stack.md for the stack and references/workflow.md for workflow detail.

Tensions

• CEP's layout migrated twice in ~6 months (agents folded into skills) — mimic the principles, expect the mechanics to drift.
• "Engineers write virtually zero code" is the honest ceiling of the claim; the punchier "100% AI-written" framing outruns the evidence. Repo history shows human review is the load-bearing step. Treat the ratio as directional.

Want a reviewer/pair persona in Every's voice? See personas/every-developer.md — it's reference material. Claude Code users can copy it to .claude/agents/ to run it as a first-class subagent; other harnesses may need their own harness-specific metadata.