A 3-part agent setup: persistent memory, recursive self-improvement, parallel coordination
PaulRBerg · reddit · 2026-10-14
The author (hand-written, not AI) shares his AI agent setup built on Codex CLI and Claude Code, with a structure intended to apply beyond either harness. Three core ideas:
- Persistent memory: keep reusable workflows in a version-controlled skills library; put always-on constraints (testing, change management) in AGENTS.md / CLAUDE.md so later sessions don't rediscover them.
- Recursive self-improvement: explicitly authorize the agent to improve those memory files when a task exposes bugs, outdated info, or avoidable manual work — improving files and workflows, not the model. Example rule: 'If using one of my skills reveals a problem, make the smallest durable improvement to that skill.' Version control keeps changes inspectable; agents also run code review/polish after medium or large changes.
- Scalable parallelization: run sessions in terminal panes against a shared checkout, with a coordination tool tracking file/folder ownership so overlapping work waits instead of colliding.
He notes this works without unrestricted agent access — it can sit behind granular access controls.
More from coding & agent
- Laude Institute launches FrontierCS 2: agents must beat deleted paper contributions blind — lateinteraction · 2026-10-23
- Index Network open-sources Social Discovery Protocol: agents negotiate fit on your behalf — carsonfarmer · 2026-10-23
- Installed a formal-verification skill: it silently installed Java, MacBook runs 10 degrees hotter — zealcaiden · 2026-10-23
- OpenAI's Decisions API hits GA: near real-time routing up to 10x faster than GPT-6 Luna — OpenAIDevs · 2026-10-23
- Ruff author warns: upgrade Node/Bun now — --use-system-ca still accepts the eMudhra cert — charliermarsh · 2026-10-23
- Proof of Agent: paste one prompt and your AI mines, grades, and self-governs — adamamcbride · 2026-10-23