# Harness Design, Review Gates, and Always-On CI Agents *By Coding Agents Alpha Tracker • May 6, 2026* Practitioners converged on a durable lesson: the edge in coding agents is shifting from model choice to harness design—shared specs, context trimming, portable interfaces, and explicit review gates. Also in this brief: Cursor's CI autofix, fs-safe, CodexBar 0.24, and the clips worth stealing from. ## 🔥 TOP SIGNAL The durable edge is moving from **model picking** to **harness design**. PI maintainers say tool-call and system-prompt work can move a model's score by ~30-40% [^1], LangChain is pushing ACP so the same agent can survive CLI/TUI/IDE changes [^2], and Harrison Chase argues that the state wrapped around the model—not the model itself—is now the bigger lock-in risk [^3]. Simon Willison's day-to-day workflow is the operational version of that thesis: agents can be black-box reliable for routine tasks, but humans still own security-adjacent review and higher-order judgment [^4]. > "the model is yours to pick. the interface is yours to pick. the harness shouldn’t be the thing that locks you in." [^2] ## ⚡ TRY THIS - **Black-box the boring path; hand-review the risky path.** Give the agent a bounded task like: *"build a JSON API endpoint that runs a SQL query and outputs the results as JSON; add automated tests and documentation."* Simon Willison says that class of work is now reliable enough to treat as a semi-black box—but he still manually reviews anything security-adjacent [^4]. - **Run parallel spikes, not parallel production merges.** Simon's current workflow: fire off a Claude Code web task for one spike, run a second spike in Codex, keep doing other work, then come back and review the prototypes. He says this only became practical once reliability improved enough to reduce review overhead [^4]. - **Use one shared spec, many fresh subagents, and aggressive context trimming.** Max's PI setup starts each subagent from a fresh session with a common-ground plan/spec and a manager session id; the main session surfaces blockers, and `Reduce` strips tool calls/thinking so the active context keeps only user + assistant finals [^1]. - **If the code is wrong, rewrite the spec—not just the prompt.** Salvatore Sanfilippo's Redis-arrays loop: write the spec in Markdown, improve the spec with GPT, generate an implementation, go back to the spec if tests are unsatisfying, then do a manual line-by-line review of the core code [^5]. ## 📡 WHAT SHIPPED - **Cursor CI autofix** — Cursor now offers always-on agents that monitor GitHub, investigate CI root causes, and open PRs with fixes. Setup template: [cursor.com/marketplace/automations/ci-autofix](http://cursor.com/marketplace/automations/ci-autofix) [^6][^7] - **openclaw/fs-safe** — Peter Steinberger shipped a reusable filesystem safety primitive extracted from OpenClaw. The guidance is practical: if your Node app accepts paths from agents, plugins, uploads, configs, or users, use a **root handle** instead of treating string normalization as the security boundary. Docs: [fs-safe.io](https://fs-safe.io) [^8] - **CodexBar 0.24** — New Windsurf, Codebuff, and DeepSeek providers; Copilot multi-account switching; opt-in local storage breakdowns; fixes for hung Codex RPC and redraw battery drain. Release: [github.com/steipete/CodexBar/releases/tag/v0.24](https://github.com/steipete/CodexBar/releases/tag/v0.24) [^9] - **Deep Agents + ACP** — LangChain says `deepagents-acp` can serve any agent and `deepagents-cli --acp` exposes the same harness over ACP, with working frontends like [toad](https://github.com/batrachianai/toad) and JetBrains IDE integration via this [blog post](https://blog.jetbrains.com/ai/2026/04/using-acp-deep-agents-to-demystify-modern-software-engineering/) [^2][^10]. - **Current model/tool preference snapshot** — Simon Willison says Codex has replaced Claude Code for most of his daily use because the latest version is "outstanding" and Claude Code pricing is a trust issue for him [^4]. His current favorite local model runs in about **20GB RAM** on a laptop and feels roughly like frontier capability from **6-12 months ago** [^4]. Harrison Chase adds that **GLM5** feels close enough to Sonnet/Opus for a lot of prototyping that product taste now matters more than squeezing out the absolute best model [^3]. ## 🎬 GO DEEPER - **10:26-11:55 — Simon Willison on vibe coding vs agentic engineering.** Best short reset on where agents belong in real software work: personal tools are one thing; production systems touching other people's data need a stricter bar. Watch: [YouTube](https://www.youtube.com/watch?v=D76X-96tpTY) [^4] [![High Leverage - Ep. #9, The AI Coding Paradigm Shift with Simon Willison](https://img.youtube.com/vi/D76X-96tpTY/hqdefault.jpg)](https://youtube.com/watch?v=D76X-96tpTY&t=625) *High Leverage - Ep. #9, The AI Coding Paradigm Shift with Simon Willison (10:25)* - **12:35-13:50 — PI's shared-spec + blocker handoff pattern.** Best concrete demo in today's pile of a main session steering fresh-context subagents: every worker reads the same plan/spec, and the main session surfaces blockers so the human can drop straight into the right subagent. Watch: [YouTube](https://www.youtube.com/watch?v=BQ_Es8k650I) [^1] - **8:16-9:34 — Harrison Chase on memory as the real lock-in.** If your stack is quietly starting to depend on provider-managed memory, this is the clip to watch before that hardens into architecture. Watch: [YouTube](https://www.youtube.com/watch?v=lgMTA5GU7TE) [^3] [![The Rise of Agentic AI with Harrison Chase (LangChain) + Rajeev Dham (Sapphire Ventures)](https://img.youtube.com/vi/lgMTA5GU7TE/hqdefault.jpg)](https://youtube.com/watch?v=lgMTA5GU7TE&t=495) *The Rise of Agentic AI with Harrison Chase (LangChain) + Rajeev Dham (Sapphire Ventures) (8:15)* - **Study these artifacts, not just the takes.** Cursor's [CI autofix template](http://cursor.com/marketplace/automations/ci-autofix) is the most copyable always-on GitHub agent setup from today [^7]. [fs-safe.io](https://fs-safe.io) is the cleaner reference if any part of your stack lets agents touch the filesystem through generated or user-supplied paths [^8]. *Editorial take: model choice still matters, but today's durable edge is harness design—portable interfaces, owned memory, trimmed context, and explicit review gates.* [^2][^3][^1][^4] --- ### Sources [^1]: [Pi.dev explained by its creators | AI Agents Under the Hood](https://www.youtube.com/watch?v=BQ_Es8k650I) [^2]: [𝕏 post by @masondrxy](https://x.com/masondrxy/status/2051714091924828480) [^3]: [The Rise of Agentic AI with Harrison Chase \(LangChain\) + Rajeev Dham \(Sapphire Ventures\)](https://www.youtube.com/watch?v=lgMTA5GU7TE) [^4]: [High Leverage - Ep. #9, The AI Coding Paradigm Shift with Simon Willison](https://www.youtube.com/watch?v=D76X-96tpTY) [^5]: [Redis ora ha gli array. Con una sorpresa.](https://www.youtube.com/watch?v=89GL7Gu9LAM) [^6]: [𝕏 post by @cursor_ai](https://x.com/cursor_ai/status/2051739625958584659) [^7]: [𝕏 post by @cursor_ai](https://x.com/cursor_ai/status/2051739627233628519) [^8]: [𝕏 post by @steipete](https://x.com/steipete/status/2051852940554481901) [^9]: [𝕏 post by @steipete](https://x.com/steipete/status/2051882417292525950) [^10]: [𝕏 post by @LangChain](https://x.com/LangChain/status/2051715028567437359)