New page idea: Agentic AI

[henryiii]

I think we should provide a new page on Agentic AI. Like the existing page, it would start by drawing from my experience with it:

• Starting With Agentic AI
• Claude Code Reviews with Fable
• Setting up Agentic AI (URL will change)
• My skills - including sp-recommends, maybe could be upstreamed here.

I think it should remain mostly tool/model agnostic, and have something like the following:

• Guidelines for an AI_POLICY.md (and maybe a few suggested examples for different levels of AI support)
• Guidelines on how to provide an AGENTS.md (and maybe a mention that one harness doesn't support it (Claude Code), and how to fix it (symlink CLAUDE.md to AGENTS.md).
• Best practices for AI usage, with user-level AGENTS.md suggestions. My strong opinion is full disclosure and transparency:
  • Linux kernel trailers
  • AI text disclaimers for generated text
  • Human interaction to human review
• Very brief overview of the skills standard (maybe a few links like skills.sh
• A suggested list of things that AI is good at, similar to the setting up page above, maybe with sample prompts
• Warnings for things AI is bad at
• Maybe a brief intro to common harness features
• A list of some common concerns (unless already covered organically above)

Maybe would also be good to help clarify the difference between low-cost models running in automated systems like OpenClaw (which tend to be the thing most associated with "AI slop") vs. a developer using an AI harness and a quality model.

We might be able to use tabs to have "all in", "moderate", "minimal", such as for the AI_POLICY.md and maybe adjusting suggestions (like adding a AGENTS.md vs. git ignoring vs. not mentioning at all, and letting users add it to .git/info/exclude).

Open to thoughts!

[drammock]

• A suggested list of things that AI is good at, similar to the setting up page above, maybe with sample prompts

• Warnings for things AI is bad at

☝🏻 These two risk getting stale much faster than the other content (the latter point moreso than the former). Otherwise these seem like good/useful content ideas