Tooling tidbits for AI-generated docs

My Claude doc-generating skills are in decent, workable shape, but I’m feeling inspired to do more after listening to a recent podcast from I’d Rather Be Writing – Orchestrating docs instead of writing them yourself — podcast with Sarah Deaton of Anthropic.

Just some quick bullet points of stuff I want to remember and maybe incorporate into my docs toolchains:

• Yes, it IS hard to force agents to adopt a voice in which the reader is actor! They just wanna use passive voice (even fable).
• For me, a new point of wordsmithing at the sentence level or clause level is to get to a point that the agent CANNOT misinterpret your writing (you want to leave no room for them to hallucinate an explanation).
• I’m gonna try incorporating persona briefs into my skills before drafting docs, and see if they do a better job of judging the ‘why’ of a feature.
• Try out Claude tag in Slack; interesting workflow option!
• An inspiring vision: ‘self-healing’ docs; auto bots that check drift btwn code and docs.
• I’m not sure if this was in the podcast or if I extrapolated, but: try adding ‘related topics/keywords’ to YAML frontmatter; nice to embed knowledge in docs metadata at the source rather than in skill.
• Mintlify docs platform now lets you see all the anoymous chats readers had w/ your serach box AI in docs; my research shows you can also achieve that with other 3rd-party AI chatbots.
• AI enambles new and exciting categories of metrics, for example: 100% accurate all the time; time to self-heal, time from feedback to implementation, maybe token costs?