POV: @andrelandgraf's talk on agent onboarding at @neondatabase: zero errors, zero doc lookups π₯
Thanks @MuxHQ for hosting + @andrelandgraf for the @2027dev shoutout!
What does this type of Agent Experience (AX) mean?
If you've been around a while and your docs are part of the training data, agents can start using your tool from memory, no doc lookup required
This is a double-edged sword: LLMs know you well, which delivers great AX today, but it flips fast when your product or docs change
Some ideas on how to manage it:
1/ Iterate on docs vs. removing; make changes additively
2/ Be machine-readable: MCP + llms.txt + context7 so agents can fetch current truth
3/ Typed SDKs: ideally, deprecations show up as type errors agents can self-correct on
4/ Self-explaining errors ("410 - use /v2" instead of a generic 404)
5/ Versioned urls with long deprecation runways
The tools that win the agent era will be both baked into the training data and machine-readable at runtime - so agents can learn what's new
What else would you add?
We've been cooking @2027dev π₯
Humans aren't your users anymore - AI agents are. Our goal is to ensure you know where they get stuck using your tool
Diving into our April update π§΅