This week I went deep into the emerging AI design discipline: AI Agent Experience (AX).
Intro for Product & UX Design available on @Medium.
Deeper dive on flocker.md blogs.
https://t.co/qOIBLgBA64
Honestly thought this would have been corrected in Claude Design by now.
Anyone else noticed Claude 4.6+ loves TINY font?
Couldn't see any articles on 11px text being good design
An artefact of training on other languages? @AnthropicAI@ClaudeDevs@ncameron@RamenClubHQ
@kcosr@ncameron@charlierward@bcherny@CopilotKit@ulidabess Agreed about the changing product surface. Definitely agree on keeping credentials out of the context. However MCP already achieves this with OAuth. My reservation would be if Anthropic is using the tunnel as a method of tracking, or if the auth tokens aren't stored locally.
How are people feeling about Anthropics move to centralise your MCP authentication on their own servers?
My opinion - completing their OAuth client implementation and bringing it up to specification seems like it would be the better first step...
https://t.co/SP2Fhf34PM
For those building with AI agents - genuinely impressed by the OSS dev signals from https://t.co/OpNQjHcJ8r.
You can secure a future .agent domain by backing their TLD ICAAN application - click below.
https://t.co/ULmwJrrCsi
@agentcommunity_@RamenClubHQ
Open source DNS backed agent discovery protocols.
Hmm. Interesting dilemma. As a founding engineer:
1) Do you have the time and resources to maintain the software?
Among other things; modern web packages (npm) are not stable, security and bugs will creep in.
2) Do you have the time and resources to provide product support?
SaaS = service = support
3) If your climbing gym's operation, and your income, hinges on this "trivial to maintain" SaaS, do you trust AI providers to keep their pricing at current levels?
Open source seems like your most viable option if youre not interested in the IP
@bhalligan Well delivered @bhalligan!
Particularly relevant right now for those creating STDIO MCPs pushed to npm. More powerful, but less IP safe without caution.
@teddy_riker Nice! A validation loop would be fantastic. I've got some ideas in mind - but AI evals feel like a level of black magic I'm not ready for.
@kcosr Love to see it. As the major players find their balance on demand vs. capability, orchestration is genuinely the only way (plus everything else awesome it unlocks). Nice work.
@flocker_md uses similar patterns, look forward to seeing more @kcosr !