I’m battling daily on giving up control yes or no. Last week I gave Claude’s agent SDK a try. The amount of filesystem tool calls made me crazy.

Here is another take on 12 factor but especially on agents github.com/humanlayer/1...

Nice. Never thought about it as full fledged PaaS replacement. Interesting thought experiment: What would Anthropic or OpenAI need todo to host the apps we’ve build on traditional PaaS?

The confidence is a function of the ignorance.

e.g.

$ youtube_transcript_api bTVavLFH7XQJMN9D9UVZn4 --format text | llm -s "These are the captions of a presentation at KubeCon 2025. Please summarize it."

Michael Krenshaw, a staff software engineer at Intuit and an Argo CD lead, shared the progress and future of Argo CD at KubeCon 2025.

wish our’s were on time.

you should try cross country train rides ;-)

We need a vendor independent SSO bringing context to all tools.

"Act as the function designer, the LLM does the work of building the body to my specification."
"Remember it’s a conversation. They’ll never complain at being told to refactor it!"
"LLMs are no replacement for human intuition and experience."

@simonwillison.net great post.

In your process, you explicitly put human in the loop. Do you think that is sufficient to gain intuition and experience? Or do we need to adopt some new practices? e.g. using LLMs to quiz us?