At some point we’ll see this as an accessibility issue for (non-human) workers with specific interface requirements.

I think the first wave of “products” will certainly be “full stack agents” but pretty soon there will be “agent tools / actions” as product. I’m excited about the second wave.

I’m trying to understand if there’s a place for agent tool “products” or if the agent is the product. Asking for an enterprise friend 😉

URL changed when I moved to Hugo, chrismcavoy.xyz/posts/250130...

Well...after less than 12 hours or so I've moved to Hugo. While Python is still almost always the answer, the themes and features of Hugo are pretty far ahead of Pelican. I'm sure I'll change this up at least a dozen more times.

Also super interested to see what happens if one of the tools allows the LLM to write more tools and add them to their toolset. Probably dangerous but it would be interesting.

I’ve wondered if there’s a potential pattern of another LLM observing tool usage and adding “learning” to the context on future calls, then culling and editing context (memory) as feedback improves.