New iOS feels like a junior designer discovered the gradient tool, and are now using it EVERYWHERE. I've been there, that was me once.

If you are developing tooling for agents (whether via MCP or function call), ALWAYS keep in mind that the end user is the AI The interface must be designed in a way that it's easy for the AI to understand Often this means you need to go beyond just exposing the underlying API

Compound AI systems is a much better technical term than AI agents

Now that dynamically generated components / web apps are here, is now finally the time for iframes to shine?

Whenever I have a long flight 😅

Visited the xAI office today, they hired so many researchers away from OpenAI they had to expand the office with tents. Truly humbled.

Introducing Agentic: RapidAPI for LLM Tools 🔥 Turn any MCP server or OpenAPI service into a paid MCP product. • Built-in support for every LLM SDK and MCP client • OAuth, Stripe billing, rate-limits, etc • Extremely configurable • Fully OSS w/ over 17k stars on GitHub! 🧵

It gets worse. I recently met a EU tech startup whose founders refuse to let anyone on their 20 ppl team use any AI tools - no chatgpt, no cursor - because they don't want to use too much water & energy... ngmi

This reminds me of a great quote by mark pincus - e/acc: all new fails Way too many agent companies are overshooting the product, when they should be tackling problems in piecemeal. It doesn't matter that LLMs can actually do more. People only adopt in piecemeal, all new fails.

There are a lot of apps where the experience would actually be better if it's just done over 100% voice (e.g. getting a house cleaner), as long as who I'm talking to is knowledgeable. Voice beats bad UI any day. And now it's economical. Is 1-800 voice apps making a comeback?

Is there a slack bot that automatically monitors slack and turn my convos into Linear tickets? I know there are some agent builders that does this, but I'm just looking for something I can drop in with no setup.

All the voice agents I've seen model the conversation flow after a state machine, which seems overly limiting. An agenda based approach is probably better. The analogy is imperative vs declarative programming - instead of mapping out every convo path, focus on what must be

Well I think the probability that we all live in a simulation just raised by an order of magnitude