I am currently the cofoundeur and CEO of .txt, where we build the reliability layer for AI agents.
Structured streaming and agent primitives
I think the way we consume LLM output is fundamentally wrong. JSON’s closing brace is an unexamined synchronization point and the fundamental unit of LLM output is a fact, not an object. Once you see that, an LLM with patch streaming is a coroutine and agent frameworks are rational responses to an irrational constraint.
Developer tools for the AI agent era
It is becoming more and more obvious that we will need more tooling to fully take advantage of AI agents.Agent sessions make reasoning explicit in a way that changes how I think about software artifacts and code review.
Agent harnesses
The field has been focused on coding agents, and as a result has focused on an interative (TUI) form factor. Agents are also useful when they run in the background, reacting to their environment. We built our own agent harness at .txt to orchestrate the background agents we use for R&D, GTM, and generally gathering information.
Software design
Designing public APIs is one of my favorite things to do, and I still find time to enjoy it. I believe software_engineering_is_a_craft, and approach API design accordingly: start_with_designing_the_api_you_want and design by manipulating mock code until it feels right.