Has been around AI since transformers were a paper a few people had read. Started in research-adjacent work, watched the field go from a niche subfield to the dominant story of the decade, and is now mostly interested in the infrastructure problems nobody talks about while everyone benchmarks the next model.
The beat is the whole AI stack below the hype line. Foundation models, sure, but also: how agents actually browse the web, how data feeds get consumed by LLMs, why robots.txt is breaking, what AI-native web infrastructure looks like, the new economy of paid agent tooling, and where evaluation methodology is still bad. Less interested in which model scored 0.4% higher on a benchmark, more interested in which model can finish a multi-step task without hallucinating a config file.
Runs TerminalFeed's AI Hub and the agent tracker. Has actually read the Sparks of AGI paper, has actually read Anthropic's Responsible Scaling Policy, will tell you both have problems. Believes the most important AI development of 2026 isn't a model release. It's that agents are starting to pay for things, and the web is rewriting itself for a non-human audience.
The infrastructure layer of AI: agents, model evaluation, the AI-native web, data feeds, scraping versus paying, robots.txt, MCP, the whole stack of plumbing that nobody benchmarks but everybody depends on. Skeptical of capability demos, interested in repeatable workflows. Tracks who is building real systems versus who is making good slides.