Posts
All the articles I've posted.
-
AI Agents Don't Need More Hype. They Need Better Systems
The most important thing about AI agents is not the model. It is the surrounding system: decomposition, context control, evaluators, permissions, and the boring harness that keeps a clever model from acting like an expensive intern with root access.
-
AI Daily Digest - 2026-04-16
Anthropic shipped Claude Opus 4.7 with a sharper pitch around long-horizon work, self-verification, and upgraded vision—useful if it holds up outside benchmark theater. Elsewhere, the agent stack keeps getting more real and more fragile: Cloudflare is turning AI Gateway into a cross-provider inference layer, Google is formalizing agentic Android workflows, and new research shows just how exposed teams are to poisoned tools and shady LLM routers.
-
AI Daily Digest - 2026-04-15
Today’s AI story is less about flashy demos and more about leverage: Anthropic is automating pieces of alignment research, OpenAI is reshuffling compute commitments under Microsoft’s umbrella, and ASML says AI demand is still pulling the semiconductor supply chain forward. The other thread is security—models are getting better at both offense and defense, and the tooling around agents is maturing fast, whether that means browser sandboxes, observability stacks, or red-team benchmarks.
-
AI Daily Digest - 2026-04-14
OpenAI widened access to cyber-capable models while tightening the gates around them, a sign that defensive AI is becoming a managed product category rather than a research side project. Elsewhere, Ukraine’s robot-only assault, Google’s Chrome “Skills,” and Cloudflare’s MCP and OAuth moves all point to the same shift: AI is moving deeper into operational infrastructure, with the usual mix of utility, risk, and a lot of vendor promises that deserve scrutiny.