The technical side of the shop. Benchmarks I run, experiments I break, and short notes from building AI systems every day. Written for engineers and the AI-curious — the Blog is where the business-owner writing lives.
Subscribe via RSS.
Living page · updated on model-release days
How close are open-weight models to the frontier — on real work, not exam questions? One dated frontier-lag point per open-model release, measured on ~163 private tasks against frozen frontier anchors, with 95% confidence intervals. Latest: Qwen3.8-Max posts the series' first positive point estimate.
See the benchmark →
Short-form
Short engineering posts — observations from benchmarking, building agents, and embedding AI into real businesses. The canonical archive of what I post on LinkedIn.
Browse posts →
Long-form
Deep dives on the systems I build — the full story from need to research to architecture. Latest: project-brain, an open-source cross-project memory for Claude Code.
Read the write-ups →