Analysis8 min
The Model Is the Commodity. The Harness Is the Moat.
Frontier models are converging. The defensible layer has moved up the stack to the harness — and the real lock-in is session state, not model weights.
Back to articles
Engineering7 min
The Sandbox War: Cloudflare and Vercel Both Solved the Same Infrastructure Blind Spot
Cloudflare and Vercel shipped competing code sandboxes within 48 hours — one using containers, the other microVMs. The real story is what this reveals.
Guides7 min
File-Based Planning Is Becoming the Universal Agent Protocol
File-based planning is quietly becoming the universal protocol for agent reliability — and it has nothing to do with model quality.
Landscape7 min
AI-Generated Physical Design Is Having Its "ChatGPT Moment"
Four June 2026 launches — robot models, text-to-CAD, and AI architecture — mark the moment physical design AI graduated from research to usable tooling.
Case Studies8 min
The Agent Cockpit Wars: A Product Category Is Being Born in Real Time
Five products launched in two weeks — developers cannot see what their AI coding agents are doing. A new category is crystallizing in real time.
9 min
Smart Glasses Are the Unexpected Next Frontier in AI Dev Tools
Smart glasses that surface coding agent state in peripheral vision solve a problem no other dev tool has touched: the human round-trip delay.
Engineering8 min
Is Grep All You Need? The War Over How AI Coding Agents Search Your Code
A new paper says grep beats vector search for AI agents. The real story is that your agent harness matters more than your search algorithm.
7 min
Gemma 4 QAT + MTP Turned Local Inference Into a Cloud Competitor
Google's Gemma 4 QAT variants with MTP support in llama.cpp deliver 120 tokens per second on a $600 GPU — making local inference genuinely competitive with cloud APIs.
Engineering8 min