A dedicated publishing area for weekly AI signal: commentary, public opinions, and practical analysis for professionals making decisions.
March 18, 2026
Shapiro's post-labour economics thread introduces a capital ownership model worth taking seriously — then undermines its own conclusion in a single clause.
March 18, 2026
Linton's agent walkthrough nails the tool-call misconception most explainers dodge, then conflates a while loop with agency and calls it done.
March 18, 2026
Yang asks the one diagnostic question about white-collar work that actually matters. Then he extrapolates one CEO conversation into a global forecast.
March 18, 2026
Willison's coding agent explainer nails the statelessness point most writers skip — then names the hard problem and walks away from it in the same sentence.
March 18, 2026
Moonka's AI stack breakdown has the best infrastructure numbers you'll find this week. It also has a one-paragraph gap exactly where DeepSeek should be.
March 15, 2026
Tane's 'slop creep' piece names the real failure mode: plausible bad output arriving faster than any review process was designed to catch. The scope claim doesn't follow.
March 15, 2026
Anthropic's 94/33 gap — theoretical AI exposure versus real-world deployment — reframes the timeline question. The rest of the piece doesn't hold up as well.
March 15, 2026
A ten-principle agent framework gets the architecture right and the priorities wrong. Two principles earn their keep. The rest need a deployment context.
March 15, 2026
The AI apps you use daily are the worst investments in the stack. The real margin sits three layers down. One piece names this directly — then overreaches with a historical analogy.
March 15, 2026
Tsinghua's OpenMAIC correctly diagnoses passive AI education tools — then offers a more elaborate chatbot dressed in classroom language. The hard question never appears.
March 15, 2026
An agent that fails silently is more dangerous than one that fails loudly. A circulating infrastructure post buries its best insight under a generic headline.
March 15, 2026
Thompson's NYT piece names the greenfield/brownfield split that explains why AI productivity claims contradict each other. It also buries its most alarming finding.
Subscribe to the AI Primer Briefing for weekly commentary on what is happening in AI and what it means for your work.