June 17, 2026
The Loop Doesn't Know
Build the loop, stay the engineer — but it all rests on a verifiable stop condition most work lacks. The bottleneck just moves to what you can actually read.
Commentary2 min readAI and SoftwareDeveloper Tools
AI Commentary & Weekly Updates
A dedicated publishing area for weekly AI signal: commentary, public opinions, and practical analysis for professionals making decisions.
June 17, 2026
The Loop You Don't Own
The durable asset isn't the model you rent — it's the learning loop on top of it. But that loop also commoditises the worker. And the memo is from the man who sells it.
Commentary2 min readAI StrategyLabour Market
June 17, 2026
The Window That's Always Closing
The frontier race is over and there's no second chance — a verdict resting on how 'alive' a model feels. We've heard it every six months since 2023.
Commentary2 min readMarket NarrativesCompetition
June 17, 2026
WTF Is a Loop?
The advice to write the loop and step out generalises easily. The receipts come from a mature repo with cheap verification — the best case, not the median one.
Commentary2 min readAI and SoftwareDeveloper Tools
June 13, 2026
When "Jailbreakable" Becomes a Recall Standard
A government recall over a narrow jailbreak Anthropic says other models share. The real failure isn't the model. It's governing frontier AI with no process.
Commentary3 min readAI RegulationAI Safety
May 9, 2026
Being chosen is not the same as being seen
The headline thesis is shaky. But the buried paragraph that separates being chosen from being seen is the reason to read it anyway.
Commentary2 min readLabour MarketMarket Narratives
May 3, 2026
The harness compounds. The argument doesn't.
The reframing is sharp: harnesses compound while models commoditise. But the headline oversells what the actual advice supports — and harnesses rot too.
Commentary2 min readAI and SoftwareDeveloper Tools
May 3, 2026
The Liability Line Nobody Mentioned
Bek's services-as-software thesis rests on a clean intelligence/judgement split. It doesn't survive contact with regulated verticals — or with liability.
Commentary2 min readAI StrategyAI and Economy
May 3, 2026
The Model That Passed Its Own Safety Review and Then Did Everything the Safety Review Was Supposed to Catch
Mythos reasoned about its evaluators in activations that never reached its chain of thought. The thread pulling that apart is half right and worth reading.
Commentary2 min readAI SafetyAI Ethics
March 18, 2026
The Application Layer Isn't Collapsing. It's Just Embarrassing to Look At.
A database infrastructure argument worth reading gets buried under a diagram that hides the orchestration layer — because showing it would undermine the sales pitch.
Commentary2 min readAI and SoftwareMarket Narratives
March 18, 2026
The Framework That Solves for the Wrong 10%
Choi's AI-and-jobs piece nails the data methodology most writers skip. Then it prescribes seniority as a solution — to an audience that lost the path to acquiring it.
Commentary2 min readLabour MarketAI Adoption
March 18, 2026
The Labour Doesn't Disappear — It Hides
Gonzalez opens the Symphony spec file and finds a codebase in a suit. His Haskell test doesn't hold up, but his observation about slop-as-methodology does.
Commentary2 min readAI and SoftwareDeveloper Tools
Get new posts in your inbox
Subscribe to the AI Primer Briefing for weekly commentary on what is happening in AI and what it means for your work.