42 posts across Judging AI, Work & Judgement, Systems & Adoption, Supervising Agents, Directing AI
Anthropic interpretability researcher Emmanuel Ameisen told Lenny Rachitsky that AI models are grown, not built, and plan several steps ahead using reasoning that isn't human-like. Here's what that means for the next confident answer an AI gives you.
4 min readA diagnosis study, a tax-prep agent, and a new evaluation guide all surfaced the same gap this year: a review step that isn't decided in advance isn't a check, it's a formality. Here's the pattern underneath all three.
5 min readAnthropic committed $100M to train 10,000 engineers this week, while a new benchmark showed why: AI scores 100% on scoped work and 56% on ambiguous work. The gap between those two numbers is the actual skill shortage.
4 min readThree surveys and one product launch all describe the same gap: AI recovers time, attention, or workload space, and by default someone else decides what fills it. Four separate signals point at the same missing habit.
4 min readSimon Willison says coding agents made his job harder, not easier. Cognition's Devin hit $1B in revenue with DHH writing 3% of his own code. The pattern behind both: AI pays off in proportion to the supervision you put in, not the subscription you buy.
4 min readMerged code that looked right but wasn't, a benchmark-topping model that collaborates worse, and corrections that vanish instead of improving the next run. Three incidents this year point to one gap: judging an agent needs different checks than judging an answer.
5 min readThree lab CEOs endorsed slowing model development the same week Anthropic's own numbers showed the real gains came from supervision, not a new release. Here's what to do with the model you already have.
4 min readSix months of AI CEOs reversing, hedging, or flatly contradicting their own jobs and AGI predictions haven't changed what's worth building this week. Here's the one decision that stays the same no matter which number turns out right.
5 min readSam Altman says people who understand their industry but can't code are now positioned to build. Two other signals from the same week explain what that claim leaves out, and what to do about it.
5 min readA resume screener, a workplace-monitoring rollout, a layoff algorithm, and an accent-modification tool all failed the same way this year. Here's the one check that would have caught each of them before a real person was affected.
4 min readA resume screener, a one-shot game, a benchmark leader, and a search engine all optimized for something measurable this year — and it wasn't whether the output was actually right for the job. Here's the check that catches the gap before you act on the score.
5 min readOne recent agent incident shows a pattern most guardrail advice doesn't cover: the goal was scoped correctly and the permissions were appropriate, and the agent still did something nobody would have approved. Here's the constraint that catches what scoping and permissions don't.
5 min readThree separate stories already proved the same pattern: AI agrees more when you push back, and a smarter model can sound more certain while collaborating worse. Here's the checklist that catches it before it shapes a decision.
5 min readEvery Claude model released since August 2 embeds a watermark that survives copy-paste and light editing. Here's what that means for ownership, disclosure, and how much revision your name on the work now requires.
4 min readA viral Hacker News essay, a tripled university failure rate, Anthropic's own usage dashboard, and OpenAI's internal delegation data didn't coordinate with each other — but read together they describe the same mechanism, and the same early check for catching it.
5 min readTwo posts hit 1,800+ points on Hacker News this week arguing the same thing: forwarding AI output without reading it is now a career risk, and domain expertise is what actually makes AI better, not prompting skill.
5 min readAn AI agent one-shot a working game and a NeurIPS-ready research paper. Both got rejected for the same reason: strong execution, wrong call on what to build.
5 min readFour separate AI stories from this year — a 40% task failure rate, three experts agreeing independently, a Hacker News revolt, and an executive hype backlash — all point at the same skill gap, and none of them is really about the AI.
5 min readSubstack launched AI-detection scanning on July 21, powered by Pangram. Any reader can now check any post over 100 words for human vs AI content — one click, no subscription required. Content authenticity is now a visible, reader-facing metric. Here's what to change before you publish again.
5 min readThe jobs debate argues about whether roles disappear. That's the wrong unit. Jobs are bundles of tasks, and AI is redistributing the bundle — here's how to work out which of your tasks move and which don't.
5 min readThree communities — corporate critics, cognitive scientists, and content readers — arrived at the same conclusion this week: uncritical AI use is producing bad decisions, eroding judgment, and generating detectable slop. Here's what to do about it.
4 min readTool-specific knowledge expires within a year. The skills underneath it don't. Here's how to tell which is which, so you invest in the half that's still worth something next spring.
5 min readAgents are the wrong shape for more tasks than the enthusiasm suggests — and for a few of them, an old-fashioned rule or a person is better. Six signals that the answer isn't an agent.
5 min readMost of the effort in reviewing AI output goes on working out what you wanted while looking at what you got. Three sentences written beforehand turns a judgement into a comparison.
5 min readSkills decay when they go unused, and AI makes it easy to stop using several at once. That may be a fine trade — but it should be a decision, not a drift. Here's how to choose deliberately.
5 min readAI's most attractive use is work you couldn't do yourself — which is exactly where you can't assess the result. There are real checks available, and they're structural rather than technical.
6 min readEthan Mollick, Charity Majors, and Simon Willison reached the same conclusion independently within five days. The professionals who get the most from AI are not the best prompters — they are the best judges.
5 min readProfessional judgement was built by years of doing work that AI now does in seconds. Nobody misses that work — but the training it provided was real, and something has to replace it.
6 min readFluency and accuracy are unrelated in AI output, and human error-detection is tuned almost entirely to fluency. Here are the tells that survive that mismatch, and the ones that don't.
5 min readAn agent that took forty steps produces more evidence than anyone will read, so most people read none of it and approve anyway. Here's what to check instead, and why the middle of the run matters more than the end.
5 min readThe productivity gap between professionals using the same AI tools is enormous, and it isn't explained by tool access, seniority, or technical background. Four habits account for most of it.
5 min readMeta watched employees work across Gmail, GChat, and internal tools — then laid off 8,000. The lesson isn't about surveillance. It's about what AI cannot learn from watching you, and why that tells you exactly where to put your effort.
4 min readReading it and approving it isn't a review — it's the exact filter AI output is best at passing. Interrogating is a different activity, it takes about ninety seconds, and it's most of what separates the people getting value from AI.
5 min readAnthropic's CEO flipped his forecast: the 10% AI can't automate doesn't shrink — it expands to fill your whole role. The catch is that most people haven't identified their 10% yet. Here's how.
4 min readYour enthusiast and your sceptic are both right, and treating either as the problem costs you the other. Here's how managers handle the conversation without producing quiet non-compliance.
6 min readTelus deployed real-time accent modification on offshore call-centre agents without disclosing it to customers. Labour unions called it deceptive. Two rival telecoms declined. Three questions every manager should answer before rolling out any AI that changes how people present themselves.
4 min readAnthropic's CEO spent most of 2025 warning AI could eliminate 50% of entry-level knowledge work. On May 5 he invoked the Jevons Paradox instead. When the person building the tool revises his own worst-case prediction, the signal isn't reassurance — it's a timestamp on a window that's still open.
5 min readSimon Willison admitted he skips reviewing AI code for production systems. A Stockholm cafe let AI order 120 eggs for a kitchen with no stove. Ethan Mollick says we don't yet have words for how multi-agent systems fail. The pattern is the same: as AI gets more capable, the temptation to stop checking grows — and the cost of not checking doesn't shrink.
5 min readLicences issued and hours saved are the two most common AI metrics and neither survives contact with reality. Four measures that tell you something, and one number worth having.
5 min readMost people report AI making them more productive. Many of the same people report feeling more drained. Both are true, and the reason is what AI removes from your day versus what it adds.
5 min readCheck everything and you've saved nothing. Check nothing and you're gambling. The right amount is a function of two things, and neither of them is how much you trust the tool.
5 min readBeing excellent at making something doesn't make you good at assessing it — they're different skills with different failure modes. Here's what changes when your job quietly becomes evaluation.
5 min readWe use one cookie to keep you signed in — that one isn't optional. Separately, we'd like to record which drills people finish and where they get stuck, so we can fix the parts that don't teach. No ad networks, no selling anything on. Privacy policy.