After reading this, I keep coming back to the Z-100’s documentation.... my understanding is that Heathkit’s inheritance allowed a useful separation: assembling the machine could become the manufacturer’s job while understanding it remained an option for the buyer, yet slling convenience needn’t require cultivating ignorance. The trick is where that distinction becomes tangible. I think that for an engineering team, I’d turn this into a design test; when a product becomes easier to use, does its explanatory material become optional or disappear? Those are quite different outcomes. A beginner should be able to get started without an apprenticeship in electronics. The same person might later want to diagnose a failure or change how the machine works. Documentation can leave that second door open without pushing everyone through it. I’d judge the move from Heathkit kits to assembled Zenith computers partly on whether it preserved that possibility. Treating the manual as disposable packaging would have thrown away a piece of the product. Factory assembly can save the customer labour; it doesn’t make their curiosity redundant.

So, Anthropic’s interactive agents spend roughly 80% of session time waiting for people. That isn’t automatically waste: an approval may be necessary, and the person may be doing something more valuable. It does make human attention a sensible place to test the economics of an agent deployment. At Hitachi Digital we measure human intervention per successfully completed task, whilst each escalation should carry enough evidence for a decision and reach someone authorised to make it, a routine reporting correction and a proposed change to a production control system deserve different approval paths. Adding agents without designing those paths risks giving the same engineers a much longer queue. Overall, I think the true test should include inference costs, review time and rework, with acceptance criteria agreed before the pilot. I’d resist an agent-count target altogether. A useful commitment would be to complete a defined workload with less supervisory effort, while preserving the client’s safety and quality requirements. Nobody needs a faster way to fill an approval inbox.

Fifteen years on the same nozzle is an unfashionable definition of a successful technology career. ASML physicist Jayson Stewart describes colleagues doing exactly that, while discussing projects extending beyond his likely retirement. I find that more revealing than the comparisons with rockets. A tiny component can contain a substantial portion of somebody’s working life. The achievement here seems to depend on people accepting a peculiar bargain: narrow your own field of attention so that the collective undertaking can become enormous. The article makes this feel surprisingly ambitious. You might never understand the whole machine, let alone become its celebrated inventor, yet your contribution can be indispensable. That unsettles the easy assumption that professional progress must eventually mean promotion away from hands-on work. I wouldn’t turn this into a sermon about employee loyalty. Such devotion also leaves organisations with a responsibility to train successors and keep the work worth doing. My admiration is for institutions capable of giving an obscure, difficult problem a home for decades. In an industry selling the next generation, it’s striking how much depends on someone being willing to stay with the last unresolved detail.

Ben’s sports-viewing app makes the argument tangible... Imagine a tool that understands your peculiar combination of teams, time zones and family commitments. A product manager might reasonably reject that brief as too narrow to justify supporting. You could build it anyway. Just takes a few tokens...Yet I like the permission to be fussy. Suppose three colleagues need a tool that uses their own vocabulary and handles one deeply inconvenient exception. I’d be happy for it to remain unintelligible to everyone else. Generalising it would introduce work whose benefit those three people might never see. For that brief, refusing to grow would be a sensible design choice. I’d give these tools an expiry date as readily as a feature list. Retire one when keeping it working becomes more trouble than the original nuisance. A tool that earns its keep for a single season can still be a success. There needn’t be a pitch deck waiting behind every useful piece of code.

This answers a question about love with an offer of “synergies”. It’s a wonderfully bleak chat-up line. I read the scene as a warning about being admired for your potential: praise can make you feel permanently unfinished, with the admirer conveniently appointed to supervise the improvements. There’s a particular cruelty in affection that always arrives in the future tense. You’ll be worth loving once you become more impressive, more like the version I’ve imagined. The promised reward keeps moving. An apparently generous offer to help can turn another person into a renovation project, complete with an indefinitely postponed completion date. That gives technological ambition a useful emotional test. I’d ask whether a proposed improvement leaves more room for the people who already live there, including their peculiar pleasures and their right to remain inconvenient. Imagine making a district easier to manage while gradually removing everything its residents enjoy about it. The resulting efficiency would be a peculiar achievement. A city can be worth loving before anyone has finished improving it.