doctrine.engineering
doctrine.engineering
Stand back kids, I'm doing a magic trick
2026-07-23<br>article
A Motorcycle for the Mind
Steve Jobs famously called the computer a “bicycle for the mind” — I<br>guess that’d make LLMs a “motorcycle for the mind”. They’re<br>certainly less effort than pedalling yourself.
I think one of the things that really separates us from the high<br>primates is that we’re tool builders. I read a study that<br>measured the efficiency of locomotion for various species on the<br>planet. The condor used the least energy to move a kilometer.<br>And, humans came in with a rather unimpressive showing, about a<br>third of the way down the list. It was not too proud a showing<br>for the crown of creation. So, that didn’t look so good. But,<br>then somebody at Scientific American had the insight to test the<br>efficiency of locomotion for a man on a bicycle. And, a man on a<br>bicycle, a human on a bicycle, blew the condor away, completely<br>off the top of the charts.
And that’s what a computer is to me. What a computer is to me is<br>it’s the most remarkable tool that we’ve ever come up with, and<br>it’s the equivalent of a bicycle for our minds.” ~ Steve Jobs
Both LLMs and motorbikes can be used with relative safety. You can tell<br>just by looking at a motorbike that safety is something to be concerned with.<br>It’s also painfully obvious as soon as you sit on one that operator ability is<br>crucial. And when you ride a motorbike, your body will tell you very<br>clearly (if you’re lucky, via your adrenal glands) if you’ve just done<br>something stupid. LLMs will not. They’ll clap. In a way, that’s<br>why they’re dangerous.
Not in the same way as riding a motorbike. Yes, there certainly are loud, sudden, and<br>explicit failure modes, but I think it’s under-appreciated that there are others.<br>Many of the failures won’t be recognised as failures at all. Some of them will<br>get people promoted — despite never really creating value. Increasingly, it can<br>be hard to tell the difference between value creation, and the plausible-seeming<br>appearance of it.
This is not to say that LLMs don’t have compelling uses. It’s very clear<br>that they do, at least in software engineering. But there’s a popular conceit<br>that as agents become more capable, it matters less and less who’s actually<br>directing them. You’ll save so much money when you fire all the smart people,<br>you’ll be able to afford all the tokens you could ever need! Clap emoji.
I think we’re going to see a rise in commercial fatality rates: companies that<br>cease to exist in a year, or two, or three, because of what the decisions they’re<br>making right now. Eaten, of course, by other companies making smarter decisions. In<br>fact, I’d suggest that it’s almost inevitable, regardless of how optimistic<br>you are about the upside of AI, as long as you concede it’s a dominant theme in<br>strategy and expenditure: over-investing and under-investing are both potential paths<br>to the corporate darwin awards.
Moreover, there are a nearly infinite number of ways to go about anything<br>incompetently. You can be dead right about the potential of something, yet catastrophically<br>fail to realize it. Rapid changes in the fitness landscape beget waves of<br>extinction and speciation.
It’s wild just how wide the gap in effectiveness is<br>between those (few) who treat agentic engineering as a real technical<br>discipline (deserving of rigour, hypothesis-driven experimentation,<br>investment in tooling, etc) and literally everyone else.
The landscape is thronging with snake-oil salesmen and hacks deep in the throes of AI psychosis,<br>so most developers either assume what they're already doing is the only way there is,<br>or outright reject any suggestion that better ways exist as superstition for the gullible.
Well, better ways do exist. But if rabid enthusiasm is not a<br>reliable indicator of credibility, what is?
The same things that always have been. Strong engineering fundamentals.<br>Having made something worth showing (other than a training course pyramid scheme).<br>The people who know what they're doing all tend to converge around the<br>same general approaches (because they work), and don't treat them as if they're<br>trade secrets for premium subcribers only. When they do introduce something novel,<br>they can give a balanced accounting of the rationale and tradeoffs.<br>And they're not claiming that one weird trick can replace being a very smart<br>person to begin with, along with skills and expert intuition earned through<br>many years of professional experience.
AI is fundamentally a very big lever — a force multiplier. And<br>the expertise, judgement, and vision of the people wielding it is<br>not obsolete. If you start with stupid and add AI, you<br>get a whole lot of stupid. If you start with Don Knuth,<br>you advance the frontiers of human knowledge.
Natural intelligence matters, along with all the other...