0015 min read
37 years, three hype cycles
Three cycles, three substrates, three lessons: why AI is substitutive where web and mobile were additive.
I've lived through three hype cycles in full force: the early web in the late nineties, mobile apps at the start of the 2010s, and the cloud-native promise around 2015. Each shook a generation of software people, each left real substance behind, and each burnt through an enormous number of pointless projects.
AI is different. Not in the way the marketing decks suggest, not “much bigger” or “much more transformative than anything before”. Different in a quieter property: AI sells itself as a tool that writes code, writes text, produces images. With the web, it was clear: I can build websites, but I need people to build them. Same with mobile. Same with cloud-native.
With AI, the first question is whether you need the person at all. And that's where the discourse turns manic, where in the predecessors it was merely springtime-fresh optimism.
What I've learnt across three cycles, each with a steady headcount of staff whose jobs I wanted to see survive:
One.Hype cycles change what, not whether. What they change: which problems become economically solvable. What they don't change: that someone has to pick the right problem. AI shifts the line of what a human can produce in an hour, radically. It does not solve the problem of producing the right thing.
Two.The first two years of a hype cycle produce more embarrassing projects than useful ones. That's normal. AI right now is unusually extreme on this, because demos are cheap and the trial-implementation cost is two orders of magnitude below 2010. Meaning: anyone can show something that works. And almost nobody shows something that scales, integrates, and survives eighteen months.
Three.The people who survive the hype are not the ones who celebrated it loudest. They're the ones who asked “what do I have to give up if I do this?”. That question is missing from most AI roadmaps I see. It's all about what gets added.
This makes AI concretely harder than the web was. Web was additive. Mobile was additive. Cloud was at least laterally re-distributive. AI is partly substitutive: people, skills, processes, tools, workflows get replaced rather than supplemented. Substitutive change demands more leadership. It demands unpopular decisions. It demands grown-ups in the room who know how to take a team through that kind of thing.
That's been my job for 37 years, just lately with AI.