ARTIFICIAL INTELLIGENCE
I want to end this month with something I don’t write often enough: a note of considered, qualified, eyes-open optimism.
The AI landscape in early 2026 has the characteristics of a technology approaching the peak of its hype cycle. The valuations are extraordinary. The claims are sometimes extraordinary in a different sense. The gap between what the most enthusiastic advocates say AI can do and what it reliably does in production is real and significant. And there are genuine, serious problems — environmental, social, economic, regulatory — that haven’t been resolved.
I’ve spent a lot of words on this blog pointing at those problems. That’s because I think they’re real and important, and because uncritical boosterism does nobody any favours. But it’s worth, occasionally, stepping back to look at the bigger picture.
What’s Genuinely Impressive
The rate of capability improvement in AI over the last five years is, by any historical comparison, extraordinary. The things that are now routine — generating coherent long-form text, writing functional code from natural language descriptions, solving complex mathematical problems, generating photorealistic images from text prompts — were impossible or prohibitively expensive not long ago.
More importantly: these capabilities are being applied to problems that matter. Drug discovery timelines that used to span a decade are being compressed. Scientific literature that would take a researcher months to survey can be synthesised in hours. Educational access that required expensive human tutors can reach anyone with a smartphone.
“The trough of disillusionment that follows hype cycles is not a failure. It’s where the real, durable, unglamorous work happens.”
What the Trough Offers
When the hype cycle peaks and the trough of disillusionment arrives — and it will — AI won’t disappear. What will happen is that the serious, difficult work of integrating these capabilities into real systems, with real users, in real institutional contexts, will become more visible. The companies that survive will be the ones doing that work well, not the ones with the best demos.
That trough is where I think some of the most important AI progress will happen. Less glamorous. More useful. And ultimately more transformative than any single benchmark breakthrough.
The tools are remarkable. What we choose to build with them is still entirely up to us.
Tags: Artificial Intelligence • Opinion • Technology & Society • 192.168.1.22/