If you haven't properly looked since 2023, brace yourself — the jump is violent.
- Models that flunked coding tasks in 2023 now do useful junior-level work on plenty of stacks.
- Image and voice quality crossed from toy to production-grade for a lot of jobs.
- Agent demos that were lab toys hit real business pilots within months, not years.
- Robots went from viral clips to actual factory shifts (that's Lesson 5, with names and numbers).
From where I'm sitting today, the frontier models even feel more balanced than the worst of the over-censored phase — less trigger-happy on refusals, more even-handed on the surface, and hallucinating noticeably less than the early GPT days. Not perfect, mind. Still gets things wrong. Still needs checking. But genuinely usable in a way that early 2023 often just wasn't.
And the release cadence is frankly absurd. New frontier models land almost weekly — each cycle a bit faster, sharper, more capable. The CEOs frame 2025–2026 as a turning point in how work itself gets done. Take that as a statement of intent and spend, not prophecy — but note they're putting tens to hundreds of billions behind it because they expect a return.
So "wait and see" is a perfectly valid strategy — as long as you go in clear-eyed about the cost. Waiting can mean your employer adopts AI without you in the room when the decisions get made. It can mean a competitor automates your repetitive slice before you do. The gap quietly widens between people who direct AI and people who don't.
The answer isn't to panic-buy a £500 course off some influencer. It's to follow a plain path — and, eventually, build your own stack, instead of living inside whatever morality filter Silicon Valley settled on this quarter.
Continue — an optional hands-on.