AI in 2026: What's Actually Changed and What Hasn't

The pace of AI development over the past three years has been genuinely difficult to process, even for those of us who follow it closely. But it's worth stepping back and asking an honest question: what has actually changed, and where are the promises still unfulfilled?

What Has Genuinely Changed

Code generation is now legitimately useful. Not perfect, not a replacement for engineers — but a serious productivity multiplier. Developers who've integrated AI pair programming into their workflows report completing tasks in a fraction of the usual time.

Multimodal understanding is here. The ability to reason across text, images, audio, and video simultaneously has unlocked applications that simply weren't possible two years ago.

Agents are doing real work. Simple, well-defined tasks — booking appointments, filing reports, monitoring systems — are increasingly being handled by autonomous AI systems.

What Hasn't Changed

Hallucinations remain a serious problem. AI systems still confabulate with alarming confidence. Any high-stakes workflow still requires human verification.

Common sense reasoning still breaks. Give an AI a genuinely novel situation and it often fails in ways that feel almost comically basic to a human observer.

The jobs apocalypse hasn't arrived. Certain roles have transformed dramatically; some have disappeared. But the wholesale displacement many predicted hasn't materialized.

The Honest Assessment

We are living through a genuine technological revolution. But revolutions move in fits and starts. The best posture is engaged pragmatism: learn the tools, apply them where they genuinely help, maintain appropriate skepticism, and resist both the hype and the backlash.