What Remains

At the end of each year I try to look back and ask what changed. This year, the more interesting question seems to be the opposite: what didn’t?

A great deal changed. The year began with DeepSeek shaking the industry’s certainties, and it’s ending with a new wave of models from several labs, each one able to do things that would have seemed impossible not long ago. Agents went from interesting demos to tools that work for hours on their own. Many of us, myself included, now spend our days directing machines more than typing. The pace never let up.

And yet, when I look at what actually mattered this year - in my own work, and in what I saw around me - I keep finding the same old things underneath.

What remained

Judgment remained. The machines got much better at producing answers. Knowing which answer was right, which question was worth asking, and when to stop - that stayed with people. If anything, it became the whole game.

Responsibility remained. When the agent deleted the database, a person had the bad day. When an AI feature failed, a person answered to the customer. The machines act. People are accountable. That didn’t change, and I don’t think it will.

Honesty remained. The most useful thing I read all year was a study showing that experienced developers believed they were faster when they were actually slower. The lesson wasn’t about AI. It was about how easily we deceive ourselves, and how much it matters to check.

People remained. The 95% of AI projects that struggled mostly struggled for human reasons. The ones that succeeded mostly succeeded for human reasons too: someone who cared, understood the problem, and had the patience to change how work was done.

The young remained our responsibility. The canaries are still in the mine. Whether the next generation of experienced people exists depends on whether this generation makes room for them.

What I learned about myself

I learned that I can be excited and skeptical at the same time, and that this isn’t a contradiction but a kind of balance.

I learned that I have to protect time for slow thinking, because the tools are so fast that they can quietly crowd it out.

I learned, again, that the parts of my work I value most - understanding people, making things simple, helping others grow - are exactly the parts no tool does for me. I find that reassuring.

What I’m carrying into 2026

A few intentions, held lightly:

To keep using these remarkable tools every day, with gratitude and with open eyes.

To keep asking what’s worth building before asking how fast I can build it.

To spend more time with people who are just starting out.

To write more, in public, because the knowledge we share is what everyone - people and machines alike - learns from.

To rest more, and trust that rest is part of the work.

A closing thought

There’s a phrase from Kohelet that I come back to often: one generation goes, and another comes, but the earth remains forever. Tools come and go. Frameworks rise and fall. This year’s astonishing model will be next year’s ordinary one.

What remains is the work of being a good person who makes good things for other people. That was the job when I started, twenty-five years ago. It’s still the job now. I suspect it always will be.

Wishing everyone a healthier, calmer, more peaceful year.

Everything changed this year. The important things didn’t.