In January, I started collecting thoughts about the AI change in this little blog. Small details you miss while riding the wave. Reading them again now, most hold up. A few mattered more than I expected.
The ones that stick out weren't the most quotable. Nobody quotes "delegation is harder than it looks" at a conference panel. But it's the one that kept showing up in real rooms: people don't hand work to a machine just because they trust it. They need a vocabulary for what that thing actually is first. And most organizations never had to build that vocabulary before, because the work lived entirely in someone's head.
That observation (the real lever is organizational, not technical) is the one I keep coming back to without having to defend it anymore. This spring it quietly stopped being a contrarian take. The big studies now say some version of the same thing: technical capability is moving faster than organizational readiness, and closing that gap is structural work, not a personal failing.
What's actually changed since the beginning of 2026: the chat interface stopped being the interesting part. Whatever conversations I have now are about agents that act across multiple steps without being told each one, and about what it takes to make that hold together. Token costs turned from a footnote into a line item. (Funny side note: dependency on a handful of American infrastructure providers is turning into a literal procurement question for a lot of people building this. Sovereignty, but the boring kind.)
My own thinking got less original and more useful, if that makes sense. Less "here is a new frame nobody else has," more "here is an old pattern, dressed differently." Most of what's demanding about this transformation has the same roots as what made digitization and cloud migrations demanding a decade ago. It just moves faster and touches identity more directly. Two things sharpened for me specifically: self-organizing teams aren't a nice-to-have for AI adoption, they're close to a precondition. And there's more upside in honesty about AI use than I gave it credit for. Not leadership lying about strategy, but individuals quietly not mentioning how much of their actual output runs through a model, because saying it out loud still feels like admitting something. I'm starting to think an organization's speed has less to do with its software licenses than with whether it's made it safe to say the unsaid out loud.
So where does this go next, not just for me, for anyone watching this from inside an organization right now? My best guess: the efficiency story stops being the main story this year. Some teams will report 10%, some 15%, plenty will report nothing measurable, and a handful will report 50% and get held up as the standard everyone else should already meet. That comparison will create more skeptics than converts. The real shift is happening somewhere else. What will actually carry the conversation by December is the smaller number of cases where work got reinvented rather than sped up, done differently enough that comparing it to the old version stops making sense. Organizations that get there will have learned, through doing rather than theory, that the push has to be at least half bottom-up, that individual people need their own moment of actually seeing it work before they believe it, and that the team itself, not just the tools the team uses, gets to be redesigned.
That last part is where I want to spend my own time next: not the strategy layer, but what it actually takes to bring people who don't write code, don't think in systems, and didn't ask for any of this into a way of working that holds together. At scale, not in a single workshop. I don't have that answer yet. That's where I'm headed next.
The technology will keep moving. The interesting question was never whether we'd keep up. It's who we become while we're trying to.