I had AI's effect on companies wrong for about a year and a half, and I was giving founders advice out of the wrong model the whole time.
Through 2023 and most of 2024, I assumed the first big effect would land on what companies shipped: new interactions, new product shapes, new categories. The teams making all of that would be roughly the teams that made the previous wave of products, just moving faster. That is what I planned around, and where most of my advice was coming from.
Extrapolating from my own desk
I picked the error up honestly. My own output as an individual contributor sped up a lot through 2023, and I extrapolated from one person to organizations, which was sloppy. One senior person speeding up doesn't tell you what happens to a team's work. I could do twice as much, and I took that to mean teams would work faster on the same surface area, instead of asking the more interesting question: what happens when each person can hold what used to take three of them.
Outside design and engineering, the original belief was closer to right. Sales, marketing, and support changed at the output layer first. Buyers got AI-written copy, agents in the loop, and summarized tickets long before those organizations got visibly smaller. I generalized from that to design and engineering, and I shouldn't have. Making and distributing are not the same thing, and the compression in making landed on the team before it landed on the product.
A gap the usual explanations don't cover
I noticed it when I watched a team (not at GrowthX; a Noord portfolio company) ship something that should have needed three or four times its headcount. I tried the usual explanations. They're senior. They're fast. They had a head start. Those cover a 1.5x gap, maybe 2x. They do not cover 4x. The only explanation I could make fit was a change in the unit of work itself.
GrowthX is where it got clearest for me, because I'm inside the team-shape decisions every week. Look at what we haven't hired. The product surfaces we ship would have needed three or four discrete roles in 2022: research, design, brand, design systems. We do them with a small senior team. Budget is not the constraint, and sourcing isn't either. The work has gotten dense enough that additional designers don't add what they used to.
Engineers got there before I did. Several of them, across teams at GrowthX and across the Noord portfolio, were already running at the compressed density through 2024, fewer roles with broader scope inside each, while I was still planning design orgs at the 2022 shape. A lot of the friction I watched in design organizations that year came from exactly that lag. We were hiring for a team shape that no longer matched the work.
The unit of work changed before the product did. What one designer or engineer can hold has roughly tripled in two years, and the teams adjusted before anything on screen did.
I said no to all three
The advice I give has changed, and that is how I know the shift is real. Three founders asked me this quarter whether to hire a second designer. Two years ago I would have said yes without much thought. I said no to all three. The reasoning differed each time: not yet; hire one level more senior and don't backfill; learn the tools well enough to write the spec yourself before you fill the role. The underlying advice was the same.
What it asks of leaders
The uncomfortable half of the argument is what it does to people leading these functions, my own cohort included. For twenty years the deal was that you traded the tools for the org chart: seniority meant your hands left the work. That deal has expired. If you lead engineering, product, or design and you are not hands-on with the toolset required to ship, or at least visibly learning it, you will struggle to attract the people who are. The strongest builders now calibrate a leader the way they once calibrated a portfolio, and they can tell in one conversation whether you have touched the tools or merely commissioned slides about them.
I don't say this from a comfortable distance. I went back to shipping production work after years of mostly writing org charts, and the first months were humbling in the specific way that matters: my taste had kept current while my hands had not. Closing that gap changed my advice more than any market map did. Double down on the judgment only people can supply, and let compute fill the gaps. Teams built that way are small because nothing is missing from them.
Late together
What makes a shift like this hard to feel is that you can be quite late and not notice, because so is everyone you talk to. The catching up is happening at every company more or less simultaneously, so conversation doesn't tell you where you are. The teams that adjusted earliest are a year to eighteen months ahead, and you can't see that lead from the outside. I can see it now mostly because of where I sit, inside one company and across a portfolio, and the spread between earliest and latest is wider than I would have guessed.
Most of the AI products people use in 2026 still look a lot like the AI products of 2024. The more important difference is behind them: the teams are smaller and more senior than they were. Product shapes will follow. They usually do.
I'm Head of Design at GrowthX and run Noord, an early-stage fund and studio in Amsterdam. The vantage point is the whole argument, so discount accordingly. The counter-argument I take most seriously: every tooling shift in memory produced a season of small-team triumphalism, and the org chart grew back once the novelty premium faded. If that happens again this essay becomes a period piece. I don't think it will, and I have been wrong about this subject once already.
© 2026 Renato Valdés-Olmos