So conventional wisdom out there right now says, if a AI-model can draft an architecture diagram, ship working code, and review pull requests, a senior engineer should be worth less, not more, right?
Well, employment data says otherwise...
In February this year, the Dallas Fed published a study covering more than 200 different occupation types since ChatGPT launched in November of 2022.
A couple noteworthy findings:
- In computer systems design (the most AI-exposed slice of the whole economy), wages grew 16.7% while average weekly wages across the US grew 7.5%.
- Over the same period, employment in those sectors fell about 1%, while overall US employment grew 2.5%.
The same data was broken out by age and found something even more interesting:
- Employment for software developers aged 22 to 25 dropped nearly 20% from 2022.
- However, developers aged 30 and over in the same AI-exposed fields grew 6 to 12% over the same period.
- While entry-level US software postings fell 67% between 2023 and 2024.
No surprise. People have been talking about entry level engineering shrinking for a while now.
But the question I keep coming back to is:
Why does it appear that experienced talent is getting more valuable in the marketplace, not less?
I believe the sooner we understand exactly what's driving this, the sooner we can restructure our teams to properly compete. My gut says the agencies that figure this out early are going to have a massive advantage over the already saturated agency landscape.
The coordination problem
Depending on the project, let's say the traditional way to ship a robust website or app means assembling these speciabullets: a frontend dev, a backend dev, solutions architect, QA engineer, researcher, strategist, and designer. Then at least one more person to manage handoffs between them, say a PM or TPM.
By Brooks' Law, a team of 8 people carries 28 communication channels. A team of 9 people has 36. Every channel is a chance for someone to mishear, miss-prioritize, or even lose the thread entirely. Each person you add doesn't just add one more voice. It multiplies the number of ways things can go sideways.
Just to hammer this point home, UC Irvine's Gloria Mark found that on average, knowledge workers get interrupted every 3-minutes and take 23-minutes to fully recover a high focused state. Microsoft's 2025 Work Trend Index put the ping rate at 275 per day per employee, with 58% of the workday going to 'coordination overhead' rather than actual skilled work.
And that's just the cost of keeping everyone in sync!
Now, as a thought experiment, imagine a senior operator who, in an extreme case, holds all seven of those jobs competently, with AI absorbing the production work. Most teams won't get there, but the point is that what used to require seven people might now require two or three. All those communication channels drop to zero because the integration happens inside one head.
There's no standup to resync on what changed yesterday, no translation layer between design and engineering, no researcher writing a deck the strategist will only half-read. And no lag between a decision and the work that follows it. When one person holds the context, feedback doesn't have to wait for the next meeting, and instructions don't get lost in the handoff. That alone shrinks the overall project timeline in a big way.
And because agencies are so focused on output, most are underestimating how much AI rewards system designers over producers. The best engineers today are building the factory first: the architecture, the constraints, the quality checks, the feedback loops, the reusable components. Once that's done, everything downstream gets faster, handoffs get shorter, and the output becomes much more predictable. The value has moved to system design, and most agencies just haven't caught up to that yet.
What this all means for how you hire
Today, the most valuable hire is the operator who can hold the whole project in their head and use AI to absorb the production work that used to take seven people. That role used to get called a "generalist", almost as an insult. A more appropriate name for it now, I think, is "integrator" or "systems operator", and the market is pay big league for it.
In practice, this means your interview process needs to change. Asking someone to complete a coding challenge or show a portfolio only tells you what they produced. The more useful question is how they think through a problem end to end: how they scope it, where they see the risks, when they'd push back on a brief. Those are the things AI can't do for them. How you actually interview for that is a whole topic on its own, and something I'll write about separately.
Conclusion
Yes, execution is getting cheaper. But also, being wrong about what and how to build is getting more expensive too. Which is why a small team of uber-talented folks who've already paid for those lessons will run circles around a bigger team that hasn't.
The market is already voting, and we're trying to stay ahead of it ourselves. There's real anxiety out there right now. Some people are genuinely unsettled by how fast things are moving, and I get it. But I think if we keep paying attention to the latest developments and stay honest about what the data is showing us, there's still time to make adjustments. And if you believe, like I do, that we're going to be okay, there's never been a better time to build.
If you're running an agency and you haven't started asking these questions yet, now's a good time to start. The org chart is changing and it's happening faster than most agencies realize.
