The AI Implementation Report is a fortnightly newsletter for leaders navigating AI strategy. Each issue highlights the research, data and trends that matter most for implementation, not the hype, not the doom, just what is actually working and what is not. Curated by Mark Phelps and Drew Horton. A Mark Phelps LLC production.
Last issue we looked at what happens when the entry-level rung gets automated away, and where the senior people are supposed to come from once it has. This issue is about the people still standing on the ladder. Their jobs held, something else moved instead.
THE ONE TO READ
If you only read one AI implementation article before your next leadership meeting, make it this one.
Two years of research asked whether AI was taking jobs and mostly found that it was not. An Apollo paper asked a different question and found the adjustment hiding in the pay data.
The Impact of AI on the U.S. Labor Market — Sania Edlich and Torsten Slok, Apollo Global Management (Jul 30, 2026) | FREE
Nearly all the AI labor research so far runs on exposure scores which estimate how automatable a job looks on paper. Edlich and Slok used observed usage data instead, then ran a difference-in-differences design with occupation and year fixed effects across 321 matched occupations from 2015 to 2025.
High-exposure occupations saw real wage growth fall 6.7% after 2023. Employment effects were not detectable. The jobs stayed. The raises did not.
The distribution is the part to think about: bottom-quartile workers saw 10.7% slower real wage growth and service workers saw 24.3%. Top earners showed no significant effect at all. The paper puts a conservative figure on the aggregate, around $28 billion in annual labor income across 5.8 million workers in high-exposure roles.
Now hold that next to the Budget Lab at Yale, which has spent a year publishing CPS updates concluding there is no meaningful AI effect on the labor market. That looks like a contradiction until you notice the two are not measuring the same thing: Yale watches employment and occupational mix, while Apollo watches wage growth. On employment, Apollo agrees with Yale.
So TLDR; the useful reading is not that one of them is wrong. It is that many firms appear to be taking the productivity gain through compensation rather than headcount, which is invisible to anyone counting jobs.
Which lands back where we were last issue. Cut the bottom rung and you defer a capability bill. Keep the rung but quietly stop paying it forward and you get a slower version of the same erosion, with no announcement and no press cycle. A paycheck that does not move is a much quieter thing than a layoff.
Worth naming: the usage data behind this paper comes from Anthropic’s Economic Index. The analysis and the conclusions are Apollo’s own.
LEADERSHIP LOAD OUT: BRING THIS TO THE MEETING
Ask your firm’s finance lead one question about the roles where you deployed AI this year. Not what happened to headcount, but what happened to the merit pool. If it flattened while output rose, you have already run this experiment on your own people, you just did not write it down. But, the teams probably noticed.
ALSO ON OUR RADAR
Three angles on the same question, plus one that follows it into the interface. Two of these land on the same finding from opposite ends: pull out the layer that was carrying accountability and it does not redistribute by itself.
Nvidia’s Huang and Google DeepMind’s Hassabis call out “lazy” AI layoff logic — HR Executive (May 31, 2026) | FREE
Huang told Channel NewsAsia that tying job cuts to AI is “just too lazy,” asking how layoffs announced two years ago could be blamed on tools that only recently became broadly useful. Hassabis made a similar point about replacing developers, calling it a failure of imagination. Read against the lead, this is the two men selling the technology telling you it did not do what you said it did. That is an argument against their own interest, which is exactly why it is worth hearing. Watch the alternative Huang offers, though. His pitch is do more with more, and more of both runs on his chips, weird.
Builders, Sellers, Measurers: Why AI Is Restructuring Every Company Whether They Plan For It or Not — Andrew Baker (Jun 29, 2026) | FREE
Long but worth it. Baker is a sitting bank CIO and he takes our last issue apart in a useful way. His Ford section puts a headcount on the sequencing problem: the automated inspection tools got worse once the engineers who understood the edge cases were gone, so the company brought 350 of them back. Then he argues the apprenticeship crisis is an environment problem rather than a people problem, because juniors learn inside teams with review gates, and the failure is not giving them AI but deciding not to hire them. His last move is the one that stays with us. Strip out the coordination layer and you also strip out the place accountability was quietly living.
When the US government switched off AI — CoreStream GRC (Jul 2, 2026) | FREE
In June the US Commerce Department ordered two frontier models suspended for any foreign national anywhere, and because nationality could not be checked in real time the vendor took them down for everyone inside 90 minutes. Access came back on July First. Two weeks after the first order a different part of the government used a different instrument on a different company with different notice, which is the part that should worry you more than the outage. If your continuity plan does not survive that paragraph, it may not be a continuity plan.
Defining ethical design for machines — Michael Buckley, UX Collective (Aug 6, 2026) | FREE
Design teams are now writing plain-text files that tell AI how to make design calls, and some of them are reaching past spacing and color into autonomy, transparency and privacy. Buckley, who teaches design at Seton Hall, works through what that actually requires and lands somewhere uncomfortable. Writing ethics into a file does not remove the judgment, it moves it into whoever wrote the file, where nobody can see it or argue with it. His one concrete rule is a good test to steal: cancelling should take no more steps than signing up. He is building his own framework here, so read it knowing that.
PARTING THOUGHT
A layoff is legible. It has a date, a memo, a number somebody has to defend on a call. Wage compression has none of that. It is a raise that was discussed and then was not, in a year when the work got easier and the output went up. Nobody has to announce it and nobody has to defend it. That is what makes it the easier thing to do, and the harder thing to notice.
Know someone navigating AI implementation? Forward this to them.
Hit reply and tell me what you think. We read every response.
And if there is something you think we should cover, send it over. We plan these issues as one connected argument, but the best suggestions come from readers, and we will give you props when yours lands.
Published for AI implementation leaders in the US and UK.
Mark Phelps is a product and design leader with more than 20 years of Fortune 100 experience at GE, Fidelity, TIAA, Bank of America and Marsh McLennan. As the founder of Mark Phelps LLC, he serves as a Fractional VP of Product Design with a focus on AI-driven environments.
For more than three years he has worked as a fractional product design leader, first at Rockefeller Capital Management, where he was engaged to provide critical process improvements, and then on an AI-powered healthcare software product, brought in for urgent user interface deliverables. Both engagements were extended well beyond their original scope. He is a named inventor on a pending design patent for the product’s interface.
He holds a Stanford AI Certificate earned with Distinction and an MBA in Digital Strategy from Suffolk University (Beta Gamma Sigma). He is a United States Marine Corps veteran.
20 Portsmouth Ave., Suite One #161, Stratham, NH 03885