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 found the productivity dividend leaking away into supervision, workers spending more time checking AI than doing the work it was meant to speed up. The fix was designing better inputs, which takes people who know the work cold. This issue is about where those people come from, and whether that supply is quietly drying up.
THE ONE TO READ
If you only read one AI implementation article before your next leadership meeting, make it this one.
The cheapest work to automate is the entry-level work. It is also where your future experts are made. A Harvard Kennedy School paper pulls six datasets together to show what that trade is costing.
Future of Work in the Age of Automation, Augmentation, and Agentic AI — Sol Rashidi, Harvard Kennedy School (M-RCBG Associate Working Paper No. 276, Jul 2026) | FREE
Start with the thing every prior wave of automation left alone. Machines took the routine physical tasks, the procedural ones, the work you could write down as steps. What they never touched was the ladder itself, the entry-level jobs where people learn the work by doing it badly, then less badly, until one day they know enough to supervise someone else. Rashidi’s argument is that AI is the first technology to automate that rung directly. The execution-heavy tasks you used to hand a 24-year-old are exactly the ones a model does now.
She calls it a talent formation fracture, and she is careful to say the concern is not job loss on its own. It is what happens to the pipeline that turns junior people into senior ones.
The numbers she assembles are not hers, and that is the point. She pulls them from six independent sources and lets them converge. Stanford’s Digital Economy Lab finds early-career workers saw employment fall 6 to 13% between late 2022 and mid-2025, while older workers in the same jobs held steady or gained. The firm-level studies land harder, a 16% relative decline for young workers. Handshake reports graduate-level job postings down 15 to 20%. The New York Fed puts recent-graduate unemployment at 5.6%, an all-time high. In software, entry-level hiring fell about 20% even as demand for senior engineers rose. Consulting, accounting and legal saw entry-level hiring drop 15 to 25%.
Any one of those is a data point. Six of them pointing the same way is a pattern, and the pattern is that firms are meeting AI by cutting the bottom of the org chart first.
That would be a manageable problem if entry-level work were just cost. But it is not, it is how expertise gets built, and you cannot automate the junior tasks and still expect the senior people to appear on schedule five years later. The supervisor who spots the model’s mistake in Issue #4, the person who knows the work well enough to design the inputs, that person is made by doing exactly the work AI is now absorbing. Cut the rung and you do not save money, you defer a capability bill to a future quarter when it is much harder to pay.
Here is where it ties back to where we started. Rashidi’s own read of the field studies is that workforce architecture redesign is what separates the 6% of firms capturing real AI value from the 94% that are not. That is the same split we opened this newsletter with in Issue #1, the small cohort pulling away from everyone else. She is describing it from the talent side. The firms that will still have experts in 2031 are the ones treating early-career work as capability formation rather than a line item to automate.
Which is the whole argument in one sentence. You can implement AI without cutting the capability you are about to need. But it is a choice, and most organizations are making the other one without noticing they have made it.
LEADERSHIP LOAD OUT: BRING THIS TO THE MEETING
Ask one question about your last round of AI-driven efficiency: did it remove tasks, or did it remove the jobs where people learn those tasks? If nobody knows, you do not yet know whether you bought productivity or borrowed against your own bench.
ALSO ON OUR RADAR
A complication to the read above, and the skill that resists the fracture.
The Companies Spending the Most on AI Are Also Spending the Most on Humans — Revelio Labs, with Ramp (Jun 30, 2026) | FREE
A useful complication to the read above, not a rebuttal of it. Across 21,000 firms the heaviest AI spenders grew headcount 10% and lifted their entry-level share, while everyone else saw nothing. The catch, which Revelio names itself, is that those firms were already growing faster before they adopted anything. So this is aggressive, well-funded companies doing what they were going to do anyway. The fracture is not adopters versus the rest. It is a narrow, well-capitalized band versus everyone else, most adopters included.
Beyond Vibe Coding: A Designer’s Case for Directed Generation — Jim Gulsen, UX Magazine (Jun 25, 2026) | FREE
Gulsen argues that designers who accept the “vibe coding” label are handing away authorship of their own practice. The real skill is judgment before generation: curating the reference, setting the constraints, reading the output critically instead of accepting it. The capability that resists the fracture, described at the level of a single practitioner.
PARTING THOUGHT
The entry-level job was never really about the output. It was the tuition an organization paid to grow its own experts. Automate it away and the savings show up this quarter. The bill shows up in about five years, and it is payable in people nobody remembered to train.
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