Analysis · 20 October 2026 · 7 min read

Pyramid to diamond: AI and company structure

Is AI turning companies from pyramids into diamonds? What Anthropic's hiring data, startup research and a new economic model show, and what it means for you.

For most of the last century, a company looked like a pyramid. A wide base of junior staff did the drafting, the data gathering and the routine checks, a narrower layer of experienced people reviewed their work, and a few leaders sat at the top. The juniors were not only cheap labour. They were the pipeline that produced the next generation of seniors.

The claim now circulating is that AI is turning the pyramid into a diamond: fewer juniors at the base, a bulge of experienced workers in the middle, and the same thin top. On 16 October 2026 Anthropic published data from inside its own company that fits that picture closely. Here is what the evidence actually shows, where the diamond idea comes from, and what it means if you are early in your career.

Where the diamond idea comes from

The image is not Anthropic's. PwC's chief AI officer predicted in 2025 that AI would shift the traditional labour pyramid to a diamond shape, and the phrase has since become shorthand for the hollowing out of entry level hiring. The most careful treatment is an April 2026 working paper by Guido Friebel, Yao Huang, Jin Li, Soumitra Shukla and Andrew Zhang, Pyramids, Diamonds, and Oscillations, which models how a firm decides the ratio of juniors to seniors when AI arrives.

Its central point is that it matters what AI does. The paper separates three effects:

• AI makes seniors more productive.
• AI makes juniors more productive.
• AI makes juniors learn faster, so they qualify for senior work sooner.

For the first two, the model says firms freeze junior hiring in the short run, which pushes the shape toward a diamond, but in the long run a pyramid stays a pyramid. Only the third effect, faster learning, permanently turns a pyramid into a diamond. On the way there, firms can swing back and forth: a hiring freeze starves the pipeline, which causes a senior shortage later, which triggers a surge in junior hiring. The authors call the people who enter during the freeze a lost cohort.

That is a model, not a measurement. But it gives the right question to ask of any data: is this a temporary freeze, or a permanent change of shape?

What Anthropic's data shows

Anthropic's report, How employment is changing at Anthropic, is the most detailed look yet inside a company that has rebuilt its work around AI. The relevant numbers:

The base of the technical pyramid shrank while the company tripled in size over the past year. Total technical hiring still grew; it is the mix that changed. Postings for junior roles fell especially sharply after December 2025, and by April 2026 around a third of machine learning managers and a fifth of software managers said they were hiring less because of AI.

Two other details matter for the shape. The middle layer is changing too. Managers made twice as many code commits as before, and there are now about 17 engineers for every product manager, a ratio that has risen since 2025, so supervising roles increasingly include building. And administrative work, which normally grows with a company, barely did. One executive assistant told the researchers that the entry level admin role had already been automated away, and that their own job had moved toward chief of staff work.

Is this one company, or a pattern?

Anthropic is an unusual employer: it builds the tools it uses, and its revenue per employee is two to four times what Google and Meta took in at the same age. On its own it proves little. But it is consistent with three other strands of evidence.

Startups built around AI are smaller and flatter. A Harvard Business School working paper by Hyunjin Kim and Rembrand Koning, AI-Native Firms, compared startups that raised their first funding between 2020 and 2024. Firms built around AI were 25 percent smaller than comparable startups, with roughly 15 percent lower shares of both entry level workers and managers, and hierarchies about half a seniority level flatter, at similar valuations.

Junior employment falls first. Stanford payroll research found a 16 percent relative employment decline for workers aged 22 to 25 in the most exposed occupations, which our entry level post covers in detail. A separate study of résumé data on 62 million US workers found that firms adopting generative AI sharply reduced junior employment.

But not everywhere. Research from Denmark cited in the diamond paper found that even as early career employment fell, the firms actually adopting AI mostly reorganised tasks internally rather than cutting headcount.

Note that the startup evidence is not a perfect diamond. Those firms have fewer juniors and fewer managers, which is a smaller, flatter organisation rather than one with a bulging middle. What the strands agree on is the bottom of the shape.

Why the bottom of the pyramid is the exposed part

This lines up with what our own index measures. Our scores rate how much of each occupation's task content overlaps with what current models can do, and the tasks that score highest are the ones juniors traditionally do: first drafts, research summaries, routine code, document review. At Anthropic, junior employment slowed most in the jobs where Claude was already handling the most complex tasks.

Two occupations where that shows up, as our index scores them: Computer Programmers (59) and Secretaries and Administrative Assistants, Except Legal, Medical, and Executive (58).

The open problem is the one the diamond paper puts at its centre. If the bottom rungs disappear, where do tomorrow's senior people learn? AWS chief executive Matt Garman made the same point bluntly in 2025: a company that stops hiring juniors will have nobody who has learned anything in ten years. Anthropic's report admits that its own measures for current staff, such as internal redeployment, do nothing for the people who were never hired.

What it means if you are early in your career

A diamond still has a way in, but it is narrower and it rewards different things. At Anthropic, hiring managers' notes show what changed: for technical candidates, comments on autonomy and judgment rose the most, mentions of creative use of AI tools tripled for technical roles and quadrupled for others, and discussion of narrow domain expertise shrank. Our post on what Anthropic's own data shows goes through those hiring changes and the rest of the report.

The practical reading is to aim for roles where you get near real decisions early, rather than two years of drafts someone else signs, and to treat fluency with AI tools as the minimum rather than the differentiator. When will AI take your job covers why these shifts arrive unevenly across firms and fields.

Our index scores 961 occupations from published research. The assessment shows how much of your own week looks like the tasks at the bottom of the pyramid. It takes about two minutes, free, no account.

Figures come from the AI Job Risk Index and were current when this was published. Scores change as new research is ingested, so the index is always the live version. See how scoring works. Informational guidance based on published research, not professional career or financial advice.