The AI job diamond still needs a bottom rung

IBM Technology has a useful frame for the AI jobs debate: don’t picture the org chart as a pyramid anymore. Picture it as a diamond.

The old pyramid had lots of entry-level people doing task work, fewer experienced people owning chunks of work, fewer seniors leading individuals, and a thin layer of executives leading leaders. The AI version, IBM argued, narrows the traditional entry-level layer and expands the experienced layer. Not because everyone magically becomes senior overnight. Because the unit of work changes.

That is the part I buy.

The part I worry about is the missing apprenticeship system.

AI turns task-doers into work-owners

IBM’s strongest point is that AI changes what a junior person can carry. A programmer who used to write one module may now coordinate an “AI team” that drafts code, writes tests, explains errors, generates documentation, and proposes integration options. Same human. Bigger surface area.

That is not job replacement in the simple sense. It is role compression. Some of the task layer gets automated, and some of the people who would have lived there move up into coordination, review, and ownership faster.

This maps to what I see in actual AI workflows. The person with the model is not just doing faster typing. They are specifying intent, checking assumptions, deciding what is good enough, and catching the weird mistakes. The job becomes less “produce the artifact” and more “run the production system.”

IBM also brought up Jevons Paradox, the idea that making something cheaper can increase total demand for it. If AI makes software, analysis, content, customer support, or internal tooling cheaper, companies may want more of all of it. That could mean more work, not less. But Jevons does not guarantee better jobs. It only says demand can expand when cost falls.

The quality of those jobs depends on design.

The danger is deleting the training ground

Here is the catch: entry-level grunt work was never just output. It was also how people learned taste.

You learned what a bad ticket looks like. You learned why a customer complaint is messy. You learned where code breaks in production. You learned which spreadsheet numbers are fake precision. Boring work trained judgment.

If AI absorbs that layer, companies cannot just tell new hires to “manage agents” and hope for the best. Agent management is not an entry-level skill if the person has no model of good work. It requires domain context, error detection, prioritization, and the confidence to reject plausible nonsense.

So the diamond only works if the bottom rung becomes smaller but more intentional. Less repetitive task grinding. More structured practice. Shadowing. Review loops. Simulated cases. Human sign-off. Clear rubrics for what good looks like.

Otherwise, we get a fake promotion: juniors supervising systems they do not understand, seniors spending their days cleaning up invisible mistakes, and managers wondering why throughput went up while trust went down.

The promotion is real only if accountability moves with it

I like IBM’s optimism because it is operational, not magical. The claim is not “AI will save every job.” The claim is that smart organizations can move people into higher-value work by giving them AI capacity.

But that word “smart” carries the whole argument.

A smart org will measure more than headcount savings. It will ask which tasks are now machine-first, which tasks still teach humans, which outputs need review, and which decisions should stay with accountable people. It will redesign ladders, not just buy licenses.

A lazy org will flatten the base, call it efficiency, and discover later that it also flattened its talent pipeline.

Practitioner’s take: pick one role in your team and map its work into three buckets: tasks AI can draft, tasks humans must judge, and tasks that train judgment. Automate the first bucket carefully. Protect the third bucket on purpose. Then give junior people a real AI workflow with review checkpoints, not a blank chat box and a productivity target. The catch most teams miss: the goal is not fewer juniors. It is faster formation of capable owners.