AI won't shrink your engineering team. It will change what the team is for.

At some point in almost every executive conversation about AI, the question arrives — usually carefully phrased, sometimes not: if the tools make developers this much faster, can we run the same roadmap with fewer of them?

It’s a fair question. We just think the evidence says it’s the wrong one.

Gartner’s Software Engineering 2030 research takes an unusually firm position here: AI fuels demand for more developers, not fewer. World Economic Forum data from 2025 projects developer roles growing 57% between 2025 and 2030 — because of AI, not in spite of it. Economists have a name for the mechanism: the Jevons Paradox. Make a resource cheaper to use, and people use more of it. Making software cheaper to build has not reduced anyone’s appetite for software. Quite the opposite.

The backlog you stopped counting

There’s a second reason the fewer-people sum doesn’t work, and it lives inside your own estate. Thoughtworks describes it as a mountain of “accidental complexities”: decades of workarounds, undocumented systems and postponed modernisation that every large organisation carries — and has quietly stopped calling backlog, because fixing it was never going to be affordable.

AI is rewriting that arithmetic, most dramatically in modernisation. One case Thoughtworks documents involved a luxury carmaker reverse-engineering a fifteen-million-line legacy codebase — previously six weeks of work for two people per module — in hours. Another organisation cut bug investigation from eight days to three. Capacity on that scale doesn’t argue for a smaller team. It argues for finally doing the work you never could afford: retiring legacy risk, unpicking technical debt, modernising the platforms your growth plan quietly depends on.

What engineers become

The role changes even where the headcount doesn’t. Gartner sketches the progression — AI-assisted tasks, then orchestrating work across people and tools, then translating business intent straight into action through agents. The scarce skills shift with it: judgement, system design, verification, and the underrated ability to sit with business stakeholders and argue about what should be built at all. Thoughtworks adds a point we’d echo from our own teams: engineers freed from routine work don’t sit idle. They finally get to the strategic work everyone has spent years complaining there was no time for.

Gartner’s warning is blunt by research-house standards: leaders who cut their workforce out of a misplaced efficiency plan will be outperformed by those who upskill and redeploy. Put more plainly — the risk isn’t carrying too many engineers. It’s discovering in two years that you released the capacity your competitors used.

The question that actually matters

So the headcount question becomes an allocation question: where is the freed capacity actually going? That one can’t be answered on intuition. You need to see where delivery effort flows, where it leaks into rework, and whether the hours AI hands back are being reinvested in modernisation and innovation — or silently absorbed by the same old friction. In our experience, without instrumentation it is almost always the latter, and nobody notices for a year.

That visibility is the part we obsess over. OnTrack AI™ exists so delivery leaders can see, rather than guess, how capacity, quality, cost and risk move as AI takes on more of the work — and steer the reinvestment on purpose. If your board is asking the fewer-people question, the better conversation starts with what you’d do with the capacity.