Harry Stebbings is betting dev teams shrink 30 to 40 percent as AI token budgets hit $100,000 per engineer

Harry Stebbings, the venture capitalist behind 20VC, has put a precise shape on what he thinks AI-assisted engineering teams will look like once the transition settles. The bet is not abstract. Each top engineer gets a $100,000 annual token budget. The dev team around that engineer shrinks by 30 to 40 percent. That is the trade, as Stebbings frames it, and he expects it to hold as the steady state.

The logic is simple enough that a department head can run it in a meeting. A senior engineer’s fully loaded cost sits well above six figures in most technology organizations. If a $100,000 token budget replaces a meaningful share of the surrounding team’s output, the arithmetic in favor of reallocation becomes hard to argue against. Stebbings is not predicting that engineers disappear. He is predicting that fewer of them, each equipped with significantly more AI capacity, will replace what larger teams currently do.

What makes the call worth tracking is its specificity. Stebbings is not gesturing at a direction. He is naming a number ($100,000 per engineer equivalent) and a ratio (30 to 40 percent team reduction). Those are figures an organization can look up in its own budget and headcount data within a year or two. A forecast this concrete either lands or it does not, and the terms of failure are as clear as the terms of success.

I think it'll land at 100 grand per engineer equivalent. I think that's what we'll I think we'll give each of our best engineers $100,000 of tokens and in return we'll cut the size of our dev teams 30 40% effectively. Harry Stebbings

The harder question is whether the headcount reduction actually follows the spending increase. Allocating more on tokens per engineer does not automatically mean organizations cut bodies. Budget lines for AI tooling could simply be added on top of existing team sizes, particularly in organizations where engineering capacity is already constrained or where leadership is reluctant to act on the math. The 30 to 40 percent figure assumes that decision-makers will make the explicit swap, not just expand both columns simultaneously. That requires a kind of organizational discipline that has historically been uneven across technology companies, even when the financial case is straightforward.

There is also the question of which engineers absorb the $100,000 budget and which roles find themselves on the wrong side of the reduction. Stebbings specifies “best engineers” as the recipients of the token allocation. That framing implies a sorting mechanism: the engineers whose judgment and direction are worth amplifying get the tools, while those whose value was primarily in execution volume become candidates for reduction. Whether organizations can make that distinction cleanly, or whether headcount cuts will be blunter than the model suggests, is genuinely uncertain. The bureaucratic reality of performance distinctions inside large engineering organizations rarely maps neatly onto the kind of tiered allocation Stebbings describes.

The structural argument underneath the forecast is that AI tooling has already crossed a threshold where the productivity gap between an augmented engineer and an unaugmented one is large enough to justify a team-design rethink, not just an individual productivity improvement. That is a stronger claim than “AI makes engineers faster.” It is a claim that the unit economics of software development are being reset at the team level, not just at the individual workstation.

What Stebbings is describing is not a gradual drift but a deliberate restructuring decision. Someone with budget authority has to decide to hand a top engineer $100,000 in token spend and then hold a harder conversation about team size. His forecast implies that decision will become common enough to define an industry norm, a new default configuration for how engineering organizations are staffed and resourced. The premise is falsifiable and the timeline is near enough that the evidence will arrive before the forecast fades from memory. Whether the organizational follow-through matches the spending trend is the part that remains to be tested.