An open debate on the size of AI squads
As coding agents gain ground, a debate has opened up on the size of AI squads. Part of the market believes the squad could eventually shrink down to a single Product Builder, who would manage the product roadmap and personally steer the agents responsible for producing the code.
This view is not absurd. Our own reading of the levels of AI engineering shows teams gradually getting smaller:
Our field experience also confirms that gains are higher with very small teams. The autonomous factory model, with a single pilot, is part of our roadmap, but for the 2029-2030 horizon and subject to mature governance and risk management.
Without trying to predict the future, three realities make us skeptical today about the viability of a squad resting on a single person.
1. The reality of the squad: pressure concentrated on people
Automation speeds up production, but it does not eliminate human work: it moves it. With cycles of 2.5 days to 1 week, the bottleneck is no longer development but specification, validation and decision-making. These steps remain human: AI produces, the Product Builder checks before delivering, and nothing is shared without that review.
The faster the factory runs, the heavier the review and decision load becomes for those overseeing it. We already see this in two-person squads (one Product Builder, one developer), where the pressure of near-real-time work is intense and the risk of burnout is real. With a single person, there is no longer a second pair of eyes on what the AI produces, and every absence becomes a risk to product continuity.
2. The reality of skills: two demands rarely found together
Product Builder and AI Engineer are new professions. The Product Builder has to go further than the Agile-era PO: product vision, discovery through to prototyping, hands-on design, steering delivery and adoption. The AI Engineer must master architecture, quality and security to be able to validate what the agents produce. A one-person squad assumes both sets of demands can be met by a single profile.
Yet across five editions of the Niji Barometer, organizations have consistently pointed to how hard it is to find good PO/PMs in Agile mode. In 2025, only 36% said they had mastered product skills. With the bar set higher in an AI model, these profiles will have to be built (career tracks, learning paths, coaching) rather than waited for.
3. The reality of companies: a trade-off between optimum and risk
Factories build and maintain the IT that keeps the business running: revenue, processes and production depend on it. The 2026 Barometer identifies the top risk as the gap between how fast AI is evolving and the time organizations need to adapt. In 2026, only 23% anticipate the impact of AI on their skills, and only one in six has built a transformation plan. The most complex question will therefore be transforming teams and defining the associated Workforce Planning, well before the question of the ideal AI squad size.
In this context, we think it unlikely that companies will go straight from an 8-to-10-person feature team to a one-person team solely responsible for business continuity. Team size will be a compromise between the ideal optimization (one or two people) and risk control. As a first step, going from 8-10 people to 3-4 is already a major change, while still providing reassurance.
Key takeaways
Our target model relies on tight squads of 3 to 4 people (Product Builder, AI Lead Engineer, AI Engineer), backed by a shared Design Authority and steering cockpit for architecture, the design system, cybersecurity, FinOps and support. Rollout happens in stages, starting with a dedicated factory alongside the existing ones. Squad size is meant to be reassessed at each stage, as governance, skills and tooling prove their robustness.
In other words, it all comes down to pace. How far teams can be reduced matters less than how fast the organization can absorb that reduction without putting either its teams or its business at risk.
