In March of 2025, we wrote something that, at the time, felt like a stretch.
We argued that the traditional engineering model—the familiar “two-pizza team” of six to ten people working through structured sprints—was not going to survive the impact of AI. We suggested that velocity would no longer come from adding more engineers, more process, or more coordination. It would come from something fundamentally different: smaller teams, deeply integrated with AI, operating more like orchestrators than builders.
At the time, that idea sat somewhere between insight and speculation.
One year later, it is no longer speculative. It is documented, measured, and already reshaping how the best engineering organizations in the world operate.
The Shift Is No Longer Subtle
Across 2025 and into 2026, reporting from sources like McKinsey, The Pragmatic Engineer, Stack Overflow, and Fortune has converged on the same conclusion. The structure of software teams is changing because the constraints that shaped them have disappeared.
The traditional model assumed that software development was bottlenecked by human effort. Writing code, debugging, testing, documenting, and coordinating across teams all required time, and that time scaled poorly. The answer, for decades, was to add more people and distribute the work.
AI removed that assumption.
In multiple real-world examples cited in the report, teams that once required ten engineers are now being replaced by three people working alongside AI systems. In some cases, those three individuals are managing not just code, but entire ecosystems of AI agents—sometimes fifty to one hundred of them—handling everything from generation to testing to deployment.
At companies like Google, large teams responsible for a single capability are being broken down into much smaller units, not because of cost pressure, but because smaller groups move faster when AI eliminates coordination overhead. At Atlassian, engineers are already operating in environments where they write little to no code themselves, focusing instead on directing AI systems that produce the output.
This is not optimization. It is a structural collapse of the old model.
Why the 2025 Prediction Was Correct
The original prediction was not that AI would make engineers more productive. That would have been obvious and, frankly, uninteresting.
The real prediction was that AI would move the bottleneck.
For decades, software engineering was constrained by production. The limiting factor was how quickly humans could translate ideas into working systems. Every process, every methodology, every team structure was designed around that constraint.
What we argued in 2025 was that AI would remove that bottleneck and expose a new one: cognition.
That is exactly what has happened.
Today, the limiting factors are no longer how fast code can be written or tested. They are whether the system is designed correctly, whether the problem is understood deeply enough, and whether the solution reflects real human needs. The work has shifted from execution to judgment, from implementation to architecture, from activity to understanding.
The report reinforces this directly. High-performing organizations are not just adopting AI tools; they are redesigning their entire operating models around them. They are redefining workflows, changing incentives, and measuring success not by output volume but by business impact and quality. This shift is producing measurable gains, with productivity and quality improvements reported in the range of sixteen to forty-five percent.
But even those numbers miss the larger point. The real change is not linear. It is multiplicative.
Orders of Magnitude in a Single Year
When we say AI has improved productivity, it is tempting to think in percentages. Ten percent faster. Twenty percent more efficient. Even forty percent better.
That framing is already outdated.
What has happened over the last year is not a percentage increase. It is a change in scale. Tasks that previously required days of coordinated effort can now be completed in minutes. Entire layers of work—testing, documentation, scaffolding, refactoring—are handled automatically. The role of the engineer is no longer to produce each artifact, but to guide a system that produces them continuously.
The report captures this through examples of two to five times output increases, but the more important insight is how those gains compound.
When AI accelerates development, it also accelerates iteration. Faster iteration leads to better products. Better products lead to clearer direction. Clearer direction reduces wasted effort. The system improves itself.
This is why the impact feels like orders of magnitude rather than incremental improvement. The gains are not isolated; they reinforce each other.
The Emergence of the Two Sandwich Team
One of the most striking confirmations of the original prediction is the collapse of the two-pizza team itself.
The report describes a consistent pattern across high-performing organizations. Teams are shrinking, not growing. The new standard is not six to ten people, but three to four, and in some cases even fewer. These teams are not less capable. They are more capable, because AI fills the gaps that previously required additional specialists.
The term "two-sandwich team” has begun to emerge to describe this model. It is not just a smaller version of the old structure. It is fundamentally different. The individuals on these teams are not narrowly specialized. They are generalists with strong architectural thinking, capable of understanding systems end-to-end and directing AI to execute across layers.
This aligns exactly with what we described in 2025. The future was not about adding more people to solve problems. It was about increasing the leverage of each individual through intelligent systems.
What This Means for Engineering
The implications for engineering are profound, and they are arriving faster than most organizations are prepared for.
The skill set that mattered most a year ago is no longer sufficient. Writing clean code, while still valuable, is no longer the defining capability. The engineers who will thrive in this new environment are those who can think in systems, who can design architectures that incorporate AI as a core component, and who can make high-quality decisions in environments of uncertainty.
Equally important is something that has often been undervalued in technical fields: empathy.
As AI takes over execution, the human role shifts toward understanding the problem space more deeply. This is especially critical in domains like healthcare, cognitive support, and accessibility, where the difference between a functional system and a transformative one lies in how well it understands the user.
This is where the opportunity becomes extraordinary.
Why This Matters for Voice of Reason
At Voice of Reason, this transformation is not just an engineering story. It is a human story.
The systems being built are not simply optimizing workflows or improving developer productivity. They are enabling entirely new forms of support for individuals with dementia, traumatic brain injury, and learning disabilities. They are creating continuity where memory fails, guidance where confusion arises, and independence where it was previously impossible.
What is most striking is that this level of capability did not exist even one year ago.
The convergence of real-time speech processing, multi-model reasoning, memory systems, and retrieval architectures has created something new: AI that can participate meaningfully in human life, not just respond to prompts.
And this is only the beginning.
The Prediction Was Right — But Incomplete
Looking back, it is clear that the core prediction from March 2025 was correct. Team structures are changing. AI is redefining productivity. The role of the engineer is evolving exactly as anticipated.
But there is one thing we underestimated.
We underestimated the speed.
The transformation did not take five years. It did not take three. It happened in one.
Which leads to a more important realization.
If this is what one year looks like, then the real question is not whether the prediction was correct.
It is whether we are ready for what happens next.