Over the past year and a half, we have started to observe early, concrete signs of AI-driven productivity gains in operational companies. These gains are not theoretical and not limited to prototypes or pilots. They are showing up in production environments, most visibly in support and coordination functions: planning, procurement, quality, supply chain, engineering support, finance operations, customer operations. In many industrial organizations, these functions represent up to half of total headcount.
Across deployments, productivity uplifts typically fall in a wide range, from roughly 30% in the most conservative cases to well above 70% in more advanced setups. Achieving those results is not easy. It requires data work, process redesign, careful tooling choices, and sustained operational effort. A playbook is slowly emerging, informed by what works and what does not, but that playbook is not the subject here.
What matters for this discussion is something more basic and more uncomfortable: despite very real productivity improvements, many organizations see little to no immediate impact on their P&L. This is not due to accounting quirks or measurement issues. It is structural, and it raises questions that are not purely technical. They are moral, cultural, and economic at the same time.
We are fortunate to work closely with a number of operational companies that approach these questions seriously and thoughtfully. The reflections below are drawn from that exposure. They are not meant as a guide or a recommendation, and certainly not as advice. They are observations from the field, shared with humility and with the awareness that many of these questions remain open.
Productivity gains do not automatically translate into economic gains
A recurring pattern is easy to describe and harder to act upon. AI-driven productivity gains, even when substantial, do not materially improve margins if headcount trajectories remain unchanged. When output per person increases but the number of people continues to grow along pre-existing plans, the economic effect is absorbed by organizational inertia. In some cases, additional capacity simply leads to more internal complexity, more projects, and more coordination costs.
This is not a failure of AI. It is a reflection of how organizations are built and how decisions about hiring, growth, and structure are made. For decades, growth in activity was tightly coupled with growth in headcount. Many planning processes still assume this coupling by default. AI breaks that assumption, but organizations do not automatically update their behavior.
The result is a growing tension between what technology makes possible and what organizations are prepared to do with it.
Stop pretending nothing’s happening : Employees are aware of what is happening
One of the most striking aspects of recent discussions is that employees across levels already understand that AI will affect their work. This is not limited to executives or technical roles. People on the shopfloor, in offices, and in shared service centers are well aware that parts of their tasks are being automated or augmented.
In that context, communication strategies that deny or minimize the impact of AI tend to be counterproductive. When leaders claim that nothing will change, employees often interpret this as either a lack of clarity or a lack of honesty. Trust erodes not because the reality is difficult, but because it is not addressed openly.
By contrast, the leaders who seem to navigate this period more effectively are those who acknowledge the reality directly. They state plainly that AI will reduce the amount of human effort required for certain tasks, that roles will change, and that not all current positions will exist in the same form in the future. Importantly, they pair that honesty with a concrete plan and a clear commitment to supporting people through the transition.
Managing the transition without immediate disruption
In practice, the most effective approaches we have observed rarely rely on abrupt headcount reductions. Instead, they focus on managing trajectories over time. This often starts with slowing down or rethinking hiring plans, especially in functions where AI-driven productivity gains are already visible. Future growth is accounted for, but so is the fact that fewer incremental hires may be needed to support that growth.
This approach requires significant investment in retraining and upskilling. As tasks change, people are asked to move away from repetitive execution toward roles that involve supervision, exception handling, analysis, and decision-making. In several organizations, this is combined with a cultural shift away from static career paths. Staying in the same role for long periods becomes less common, and internal mobility is encouraged.
The economic gains from AI may materialize more slowly under this model, but the organizational benefits are tangible. Employees tend to engage more actively when they feel that leadership is taking responsibility for the transition rather than deferring it or externalizing it. While this does not eliminate all tension or anxiety, it creates a basis for trust.
The nature of work changes as much as its volume
A frequent misunderstanding about AI-driven productivity is that it simply removes work. In reality, it changes the composition of work. While AI systems can handle a large share of routine processing and execution, a significant portion of human effort shifts toward higher-level tasks. These include controlling automated processes, interpreting outputs, resolving edge cases, and making judgment calls in situations where data is incomplete or ambiguous.
This shift places higher demands on job knowledge rather than lower ones. As individual decisions carry more weight, the consequences of errors increase. People are expected to understand their domain more deeply, not less. In that sense, AI raises the bar for human contribution rather than lowering it.
This also explains why retraining is not a superficial exercise. Teaching people how to use new tools is insufficient if it is not accompanied by a deeper understanding of the underlying processes and objectives. Organizations that underestimate this tend to struggle, not because the technology fails, but because the human layer is not adequately supported.
What seems to differentiate early winners
Based on what we have seen so far, the organizations that appear to be navigating AI-driven transformation more effectively share a few characteristics. They communicate with a level of honesty that may feel uncomfortable but proves stabilizing over time. They avoid overly optimistic narratives and instead focus on explaining trade-offs and constraints. They invest heavily in people, not only through training budgets but through time, attention, and managerial involvement.
Perhaps most importantly, they are willing to rethink parts of their organizational structure that were previously taken for granted. This includes questioning team sizes, role definitions, and career paths. These changes are not easy and often meet resistance, but postponing them tends to compound difficulties rather than resolve them.
An ongoing learning process
AI is already reshaping how operational organizations function, and this process is far from complete. The technical challenges are significant, but the organizational and human challenges are at least as important. Productivity gains will not automatically lead to economic gains unless organizations adapt how they plan, hire, train, and lead.
We are still early in this transition, and many assumptions will need to be revisited as experience accumulates. The reflections shared here are not conclusions, but snapshots from a moving landscape. We remain grateful to the founders, teams and leaders who are willing to engage openly with these questions and to let us learn alongside them, building and learning at pace.