Navigating the Future with AI, Part 7. Welcome to this new episode of Navigating the Future with AI.
A fundamental question now faces every executive, enterprise architect and economic decision-maker: if generative AI and agentic systems multiply our individual productivity by two, three, five, or even ten on certain activities, should we completely rethink employment contracts historically tied to a fixed number of working hours?
These multipliers are not a uniform promise. They raise a question: when the gain is real, measured and sustained, who benefits?
From the stopwatch to velocity: the illusion of hours in white-collar work
To understand this shift, we need to return to the origins of our relationship with working time. In manufacturing, craft and mechanised production, the relationship between time on site and output is often more directly observable. It still depends on machinery, skills and the physical constraints of the activity.
Take a bakery. Producing bread and pastries follows a physical clock that is difficult to compress. In our example, kneading and baking begin at 5 a.m. to meet the first wave of customers at 6:30 or 7 a.m. The operating rhythm follows the synchronisation between production and the act of selling.
For white-collar workers, from public administration to software engineers, architects, designers and Product Managers, the equation is radically different. Although value is notoriously difficult to quantify through time spent alone, employment contracts remain largely organised around hourly reference points inherited from the industrial era.
These reference points have different legal meanings. France has a statutory reference week of 35 hours, while ordinary working time in Luxembourg is generally limited to 40 hours. In the United States, 40 hours is notably the federal overtime threshold for covered, nonexempt employees. A 42-hour contractual week is possible in Switzerland, but it is not a uniform statutory duration: contractual hours and legal maximums are distinct.
Behind a 40-hour working week, the enterprise is not buying time alone. It expects an implicit level of performance and velocity.
The anatomy of an IT working day and the relativity of performance
Look at the IT industry, now on the front line of generative AI and autonomous agents. A technology professional’s working day is a mosaic of different cognitive activities:
- Systemic analysis and the formalisation of business requirements;
- Writing code, configuration and testing;
- Continuous deployment through environments to the production release;
- Resolving complex incidents and providing third-level support for users’ business problems;
- Strategic alignment with management and cross-functional teams;
- Administration, including timesheets and project allocation reporting, alongside the breathing space needed for cognitive balance.
In this ecosystem, a manager should not assess someone through screen presence alone, but through relative performance and the value of the work delivered. That assessment depends on professional seniority, the onboarding needed to understand the company’s culture, business domain and existing codebase, and how the rest of the team works.
This movement from the technical act towards understanding, orchestration and judgement is precisely what I explore in Prompt Over Code, the second chapter of the series.
The empirical shock of AI: from measured gains to 10x ambitions
This reference point for velocity is being challenged. AI tools can accelerate certain tasks dramatically. But accelerating a task, increasing a team’s throughput and creating more economic value are three different measures.
- AI Engineering and the 10x ambition. Didier Girard, SFEIR’s co-CEO, champions an AI Only vision and a Software Factory 10x approach. That ambition changes how we design the software lifecycle. It is not, by itself, evidence that an entire working week becomes ten times more productive in every setting.
- GitHub Copilot. In a controlled experiment involving a programming task, the equipped group completed the task about 55% faster. The finding applies to the experimental task, not to everything a developer does.
- MIT. Noy and Zhang’s study of professional writing tasks found an average 40% reduction in completion time and an 18% improvement in quality. It was not a software development study.
- Google. On 29 October 2024, Sundar Pichai said that more than a quarter of Google’s new code was generated by AI, then reviewed and accepted by engineers. The share of generated code is not a direct measure of productivity gains or of code deployed to production.
- The 2025 DORA report. 90% of respondents reported using AI at work, and more than 80% believed it had increased their productivity. The report observes a positive relationship with delivery throughput, alongside risks to stability. The organisation of work remains decisive.
We must also examine results that challenge automatic acceleration. METR measured a 19% slowdown using early-2025 tools among experienced developers working on their own repositories. Its February 2026 update suggests improvement with newer tools, while highlighting selection effects that prevent a reliable, general estimate of the gain.
The point is therefore not to declare everyone 10x. It is to measure the net gain across a real scope of work, then decide how to allocate it. This extends the idea of Digital Augmentation: increasing our capacity to act, without confusing the production of content with the creation of value.
Three paths for rebuilding the enterprise’s social contract
If a 40-hour contractual baseline previously corresponded to a given volume of output, V, and an AI-augmented employee now delivers that same volume with double or triple the productivity, we face a major societal and managerial choice. Three ways of allocating this productivity dividend emerge. They can be combined.
1. Returning time to the worker and the architectural prerequisite
If we base the contract on value produced rather than presence, delivering the expected weekly workload faster should allow some time to return to the employee, at unchanged pay. That time becomes space for personal life, learning or civic engagement.
In the theoretical case where the entire workload falls from 40 to 20 hours, a 2x productivity gain allows it to be completed in 2.5 eight-hour days. A three-day or 3.5-day week represents a different balance, retaining more availability. Accelerating coding alone does not mechanically halve the entire working week.
This transition requires removing the synchronisation bottleneck. In IT and other service functions, a significant part of presence meets a need for synchronous availability: support, incident management or cross-functional alignment. For a shorter week to preserve collective velocity, enterprises must organise continuity through human handovers, multi-agent systems and living knowledge bases. The worker can then decouple from part of the continuous flow, within the limits of what can be delegated.
That continuity is a question of organisational architecture. It is also the subject of my white paper Enterprise Morphogenesis: keeping the enterprise coherent as it changes.
2. Strategic reallocation towards resilience and offensive growth
If the 40-hour framework remains, time saved on conventional delivery should not feed a blind race for volume. The enterprise can reinvest it in structural work too often sacrificed for lack of bandwidth:
- A more offensive approach to sales and innovation;
- Stronger cybersecurity and system safety;
- Higher operational resilience and regulatory compliance, in the spirit of frameworks such as the Digital Operational Resilience Act, applicable to the financial sector since January 2025;
- Excellence in Knowledge Management, living architecture documentation and Business Continuity planning.
The European DORA regulation and the DevOps DORA research programme are distinct references. In both cases, acceleration without the conditions for reliability exposes the enterprise.
This reinvestment extends the question raised in The $1 Trillion Agent Factory: what do we do with the additional power, beyond productivity alone?
3. Pay adjustment and the algorithmic performance bonus
If an employee maintains a five-day commitment while multiplying output through mastery of AI tools, is it not legitimate to reconsider compensation?
A naive approach would demand that salaries mechanically double. It would quickly encounter an economic ceiling: compensation depends in part on the value the enterprise can actually sell or preserve in its market.
However, when these gains translate into additional sales, improved customer satisfaction, avoided costs or a net return on investment, I believe a share of that value should return to the worker. First, we must deduct licensing, infrastructure, token, integration, verification and operating costs. AI is not free.
To preserve the enterprise’s financial balance against economic uncertainty and competition, this mechanism need not necessarily make fixed salaries more rigid. It could take the form of a dynamic hyper-performance dividend or bonus, linked to the periodic results of technological augmentation. Collective contribution and quality must count, not just individual volume.
Realism in hyper-competition: a decision about the architecture of society
Looking at these three trajectories through globalisation and competition within an industry, I consider a generalised move to 2.5-day or three-day weeks the least likely path in the short term. This is a forward-looking judgement, not an economic law.
An enterprise that unilaterally reduced its commitment without securing continuity and delivered value could be overtaken by international competitors retaining their hours while increasing their velocity through AI. In this global race, I see a combination of offensive reinvestment and adjusted financial compensation as more likely to prevail.
But let us not mistake this for a purely technical inevitability. Choosing between more free time and more economic power is fundamentally a societal choice. In the fullest sense, it is a decision about the architecture of society.
From the industrial revolution to autonomous robotics: avoiding displacement as the default
Economic history shows that many technological advantages eventually become an industry norm. As economies moved from horse-drawn transport to cars, railways and air freight, mechanisation spread. What differentiated one actor could become, decades later, a condition for remaining in the market.
Today, this movement extends beyond software. Alongside cognitive AI, multifunctional robotics is advancing in agriculture, logistics, drones and on-demand transport. Waymo and Baidu’s Apollo Go already provide examples of driverless robotaxi operations within defined service areas.
During this transition, AI creates legitimate anxiety. What I challenge is the reflex of investing in R&D and automation solely to reduce headcount. An enterprise exists to serve customers and reward shareholders, but also to help its employees and society prosper.
Our societies have already shown that they can reconsider the metric of working time. In France, the Aubry laws of 1998 and 2000 organised the move from 39 to 35 hours, taking effect in 2000 for enterprises with more than twenty employees and in 2002 for the others. Their stated purpose concerned job creation and reducing unemployment. RTT days are one mechanism for organising that reduction; they were not uniformly awarded to every employee.
Just as public debate revisits retirement age in response to demographic ageing, we need the political and managerial courage to recalculate the equation of time and value when AI reduces the effort required for certain activities.
Creating a positive-sum incentive
Rethinking employment contracts, through more free time or a financial share of AI’s ROI, is not only a matter of social justice. It is a major lever for technological adoption.
By offering a direct, tangible incentive to the employee who learns to use these tools, we align individual interests with collective performance. The worker can then see AI as something other than a threat of obsolescence or a demand for exhaustion: a catalyst for emancipation and meritocratic prosperity.
That is how we can build augmented, resilient enterprises and an economy in which every actor in the value chain can benefit.
This connects with The Human Moat in the AI Economy: the human role is also built through understanding systems, exercising judgement and developing the capacity to act.
