AI is no longer a passive tool for automation — it is an active agent that is reshaping cognition, task allocation and collective dynamics.
Summary of the Wavestone · ANDRH · French Tech Grand Paris research on changes in the world of work
For over 40 years, Human Design Group has been supporting the most demanding organisations — in the aerospace, defence, energy and space sectors — with the integration of complex systems in critical environments. The advent of generative AI and autonomous agents does not alter our fundamental principles: it makes them more essential than ever.
Research by Acemoglu et al. (2026) shows that agentic AI, by providing directly usable solutions, risks stifling the production of general knowledge — that informational public good which is essential to organisational resilience. Critical competence is no longer limited to technical expertise: It is the ability to interact and cooperate with the machine.
Any integration of AI into a work system raises a fundamental question: What can the algorithm do, and what can only a human do? This boundary is not set in stone — it depends on the context, the level of expertise, the nature of the decisions and the associated operational risk.
Our IFH approach maps out this allocation in detail based on an analysis of actual activity. It reveals areas of invisible friction — where AI generates a new form of chrono-ergonomic dissonance: whilst freeing up execution time, it paradoxically imposes cognitive strain due to the asymmetry in rhythm between algorithmic time and human biological time.
Critical competence is no longer simply technical mastery, but interaction and cooperation with machines. — Wavestone · ANDRH · FT Grand Paris
What the analysis reveals:
Junior staff may confuse speed of execution with genuine mastery (the ‘Illusion of Competence’). Senior staff, on the other hand, must make the transition from technical experts to discerning practitioners. Our IFH assessment enables us to objectively analyse these dynamics and steer skill development pathways in an AI context.
As AI systems become increasingly autonomous — evolving from simple tools to assistants, and then to autonomous agents — the role of humans shifts from that of a creator to that of a supervisor. This shift requires a radically different approach to the design of interfaces, protocols and decision-making loops. Without a properly designed human-in-the-loop system, supervision becomes a mere illusion of control.
The study by Arshid et al. (2026) is unequivocal: 56 % of the AI deployments analysed breach the models’ terms of use or the applicable EU AI Act regulations. 25 out of 45 projects operate in high-risk sectors — healthcare, finance, autonomous driving — without the required human supervision. The risk is not hypothetical.
Our supervision protocol
Human Design Group has developed a unique protocol for assessing the performance of human-system collaboration, tested under real operational conditions (Ministry of the Armed Forces, France 2030). This protocol measures the efficiency of the overall system — not the power of its individual components — by focusing the assessment on the operator within their designated objectives, including in degraded situations, breakdowns and errors beyond nominal operating conditions.
Transparency regarding the use of algorithms is the ultimate user experience. Maintaining ultimate human control — particularly when it comes to mobility and assessment — is not a regulatory constraint: it is the foundation of sustainable performance.
Human-centred deployment triptych — Wavestone / ANDRH / French Tech Grand Paris
The real work is invisible to AI. Beneath the surface — the standardised deliverables, the measurable algorithmic traces — lies a hidden mass: social work, mutual support, informal coordination, the emotional burden. It is this invisible work that keeps the system going. The role of the manager and the designer is to act as a mediator of meaning in order to recognise the value of these contributions, which AI cannot see, assess or replace.
Our systemic assessment is based on the IFH’s distinctive holistic approach: it zooms in and out between the micro (individual), meso (team, processes) and macro (organisation, governance) levels. It incorporates factors relating to collective cohesion, intergenerational transmission and digital inclusion — because equitable access to AI is a prerequisite for legitimacy and collective performance.
The systemic risks your dashboard overlooks
Knowledge Collapse (the decline in general knowledge due to excessive reliance on AI), organisational cognitive debt, the fragmentation of informal connections through algorithmic intermediation, internal divides between tech and non-tech roles, and a ‘me-too’ bias in deliverables that homogenises results at the expense of the value of expertise.
Three examples of how our IFH approach has been applied in real-world complex environments. Results that have, at times, surprised even those who were sceptical.
Assessment of the suitability and methods for integrating AI into a combat cockpit. Over a period of five days, various forms of AI were simulated in a VR immersion scenario with pilots fitted with biometric sensors. The «Wizard of Oz» method was used to test different types of interaction and objectively measure performance gains using neuropsychological indicators.
«Over the course of 10 days, the pilots looked ahead to 10 years of AI integration.»
Integration of AI coupled with a simultaneous review of processes and organisational structures. Redesign of information flows, hierarchical presentation of content, and optimisation of interactions between operators. Direct impact on cognitive load, working conditions and the HR function — which is freed up to focus on career management.
Performance that was «more than multiplied» — results so significant that external organisations found them hard to believe.
Competition organised by the Ministry of the Armed Forces. Development of a single protocol for assessing the performance of human-system collaboration, focusing on overall efficiency rather than the power of individual components. Evaluation of multi-system orchestration (sensors, AI agents, distributed subsystems) with humans in the loop, in both nominal and degraded conditions.
A unique evaluation protocol — a benchmark for task allocation in complex autonomous systems.
These use cases contributed to the ‘AI Convergence (France 2030)’ white paper presented at the AI Summit 2025, organised by Bunka.ai.
Our methodology is based on a well-established body of scientific research and 40 years’ experience in the most demanding sectors. It is the only approach that enables an AI system to be assessed on the basis of its actual operational value — rather than its theoretical promises.
01. Analysis of actual activity
02. Cognitive and systemic assessment
03. Designing collaboration
04. Assessment of overall performance
