The rise of AI agents is disrupting business
Agentic AI—systems capable of autonomous planning, action, memorization, and learning—is now advancing at a pace that exceeds the adoption rate of previous technologies. Their potential is emerging at a time when many leaders are going through a "wildfire": the first use cases for text or code generation have sometimes disappointed, but four approaches are already emerging to break the rut: rethink ambition, act with urgency, prepare for scale from the start, and make every employee an agent supervisor.
This acceleration is based on a spectacular dynamic of innovation: the number of cutting-edge language models has jumped, as have the size and complexity of the tasks that agents successfully complete, which double approximately every seven months. At the same time, inference costs are falling: for a large model from 2023 to 2024, they have been divided by ten, paving the way for intensive uses that were previously impossible; the cost per million tokens has been reduced from $36 to less than $4. The economics are changing scale: a large financial institution entrusted the modernization of its application portfolio to an "agent factory"; one hundred agents, supervised by five humans, halved the time and expense of rewriting code.
To understand these upheavals, we must distinguish three levels. At the first, "agentic labor," agents support an individual: drafting a memo, taking meeting minutes, proposing contractual clauses. Individual productivity gains climb to 20-30%, but the aggregate value remains limited if the company does not change anything in the process; adoption stagnates due to a lack of support.
The second level automates existing tasks or flows. In a contact center, for example, 60 to 80% of simple requests can already be handled end-to-end by an agent, simultaneously improving speed and customer satisfaction. However, these islands remain isolated; without global orchestration, human bottlenecks persist.
The third level—the true "agentic engine"—reinvents a complete, sometimes cross-functional, journey. Teams of agents then manage financial planning or lead-to-order, eliminating manual transfers and reducing the unit cost of certain processes by 70 to 80%. At this stage, the challenge is no longer technological but organizational: responsibilities, governance, and culture must be redesigned.
To industrialize at scale, most pioneers set up a dedicated "factory": a central team that designs, tests, and deploys agents while producing libraries of blueprints, metrics, and safeguards. Its mission is also to prevent the uncontrolled proliferation of agents and ensure their regulatory compliance. Another pillar, an observability platform, manages the lifecycle: costs, performance, access rights, and security must be finely controlled.
Operationally, an aggressive roadmap is required. The first year aims to break inertia: massive deployment of personal agents, automation of key processes such as report writing or purchase validation, and launch of a "flagship" that reinvents an entire flow (e.g., order-to-cash) with a target of > 70% automation. The goal is as much to learn as to save up to 10% of costs in the targeted areas.
The second year shifts the focus to P&L impact. An initial customer journey reaches 90% automation, nearly all critical value chains shift to agents, and legacy business models—built on linear handoffs between departments—begin to disappear. At least 75% of employees are expected to use agents daily, with some orchestrating dozens.
This transition is profoundly disrupting employment. By 2030, up to 30% of current working hours could be automated. Companies must therefore invent a hybrid human-agent model where new roles are promoted: "agent orchestrator" and "agent trainer," with KPIs focused on the ability to manage a portfolio of agents and the quality of the outputs produced. Balance sheets and budgets will need to reflect the shift from labor spending to technological capabilities.
Beyond technology, trust is a prerequisite for adoption. Agents must explain their decisions, point out their limitations, and adhere to explicit ethical frameworks. Algorithmic transparency and bias governance are no longer luxuries; without them, team engagement plateaus, and regulators intervene. This is also why data foundations deserve to be consolidated at the very beginning.
Conceived as a two-speed movement, the transformation combines rapid experimentation with structural projects: while field teams test concrete uses, senior management drives a broader agenda that redefines the value proposition, talent, technology, and culture. Otherwise, the proliferation of autonomous agents could lead to a "digital jungle" where duplication, cost overruns, and security risks are mixed.
The time pressure is real: standing still means ceding control to bolder actors. Now is the time to set a clear course, align roadmaps and funding, and personally embody change. For those still hesitant in the face of uncertainty, remember that articulating an ambitious vision, investing in AI-ready leaders, and governing change at a sustained pace are now key leadership responsibilities.
Ultimately, agentic AI marks a turning point comparable to the arrival of the Internet or the cloud, but its impact on the very structure of organizations will be more profound: redesigned flows, flattened hierarchies, reinvented professions. Seizing this wave means moving quickly from experimentation to industrialization, without sacrificing technological rigor or human engagement. Those who succeed in harmonizing agents and talent will not only do things better and more cheaply; above all, they will be able to imagine products, services, and business models inaccessible to companies still in the pre-agentic paradigm.
To go further: how unlocking business value through agents, how prioritize use cases for agentic AI et innovate without restraint thanks to the agentic model.