2025 was the year of the copilot; 2026 is shaping up to be the year of the agent. Gartner projects that by 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024. In our work with more than 300 enterprises, we are seeing agent pilots move from procurement and IT into finance, HR, customer operations, and software delivery within a single planning cycle.
Agentic AI differs from the automation that preceded it in a fundamental way. Traditional robotic process automation executes a fixed workflow, while an agent pursues a goal, selects tools, breaks work into steps, and self-corrects when a plan fails. That shift from deterministic process to probabilistic, goal-directed behavior changes what a role actually contains: execution tasks become delegation and oversight tasks, and failure handling becomes a genuinely new job.
The L&D implication is not simply 'more AI training.' It is role decomposition and re-aggregation. When a workflow is decomposed, the constituent skills scatter across humans and machines, and the human remnant is more cognitively demanding than the original whole. Our client pilots show that the highest-value human work left behind is exception handling, ethical judgment, stakeholder communication, and quality assurance of agent output.
New roles are emerging in parallel. Agent supervisors, prompt and workflow engineers, human-in-the-loop controllers, and AI output auditors now appear in enterprise job architectures. We estimate that organizations introducing agentic workflows will need to re-skill 15–25% of affected roles within 18 months, which is why readiness assessment has to begin before deployment, not after it.
For L&D leaders, the practical question is sequencing: assess the agent surface area, decompose affected roles, define the oversight operating model, and only then build learning journeys. Our agent-readiness framework helps enterprises map which capabilities are augmented, which are replaced, and which are newly created, so that scarce training budget goes to the judgment work that agents cannot do.