AI adoption will fail if leaders redesign training before they redesign work

Before you fund enterprise AI training at scale, ask a harder question: what work has actually changed, and who now carries the judgment, review and intervention burden? My view is bounded but firm. When AI changes tasks, workflows and collaboration patterns, the gating decision is work redesign, not learning rollout. If leaders reverse that sequence, they can create visible training activity while leaving productivity, manager load, accountability and risk in the old design.

Deloitte is looking at changes to the AI operating model. Set that beside evidence that AI is already reshaping software development work, including user roles, workflows and collaboration patterns, and the executive issue is easier to see. The point is not only that people need new skills. It is that the work may now need different structures and combinations of human and AI contribution. Capability investment is unlikely to land well if the operating model still assumes the pre-AI version of the job.

That is the mistake I would avoid. Do not ask the CLO or L&D function to move first with broad literacy, role academies or manager toolkits before the business has specified where decisions moved, where handoffs changed and where review work increased. In most companies, that is owned by the COO, business unit leader or transformation lead, with CHRO and CLO translating the answer into talent, capability and support.

Otherwise learning teams are given a moving target and judged on completions rather than whether throughput, decision quality or readiness improved.

I would start with a short operating diagnostic. Pick the workflows where AI is being inserted. Identify where a person is now expected to challenge or override an AI-supported output. Then look at whether that person has the competence, authority, time and practice to do it reliably.

If not, the problem is not mainly training volume. It is a design gap in work, governance or capacity. Many organizations create human oversight on paper and extra managerial burden in practice.

The evidence does not prove that every sector needs structural redesign, or that redesigned models will produce better outcomes. Some organizations may see only modest role change, especially where AI use is narrow or operating discipline is already strong. Deloitte's work supports an operating hypothesis, not market-wide proof of adoption or ROI.

Even with those limits, I would treat large-scale capability funding as premature until leaders can name the changed task, the shifted decision right and the new oversight burden. I would change that view if stronger cross-enterprise evidence showed that training-first approaches improve performance, readiness or risk control without explicit changes to role design, workflow or governance. Short of that, I would assume the bigger failure mode is not a lack of AI courses. It is teaching people for jobs the organization has not yet redesigned.