AI Value Depends Less on Training Than on Work Redesign

Microsoft has put a useful correction into the market. Its account of its own AI transformation says broad licensing, tools and training did not by themselves produce transformation. In Microsoft’s telling, value came when teams simplified and redesigned end to end workflows, changed human agent decision boundaries, and relied on manager led adaptation around real work. This is evidence from one company, not the whole market. It still deserves attention because it challenges one of the most common enterprise responses to AI: treat the problem as a skills rollout.

Many organizations may be treating a work design problem as a training problem. If that diagnosis is wrong, more enablement content will not fix it. People can be licensed, trained and encouraged, yet still work inside processes that were designed for a different division of labor between humans and systems. In that setting, the limit is not awareness or basic proficiency. The limit is how work is designed, who decides, when issues escalate, and who is responsible for the result.

Microsoft’s evidence does not prove this is true everywhere. It is one company’s internal transformation evidence, and the transferability across industries, regulatory settings and company sizes remains unclear. Deloitte emphasizes organization design in realizing AI value, including clear roles, decision rights and accountability. SAP reports a readiness gap in Canadian businesses adopting agentic AI, with weaknesses in foundational data, processes and governance. Those sources do not prove enterprise wide redesign is already happening. They do support a simpler point: many firms may be less ready in their processes and governance than their training plans assume.

For enterprise learning, this shifts where L&D creates value. If an organization is adopting AI in meaningful workflows, L&D’s contribution may need to start earlier, with performance diagnosis and work analysis, not later with course production. The work to sort out is straightforward: what work is changing, which decisions move to the system, what evidence a person should review, which exceptions still need human judgment, and how managers help teams adapt in the flow of work. Training still matters, but as one instrument inside a broader capability design problem.

Speed without diagnosis scales waste.

There is a fair counterpoint. Microsoft has unusual scale, technical depth and management capacity. What worked there may be too demanding for smaller firms, or for organizations still trying to establish basic AI literacy. For some organizations, broad access and foundational training may still be the right first move because they have not yet identified where AI can materially improve work. Even then, the point depends on actual adoption. If leaders expect operational value from deployed AI use cases, they should not treat training completion as a proxy for transformation readiness.

For any priority AI use case, test whether the blockage is actually skill, or whether it sits in process friction, unclear authority, poor data, weak governance, or manager habits. Collect only enough evidence to change the decision. If the problem is work design, fund redesign. If the problem is judgment quality, clarify decision boundaries. If the problem is confidence or fluency, then training is the right answer. For organizations that adopt AI beyond experimentation, capability strategy should begin with work redesign and governance, with training attached to that design rather than standing in for it.