Three years after generative AI tools entered the mainstream, AI is now near-universal in the enterprise — but enterprise-level value remains rare. McKinsey's State of AI 2025 survey of 1,993 respondents in 105 countries reveals that 88% of organisations now use AI in at least one function, yet only 6% are realising meaningful EBIT impact. The dividing line between AI leaders and laggards is no longer access to technology — it is workflow redesign, agent-readiness, and operating-model discipline.
Adoption is universal. Value is concentrated.
The headline numbers from McKinsey's November 2025 State of AI report are striking. Eighty-eight per cent of organisations regularly use AI in at least one business function, up from 78% in 2024 and 55% in 2023. Use of generative AI specifically reached 72% in 2025, more than doubling from 33% the year before1.
Yet the economic picture is more sobering. Only 39% of respondents report any measurable EBIT impact from AI, and most of those say AI accounts for less than 5% of EBIT. Nearly two-thirds of organisations have not yet begun scaling AI across the enterprise1. McKinsey defines a small group of "AI high performers" — approximately 6% of respondents — whose organisations attribute more than 5% of EBIT to AI and report significant overall value.
The agent shift is real — but uneven
The most significant change in 2026 is the emergence of agentic AI: autonomous systems built on foundation models that plan and execute multi-step workflows, going beyond simple question-answering. McKinsey found that 62% of organisations are at least experimenting with AI agents, and 23% report scaling agents in at least one business function1.
Gartner placed this transition at the centre of its 2025 strategic technology trend forecast. The firm projects that 40% of enterprise applications will integrate task-specific AI agents by the end of 2026, up from less than 5% in 20252. Looking further out, Gartner forecasts that approximately 33% of enterprise software applications will include agentic AI by 2028, up from less than 1% in 20243.
What separates the high performers
The McKinsey data reveals a consistent pattern. The advantage of AI high performers is overwhelmingly organisational, not technological. They are not using better models. They are using the same models differently.
Three behaviours distinguish them most clearly:
- End-to-end workflow redesign. McKinsey reports that the single strongest correlation with EBIT impact is fundamental workflow redesign — not piecemeal AI adoption layered onto existing processes1.
- Growth ambition, not just cost reduction. Eighty per cent of respondents say their organisations set efficiency as an objective for AI. The companies seeing the most value also set growth and innovation targets1.
- Scale across functions. High performers deploy AI in more business functions and are 3× more likely to be scaling AI agents across the organisation1.
"The question for 2026 isn't 'Are you using AI?' It's 'Are you redesigning your business around it?'"
Where AI is paying off — at the function level
While enterprise-level EBIT impact remains modest, function-level returns are substantial. McKinsey's 2025 data shows software engineering and IT functions reporting 10–20% cost reductions tied to AI-powered code generation, automated testing, incident resolution, and infrastructure optimisation. Marketing and sales, strategy and corporate finance, and product and service development show the strongest revenue uplift, often above 10%1.
Gartner separately predicts that AI agents could drive approximately 30% of enterprise application software revenue by 2035, surpassing $450 billion, up from 2% in 20252.
The risks the data exposes
The same surveys that confirm AI's mainstream adoption also surface the risks of moving too fast. McKinsey reports that approximately half of firms surveyed have experienced AI-related incidents1. Gartner has separately predicted that more than 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls5.
The most common failure modes are predictable: pilots launched without an end-to-end workflow strategy, agents deployed without governance or human-in-the-loop checkpoints, and a lack of measurement infrastructure that ties AI initiatives to business KPIs.
What CTOs should do now
- Pick one workflow and rebuild it end-to-end. Don't sprinkle AI across twenty processes. Choose your highest-value workflow and redesign it with AI at the core.
- Build the agent-ready stack. Invest in retrieval-augmented generation (RAG), tool-calling infrastructure, policy guardrails, and audit trails.
- Govern for safety and speed. Establish an approvals matrix by risk tier, pre-approve tools and datasets, log prompts and outputs, and define rollback procedures.
- Measure leading and lagging indicators. Track adoption, quality, and business results. Tie use cases to a financial model from day one.
- Set growth ambition, not just cost targets. The data is clear that organisations setting growth and innovation objectives are the ones seeing transformational impact.
The bottom line
Three years into the generative AI era, the gap between AI leaders and laggards is no longer about whether AI is being used. It is about whether the operating model has been rebuilt to capture its value. The 6% of organisations realising meaningful EBIT impact from AI did not get there by deploying better models. They got there by rebuilding their workflows, setting bold ambitions, and treating AI as an organisational transformation programme.