Enterprise AI Transformation Programmes (2025–2026)

Narrative

The single most contested data point of the period is MIT Project NANDA's "GenAI Divide" report, which found that about 5% of AI pilot programs achieve rapid revenue acceleration while the vast majority stall, delivering little to no measurable impact on P&L. The report's own methodology, based on 150 interviews with leaders, a survey of 350 employees, and an analysis of 300 public AI deployments, drew sustained scrutiny. Marketing AI Institute's Paul Roetzer argued the study used a narrow six-month, P&L-only definition of success and warned "please don't put any weight into this study," calling it not statistically valid, while other critics noted the "zero return" finding rested on just 52 interviews the report itself called only directionally accurate. Despite the pushback, the report's secondary findings proved durable and were echoed across outlets: purchased vendor tools succeeded roughly 67% of the time, while internal builds succeeded only one-third as often, and only 40% of companies have official LLM subscriptions, while 90% of workers surveyed reported daily use of personal AI tools for job tasks, a shadow-AI pattern independently confirmed elsewhere.

McKinsey's November 2025 State of AI survey, fielded across nearly 2,000 respondents in 105 countries, offers the more methodologically robust cross-industry ledger. It found 88 percent report regular AI use in at least one function, yet nearly two-thirds of respondents say their organizations have not yet begun scaling AI across the enterprise, and just 39 percent report EBIT impact at the enterprise level. On agentic AI specifically, McKinsey found twenty-three percent of respondents report their organizations are scaling an agentic AI system somewhere in their enterprises, and an additional 39 percent say they have begun experimenting, with adoption concentrated in IT and knowledge management, where agentic use cases such as service-desk management in IT and deep research in knowledge management have quickly develop[ed]. High performers, a group representing roughly 6% of respondents with 5%+ EBIT impact, distinguish themselves by workflow redesign: they are 2.8x more likely to report fundamental workflow redesign (55% vs 20% of others) and far more likely to have defined human-in-the-loop validation, at 65% vs 23%.

DORA's 2025 State of AI-assisted Software Development report, drawing on insights from over 100 hours of qualitative data and survey responses from nearly 5,000 technology professionals from around the world, reframes the debate around organisational systems rather than tool capability. Its central finding is that AI doesn't fix a team; it amplifies what's already there. Strong teams use AI to become even better and more efficient. Struggling teams will find that AI only highlights and intensifies their existing problems. AI use has become near-universal among developers, with AI adoption near-universal: 90% of survey respondents report using AI at work, more than 80% believe it has increased their productivity, however skepticism remains as 30% report little or no trust in the code generated by AI. DORA's seven-capability "AI Capabilities Model" identifies platform engineering, data ecosystem health, and mature version control and code review as preconditions rather than AI-specific novelties, echoing ThoughtWorks' Technology Radar, whose 2026 edition explicitly returned to "pair programming to zero trust architecture, and from mutation testing to DORA metrics" as necessary counterweights to AI-driven complexity, alongside newer agent-specific concerns like the "lethal trifecta" of private data, untrusted content and external action in unsafe agent design.

On adoption strategy, the "token-maxing" thesis has a concrete named exponent: Nvidia CEO Jensen Huang, who argued that if a $500,000 engineer did not consume at least $250,000 in tokens, he would be deeply alarmed, a philosophy practitioner commentary calls dangerous for mid-market firms where the average enterprise is burning through 13 times more AI tokens today than they were just one year ago without corresponding governance. Deloitte's Tech Value survey corroborates the underlying budget pressure: AI is now the fastest-growing expense in corporate technology budgets, with some firms reporting that it consumes up to half of their IT spend. Cloud computing bills are rising sharply - up 19% in 2025, even as the unit price of AI tokens is falling, overall enterprise spending on and scaling of AI systems is rising due to elastic demand from longer context, agent loops and reasoning chains. AT&T is cited as a concrete case, reportedly scaling from roughly 8 billion to 27 billion tokens per day after deploying multi-agent systems. The FinOps Foundation identified this as its top practitioner challenge, noting the FinOps Foundation's practitioner survey identified managing the cost and use of tokens in SaaS-model AI as the top challenge facing practitioners today, with root causes described as developer-led purchasing, opaque billing, no native allocation mechanisms, and pricing models that vary dramatically across model tiers and use cases.

On governance architecture, ISO 42001 and NIST AI RMF are converging into a layered stack rather than competing alternatives, with practitioner guidance suggesting organisations start with ISO 42001's management system structure, use NIST AI RMF's functions for risk management methodology, and layer EU AI Act's prescriptive obligations for high-risk systems. Notably, none of the three frameworks was designed for agentic AI, leaving a documented gap for autonomous systems that cascade failures across tool calls. The EU AI Act's enforcement timeline itself became a live variable in 2026: a May 2026 political agreement under the "Digital Omnibus" pushed high-risk rules would apply from December 2, 2027 for stand-alone high-risk AI systems and from August 2, 2028 for product-embedded high-risk AI systems, later than originally planned, while penalties remain severe at up to €35 million or 7% of global annual turnover for prohibited AI practices. MIT Sloan Management Review's June 2026 "Scaling AI With Adaptive Governance," based on interviews conducted from 2022 to 2025 with senior leaders and practitioners responsible for AI governance, risk, compliance, data, and product decisions...at Microsoft, Barclays, Kyriba, Nasdaq, Lloyds Bank, Danske Bank, and the Abu Dhabi Department of Finance, argues that ad hoc governance is inadequate at scale and proposes matching control intensity to system risk type rather than applying uniform rules. Retool's 2026 survey of 307 senior technology and security leaders found governance confidence has not kept pace with shadow AI proliferation: only 5% of senior tech and security leaders are very confident they have full visibility into what's running in their own production environments. 43% are not confident, even as 75% of builders now work under AI directives (up from 66% in October 2025), capturing the top-down mandate versus bottom-up sprawl tension directly.