--- title: "AI Transformation" date: 2026-07-12 description: Most organisations are stuck at the pilot stage; crossing the divide is a Board judgement about delegation, architecture, and governance, not a technology purchase. author: Mario Thomas canonical: https://mariothomas.com/briefings/ai-transformation/ --- ## Start here Two short reads before you go deeper: what the 'GenAI Divide' actually is, and the order in which to cross it. ### The Divide Is Governance, Not Technology MIT's 2025 research put a number on something I had been arguing since 2024: more than **80%** of organisations have explored or piloted generative AI, and only **5%** have taken it into production. AI transformation is the work of crossing that divide, and the first thing to understand about it is that the divide is not technical. The organisations on the far side are not running better models than everyone else. They have built the governance, the architecture, and the culture that let pilots become operations. Most of the Boards I meet get this wrong in a consistent way: they treat AI as a technology decision, something to be evaluated, procured, and delegated to the CIO. The evidence points the other way. What separates the 5% is coherent, Board-level governance: coordinated investment rather than departmental enthusiasm, systems that learn rather than tools that sit idle, and workflows redesigned around the capability rather than bolted onto it. Agentic AI sharpens the question rather than changing it. Strip the hype away and an agent is generative AI in a loop, with the machine driving the iteration a human used to drive. That makes the real decision one of delegation: where should decision-making authority transfer from people to systems, at what scale, and under what controls? A Board that frames agentic AI as a purchase is likely to struggle. A Board that frames it as conscious delegation, governed by just enough structure to make it safe, is asking the right question. These articles take a more contrarian position on legacy technology. Technical debt is the reason many Boards are given for deferring agentic AI; I argue it is the opportunity, because the same agents constrained by legacy architecture are uniquely capable of retiring it. And timing matters: consumer adoption of always-on agents is running the same pattern ChatGPT ran in 2022, which means shadow agentic AI is arriving inside organisations whether they have governance for it or not. By the end of this briefing you should be able to make one judgement with confidence: where in your organisation delegating the loop to machines would create advantage, and what has to be true of your governance and architecture before you allow it. ### From the Divide to the Migration Take [the articles](#core-reading) in the core order; the sequence is deliberate. Start with the 'GenAI Divide' piece, which sets out the evidence for why 95% of pilots stall and what the successful 5% do differently. It gives you the diagnostic frame that everything else in the briefing builds on. The agentic pieces come next and belong together. The explainer strips agentic AI back to what it actually is, generative AI in a loop, and puts the delegation question in usable form. The compound loop piece then widens the lens: the loop can run any AI discipline, not just generative, and the value multiplies when it coordinates several at once. The core sequence closes with the ground under your feet and the road ahead. The technical debt article makes the case that legacy architecture is both the binding constraint on agentic AI and the thing agents can systematically retire. The inference migration piece reads consumer agent economics as the leading indicator for enterprise demand, and explains why the window to put governance in place comes before shadow agentic adoption arrives, not after. The wider reading deepens whichever front matters most to you: the personal agent piece for where the shadow agentic thread leads once people own better agents than their employer provides and knowledge ownership inverts, world models for what follows agentic AI, the culture piece for why capability without engagement stalls, the AI Centre of Excellence piece for how pilots are scaled once they work, the origin piece for the AI Stages of Adoption model as a way to read your own readiness, and the 2024 generative AI explainer if you want the foundations first. [Remake](#remake-assets) holds the mechanisms that make this operational, and [the questions](#faqs) are the ones I would take to your next Board meeting. ## Core reading The core sequence runs from the evidence on why pilots stall, through what agentic AI actually is and what it can retire, to the always-on wave now forming. 1. [Crossing the GenAI Divide: Solving The 95% Problem With The Complete AI Framework](https://mariothomas.com/blog/crossing-genai-divide-boardroom/) (12 minute read, 28 August 2025): MIT confirms what I have argued since 2024: 5% of organisations take generative AI from pilot to production. The Complete AI Framework answers that divide. 2. [Agentic AI: Strip Away the Hype and Understand the Real Strategic Choice](https://mariothomas.com/blog/agentic-ai-explainer/) (17 minute read, 2 November 2025): Agentic AI is this year's poster child, and most of the confusion is about what agents actually do. The Board's decision is strategic, not technical. 3. [The Compound Loop: Why Agentic AI's Real Power Lies Beyond Generative AI](https://mariothomas.com/blog/agentic-ai-compound-loop/) (9 minute read, 9 November 2025): Agentic AI is more than generative AI in a loop. Compound loops coordinating several AI disciplines are where returns stop being linear. 4. [How Agentic AI Turns Your Biggest Tech Problem into Competitive Advantage](https://mariothomas.com/blog/agentic-ai-technical-debt/) (11 minute read, 3 August 2025): The legacy estate that constrains agentic AI is also its largest opportunity. Retiring technical debt is what clears the path for autonomous systems. 5. [The Inference Migration: What Consumer Agents Mean for Enterprise AI's Next Phase](https://mariothomas.com/blog/inference-migration/) (12 minute read, 8 February 2026): Consumers are already paying for always-on AI agents. That consumer-to-enterprise pipeline is the one ChatGPT ran, and it is running again. Podcast edition: 17 minute listen. ## Further reading - [World Models: The Next Horizon in AI for Predictive Enterprise Intelligence](https://mariothomas.com/blog/world-model-ai-next-horizon/) (10 minute read, 23 November 2025): World models move AI from reacting to anticipating: systems that simulate future scenarios, with aviation and finance already seeing the operational gains. Podcast edition: 15 minute listen. - [Europe's AI challenge: Why culture trumps capital in technology adoption](https://mariothomas.com/blog/europe-ai-challenge/) (7 minute read, 13 December 2024): Europe's lag in AI adoption is not a shortage of capital. Seen from San Francisco, the difference is culture: openness to transformation, not money. - [The executive's guide to generative AI](https://mariothomas.com/blog/executives-guide-to-generative-ai/) (10 minute read, 8 April 2024): From my conversation with Duncan Jefferies for the AWS executive's guide: how business leaders can put generative AI to work on organisational change. - [AI Centre of Excellence: Scaling Beyond Pilots to Enterprise Transformation](https://mariothomas.com/blog/ai-coe-pilot-production/) (12 minute read, 27 July 2025): Successful pilots mask a harder problem: scaling them into enterprise-wide transformation. After the first 90 days, the AI Centre of Excellence's real test begins. - [Introducing the AI Stages of Adoption: A framework for understanding AI readiness in your business](https://mariothomas.com/blog/ai-stages-of-adoption/) (15 minute read, 13 June 2024): The AI Stages of Adoption, introduced for private equity firms and their portfolio companies: a five-stage framework for reading AI readiness and value creation. - [The Personal Agent Economy: When Your Best AI Isn't On Your Balance Sheet](https://mariothomas.com/blog/personal-agent-economy/) (8 minute read, 15 February 2026): Individuals now own AI agents that encode their judgement and expertise, capability that belongs to them rather than their employer. The assumption has inverted. Podcast edition: 10 minute listen. ## Remake The mechanisms beneath the thinking: the model, diagnostic, methodology, and principle from the Remake Library that turn this briefing into apparatus a Board can use. - **Model: AI Stages of Adoption**. A framework describing five stages of the AI journey, Experimenting, Adopting, Optimising, Transforming and Scaling, plotted on an investment-value graph, recognising that different functions progress simultaneously at different paces. [Remake Library](https://mariothomas.com/remake/library/ai-stages-of-adoption/) - **Diagnostic: Five Pillars of AI Capability**. The five capability domains of an AI capability model that cut across every level of maturity: Governance and Accountability, Technical Infrastructure, Operational Excellence, Value Realisation and Lifecycle Management, and People, Culture and Adoption. [Remake Library](https://mariothomas.com/remake/library/five-pillars/) - **Methodology: CoE 90-Day Sprint**. The time-boxed execution method that launches an AI Centre of Excellence's first pilot portfolio: initiatives scored and selected with the AI Initiative Rubric, run as a ninety-day sprint portfolio that balances quick wins against strategic bets, with governance mechanisms evolving alongside the work rather than ahead of it. [Remake Library](https://mariothomas.com/remake/library/#coe-90-day-sprint) - **Principle: Minimum Lovable Governance**. Governance embedded in how work happens: proportionate to risk, continuous rather than episodic, and used because it works. [Remake Library](https://mariothomas.com/remake/library/minimum-lovable-governance/) ## Questions The questions directors put to me most often on this subject, answered from the work in the briefing. ### Why do most of our generative AI pilots never reach production? Because pilots test technology while production tests the organisation. MIT's 2025 research found only **5%** of organisations achieve production implementation, and the difference is not model quality but coherent governance: coordinated investment, workflow integration, and Board-level accountability. I set out the evidence and the systematic response in [Crossing the GenAI Divide](/blog/crossing-genai-divide-boardroom/). ### Is agentic AI a structural shift or another hype cycle? Both, which is why precision matters. An agent is generative AI in a loop, with the machine driving iteration instead of a human, so the technology is familiar but the delegation of decision-making authority is new. The strategic question is not whether the architecture impresses but where transferring agency creates advantage, and [my explainer](/blog/agentic-ai-explainer/) and [the compound loop](/blog/agentic-ai-compound-loop/) set out how to answer it. ### Does our technical debt rule agentic AI out? It constrains it, but it is also the opportunity. Agents deployed on legacy architecture operate at a fraction of their potential, yet the same capabilities that make them struggle, understanding complex systems and implementing change autonomously, make them uniquely suited to retiring the debt itself. McKinsey's 2023 analysis estimates technical debt consumes around **40%** of IT budgets in large enterprises; [I argue](/blog/agentic-ai-technical-debt/) that agent-led modernisation converts that cost into the foundation for whatever comes next. ### Can we wait for the market to mature before acting? The ChatGPT precedent suggests not. Consumer adoption led enterprise demand by roughly three years, with shadow usage bridging the gap, and always-on consumer agents are now running the same cycle. Employees are likely to connect personal agents to work systems before formal platforms arrive, so the choice is between establishing proportionate governance now or retrofitting controls onto adoption already underway. I trace the timeline in [The Inference Migration](/blog/inference-migration/). ### We have the capital and the capability. Why is adoption still slow? Culture, more often than not. Gallup's 2024 benchmark of major European companies found that adoption lags stem from cultural barriers rather than financial constraints, with only **13%** of Europe's workforce engaged. Transformation is a continuous state rather than a project, a discipline I argued for [during the pandemic](/blog/reinventing-without-a-road-map-foot-on-gas/) and [in the downturn that followed](/blog/organisational-change-downturn/), and it applies to AI now. The cultural diagnosis is in [Europe's AI challenge](/blog/europe-ai-challenge/). ### Where in our organisation should we delegate the loop to machines first? Where scale beats expertise. The filter I use is simple: work that arrives in high volume, has definable success criteria, and fails recoverably is the candidate for autonomous iteration; routine customer enquiries, document extraction, and multi-step data processing qualify. Work that rests on proprietary knowledge, carries high stakes, or needs ethical judgement keeps a human in the loop, and high-volume proprietary work often suits a hybrid where experts review the edge cases. Start in a bounded domain, define success metrics before the pilot begins, and treat each approval as a conscious transfer of decision rights. [My explainer](/blog/agentic-ai-explainer/) sets out the full filter. ### How do we govern agents that act continuously without a human starting each task? With proportionate structure, not a rulebook written for chatbots. An always-on agent holds context for months and acts on enterprise data around the clock, which most policies I see never anticipated. The Minimum Lovable Governance principle applies: risk tiers, operating boundaries, escalation triggers, immutable audit logs, and kill-switches, built to make delegation safe rather than prevent it. Add policy on persistent agent access, monitoring for personal agents connecting to work systems, and an amnesty pathway from shadow usage to governed capability; BCG's 2025 research found **54%** of employees would use unauthorised tools when sanctioned ones fall short. I set the case out in [The Inference Migration](/blog/inference-migration/). ### Does our agentic AI have to be generative AI? No, and assuming so is where value leaks. The loop is the innovation; what runs inside it can be any AI discipline, and the strongest implementations coordinate several: machine learning for prediction, computer vision for extraction, robotic process automation for execution, generative models for synthesis. That buys maturity arbitrage, because fraud models and inspection systems carry decades of production history generative AI lacks, and it simplifies oversight, since governance sits at the coordination layer rather than per technology. McKinsey's 2025 State of AI research found **88%** of organisations use AI but only **23%** have scaled agentic systems, fewer still across disciplines. [The compound loop](/blog/agentic-ai-compound-loop/) makes the case. ## References Further reading drawn from the external research cited across this briefing's articles. - **McKinsey & Company** (5 November 2025): [The State of AI](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai). The State of AI 2025: 88% of organisations use AI in at least one function, but only 23% have scaled agentic systems. - **Gartner** (25 June 2025): [Gartner: Over 40% of Agentic AI Projects Will Be Canceled by End 2027](https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027). Gartner's 2025 prediction that over 40% of agentic AI projects will be cancelled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls. - **Boston Consulting Group** (26 June 2025): [AI at Work: Momentum Builds, but Gaps Remain](https://www.bcg.com/publications/2025/ai-at-work-momentum-builds-but-gaps-remain). AI at Work 2025: evidence that redesigning workflows, not deploying tools, is what converts AI into productivity. - **PwC** (2025): [The Fearless Future: 2025 Global AI Jobs Barometer](https://www.pwc.com/gx/en/issues/artificial-intelligence/job-barometer/2025/report.pdf). The 2025 Global AI Jobs Barometer: a 56% wage premium for AI skills and three times the revenue growth per employee in the most AI-exposed industries. - **Deloitte** (18 November 2024): [Autonomous generative AI agents: Under development](https://www.deloitte.com/us/en/insights/industry/technology/technology-media-and-telecom-predictions/2025/autonomous-generative-ai-agents-still-under-development.html). Projects agentic AI adoption growing from 25% of organisations piloting in 2025 to 50% by 2027. - **Menlo Security** (4 August 2025): [Menlo Security’s 2025 Report Uncovers 68% Surge in “Shadow” Generative AI Usage in the Modern Enterprise](https://www.menlosecurity.com/press-releases/menlo-securitys-2025-report-uncovers-68-surge-in-shadow-generative-ai-usage-in-the-modern-enterprise). Menlo Security's 2025 report on AI in the modern workspace, documenting a 68% year-on-year surge in shadow generative AI use (vendor research, press release). - **California Management Review** (15 August 2025): [Adoption of AI and Agentic Systems: Value, Challenges, and Pathways](https://cmr.berkeley.edu/2025/08/adoption-of-ai-and-agentic-systems-value-challenges-and-pathways/). Berkeley's analysis of the value, challenges, and pathways of adopting agentic systems in the enterprise. - **Social Market Foundation** (9 February 2025): [High energy prices threaten UK AI world-leading status, as data centres can’t keep up with AI ambitions](https://www.smf.co.uk/high-energy-prices-threaten-uk-ai-world-leading-status-as-data-centres-cant-keep-up-with-ai-ambitions/). Why UK energy prices threaten AI competitiveness; the energy dimension of always-on inference. - **MIT NANDA** (July 2025): [The GenAI Divide: State of AI in Business 2025](https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf). State of AI in Business 2025: over 80% of organisations have piloted generative AI and 5% have reached production, the GenAI Divide this briefing is built on. - **Gallup** (2024): [Gallup Culture of AI Benchmark Report](https://www.gallup.com/workplace/652784/culture-of-ai-and-adoption-report.aspx). The 2024 benchmark of major European companies: 13% of the workforce engaged, the cultural constraint behind the adoption question. - **McKinsey & Company** (25 April 2023): [Breaking technical debt’s vicious cycle to modernize your business](https://www.mckinsey.com/capabilities/mckinsey-digital/our-insights/breaking-technical-debts-vicious-cycle-to-modernize-your-business). McKinsey's 2023 analysis of technical debt, the source of the estimate that it consumes around 40% of IT budgets in large enterprises. ## The ideas beneath this briefing Ideas I've named and matured writing about AI Transformation: what each one means, and where it started. - **Compound Loop**: An agentic system that coordinates several AI disciplines running in parallel within a single loop, so their interaction effects multiply value and governance happens once at the coordination layer rather than per technology. [Read more](https://mariothomas.com/blog/agentic-ai-compound-loop/) - **Money for Old Rope**: Transforming forgotten or seemingly worthless legacy assets into new revenue streams, a phrase coined during a 1998 newspaper archive digitisation project and applied since to data and AI opportunity. [Read more](https://mariothomas.com/blog/ai-stages-of-adoption/) All concepts: https://mariothomas.com/glossary/concepts/ ## More Board Briefings More complete resources on AI and emerging technology for the Boards that need the full picture. - [The Great Remaking](https://mariothomas.com/briefings/the-great-remaking/): AI is restructuring how organisations think, decide, create, and deliver, and the gap between those that redesign work and those that wait compounds. - [Operating AI](https://mariothomas.com/briefings/operating-ai/): Most AI pilots never reach production; the AI Centre of Excellence is the operating capability that turns scattered experiments into governed, scaled adoption. - [AI Strategy](https://mariothomas.com/briefings/ai-strategy/): Approving good AI projects is not a strategy, and the Board's move is from accumulating pilots to a strategy it owns. - [AI & the Workforce](https://mariothomas.com/briefings/ai-and-the-workforce/): The workforce question is not how many roles AI removes, but whether people build capability that earns premiums or credentials that incur penalties.