Part of our guide: AI strategy
The pace of change in AI is unprecedented, with new foundation models released quarterly and the vendor market consolidating and fragmenting simultaneously. Agentic AI went from research concept to enterprise deployment in under eighteen months, and regulatory positions evolve faster than most governance frameworks can track.
For a leadership team trying to be strategic about AI, this creates a real dilemma, because your annual planning cycle takes three months to produce a strategy and twelve months to execute it. In that time the technology market has changed materially and perhaps multiple times, so the temptation is to abandon strategy altogether and simply move fast, experiment and course-correct.
Why "just move fast" fails
The evidence is sobering without needing the headline numbers boards usually quote. In McKinsey's November 2025 survey, 39 per cent of firms attribute any enterprise-level EBIT impact to AI and just 7 per cent report full scaling, so most effort is not yet turning into value at scale. We set out which failure statistics survive checking, and which do not, in the AI statistics that survive checking. The pattern underneath them is consistent: speed in the absence of architecture amounts to chaos rather than agility.
The organisations that move fast without strategy are the ones accumulating AI debt, tool sprawl, integration complexity, governance gaps and skills concentration. They end up with plenty of AI experiments and very few AI capabilities that are operational, governed and delivering measurable value.
What durable AI strategy looks like
A durable AI strategy is not a fixed plan: it works as a structured framework in three layers that tells you how to make decisions as the market changes.
Fixed layer: business capability model. Your business capabilities, what your organisation needs to do well to deliver its strategy, do not change every quarter. A claims processing capability is a claims processing capability regardless of whether the underlying AI model is GPT-4, Claude or something that does not exist yet. Map your business capabilities, score them for AI opportunity and use the result as the stable foundation for all your AI decisions.
Adaptive layer: technology and vendor choices. The specific AI tools, platforms and vendors you select will change, so design for this by building abstraction layers that allow you to swap out underlying models. Negotiate contracts with exit provisions, maintaining internal understanding of what the AI is doing rather than outsourcing that knowledge to a vendor. See The CTO's Guide to AI Vendor Selection for the practical detail.
Governance layer: decision rights and review cadence. Your governance framework should define how AI decisions get made, who has authority to approve new capabilities and how often the portfolio is reviewed, making speed safe. When a new opportunity emerges, and it will, you do not need to restart the strategy conversation. You evaluate the opportunity against your capability model, apply your governance framework and make a decision in days, not months.
Quarterly strategic reviews, not annual strategies
The cadence that works is quarterly, with every three months bringing a review of your AI portfolio against your capability model. Three questions carry that review, and the first is whether the capabilities we deployed are delivering the value we expected. The second is whether the technology market has changed in ways that affect our current choices, and the third is whether there are new opportunities that score higher than what is currently on the roadmap.
This review takes a day, not a month, and it produces an updated roadmap rather than a new strategy document. The strategy, the capability model, the governance framework, the operating model design, remains stable, and what changes is the sequencing and the technology choices.
Gartner's prediction and what it means for you
Gartner predicts that 40 percent of enterprise applications will feature task-specific AI agents by the end of 2026, up from less than 5 percent in 2025. That is an eightfold increase in one year, and a strategy that cannot absorb that kind of change is a wishlist rather than a strategy.
The organisations that will handle this well are the ones with a clear capability model, proportionate governance and the discipline to evaluate every new opportunity against their strategic architecture rather than reacting to vendor hype.
For the capability-led approach that makes this possible, see Why AI Strategy Must Lead Technology, and for the board-level framing, see How to Write an AI Strategy Your Board Will Back.
Breathe delivers a capability-led AI strategy in 5-10 working days, including the governance framework that makes quarterly reviews possible.