The CEO’s AI Paradox: Why Slowing Down May Be the Fastest Way to Win

Meta’s latest earnings announcement offered another indication of the extraordinary sums being committed to artificial intelligence. In October, the company narrowed its 2026 capital expenditure forecast to $135 billion to $145 billion, raising the lower end by $5 billion, with AI infrastructure a major driver.

Yet just days earlier, Meta and Apple had demonstrated how powerful AI models could run locally on Apple hardware, potentially offering businesses greater control and less dependence on costly cloud-based services. The contrast is striking. Even as investment accelerates, the technology and economics of AI continue to evolve. For most business leaders, the underlying question is familiar: how much should we commit today when tomorrow’s technology, costs and competitive landscape remain uncertain?

A contradiction lies at the heart of the AI debate. We are constantly told that the greatest risk is moving too slowly. Yet some of the most consequential mistakes a CEO could make may come from moving too quickly into decisions that are difficult and expensive to reverse.

This is not an argument for waiting on the sidelines. It is about distinguishing between moving fast to learn and moving fast to commit. They are not the same thing.

The Pressure to Respond Is Real.

AI is expected to change most industries. What is still unclear is where value will appear first, how much it will cost to capture, and which business models will be disrupted.

Stanford’s 2026 AI Index reports that 88% of surveyed organisations were using AI in at least one business function in 2025. But adoption is not transformation. Research from the consulting industry points to a persistent gap between AI experimentation and enterprise-wide impact. While many organisations are achieving productivity gains and improvements in individual functions or processes, relatively few have translated these successes into substantial, measurable improvements in overall business performance.

A CEO can be surrounded by successful demonstrations, enthusiastic employees and impressive vendor presentations, yet still struggle to answer a simple question from the board: what has actually changed in the economics of our business?

Everyone Has an Opinion. Not Everyone Has the Same Incentive.

Imagine a CEO preparing to approve a major AI investment. The technology provider is confident its platform can deliver the future. The technology team wants resources to build. Business leaders want a competitive advantage. Finance wants credible returns. The board wants assurance that management is neither falling behind nor gambling the company’s future.

Most are acting in good faith. But they see the decision through different lenses, and their incentives don’t align.

The boundaries between selling technology and advising on its use are blurring. In May, OpenAI announced a deployment company to help enterprises integrate AI inside their operations, alongside an agreement to acquire applied-AI firm Tomoro. The move shows how difficult it is to take AI from demonstration to everyday use. It also illustrates why buyers should understand who is advising them, who is implementing, and who benefits from the technology selected.

Good vendors are of great value. But even the best cannot be expected to be indifferent to which platform a customer chooses, how much it spends, or how dependent it becomes.

The CEO’s responsibility is not to distrust everybody. It is to ensure different perspectives are tested before they become a single, apparently inevitable recommendation.

The Problem with Betting on a Moving Target

Traditional investment decisions assume a reasonably stable view of what you’re buying. AI challenges that assumption. Capabilities are improving, prices are shifting, and tasks that once required specialist software may now be achievable differently.

There are already signs of this shift, with many organizations reporting that they have decided against purchasing software products or new features because AI-powered coding tools have made it possible to develop the capability internally.

This does not mean building is always better than buying. It means the alternatives may change while procurement is still underway.

CEOs need to ask what happens if a better option appears in six months. Can the business change models or providers? Can its data move? Does it retain control of critical processes? Are commercial commitments matched to evidence rather than optimism?

A large investment may still be sensible. But it should be a deliberate bet with clear assumptions, not a reaction to a competitor’s press release.

Slow Down the Decision, Not the Learning.

One of the most useful distinctions a leadership team can make is between reversible decisions and irreversible ones.

Small experiments, targeted workflow redesign and controlled trials can move quickly. They reveal where employees save time, where customers benefit, and where technology fails. They also expose less glamorous problems: poor data, unclear accountability, security concerns and processes nobody has challenged for years.

Large platform commitments, outsourcing arrangements, restructuring and the removal of critical human controls deserve greater scrutiny. Their consequences may outlast the technology assumptions behind them.

This is where slowing down can make an organization faster. Weeks spent exposing weak assumptions may prevent years of implementing the wrong answer. Similarly, letting teams experiment within sensible boundaries may teach a CEO more than waiting for a perfect enterprise strategy.

McKinsey’s research suggests high-performing organizations are more likely to redesign workflows and pursue growth and innovation alongside efficiency. That is as much a management challenge as a technology challenge.

What the Board Needs to Hear

Boards are being asked to support consequential decisions where even experienced executives cannot claim to know how everything will unfold. Demanding certainty is unrealistic. Accepting enthusiasm instead of evidence is equally unhelpful.

A better board conversation starts with practical questions. What problem are we solving? What evidence supports the expected returns? Which assumptions could prove wrong? What can we learn before committing more capital? What would cause us to stop or change direction?

The conversation must also include people. What work will change? Which employees need new skills? Who remains accountable when AI makes a recommendation or takes an action? What will customers actually experience?

Boards do not need to become rooms full of AI engineers. They need sufficient understanding and independence to challenge management constructively. CEOs, in turn, need boards that distinguish between a well-managed experiment that fails and a poorly examined investment that succeeds by luck.

The Advice CEOs Should Value Most

Executives have no shortage of access to AI expertise. Technical specialists are essential. But technology expertise is not the same as sound judgment about a company’s strategic choices.

The most valuable adviser may be the one willing to ask an uncomfortable question when everyone else wants to proceed. Why are we solving this problem? What are we assuming about customers? Are we redesigning work or automating a bad process? What would we do if this were our own money?

Good advice also requires the confidence to say, “We do not know yet.” That admission should trigger a disciplined way of finding out, not paralysis.

An adviser who brings competing voices together, exposes trade-offs, and helps leaders decide amid uncertainty offers something different from another prediction about the next model release.

There is an irony here. As AI makes analysis, presentations, and technical information easier to produce, the premium on human decision-making may rise. Listening, questioning, understanding organizational politics, and establishing trust across a divided executive team become more important, not less.

The Real Race

Every CEO knows standing still carries risks. Competitors are learning. Employees are experimenting. Customers are changing their expectations. The answer is not to slow everything down.

It is to move quickly where speed creates learning, remain disciplined where decisions create long-term obligations, and recognize the difference.

The next time an AI proposal reaches the executive committee, perhaps the most useful question is not, “How quickly can we approve this?” but “What must we understand before committing, and how quickly can we find out?”

In a market that rewards the appearance of urgency, the CEO who makes room for genuine debate may ultimately move further, and faster, than the one who never stops to question the direction.