Dataiku, a global AI platform company, surveyed 600 CIOs (chief information officers) around the world. The results are striking.

Seventy-four percent said they regret at least one of the AI vendor or platform decisions they've made over the past 18 months. Sixty-two percent have faced direct pressure from their CEO over those decisions. Twenty-nine percent said they were repeatedly asked to justify their AI investments because they couldn't adequately explain the results.

Eighty-five percent expect their own compensation to be tied directly to AI performance, and 71% said their budgets would be cut or frozen if they failed to show results by the first half of this year. "Are you using AI well?" has become a question of survival.

Chosen in a hurry, with no results to show

So why do 74% have regrets? Did they simply pick the wrong technology? Sure, that happens sometimes—but the bottleneck the report points to lies elsewhere, and that's the crux of it. "The bottleneck today isn't building AI; it's proving you can trust it, govern it, and defend it."

Unpacked, it goes like this. AI tools themselves are everywhere. Adoption is fast. The trouble starts after you adopt. Can you explain why this AI produced the result it did? Can you trace what happened when something goes wrong? Do you have the flexibility to switch to a different tool? If you can't answer those questions, you've adopted the tool but can't operate it; because you can't operate it, you can't prove results; and because you can't prove results, your budget gets cut.

This is borne out by the fact that 85% of CIOs reported AI projects being delayed or halted at the actual operational stage because of a lack of explainability or traceability.

Spreading faster than anyone can control

There's an even more worrying figure. Fifty-four percent of CIOs have discovered unsanctioned AI inside their organization—so-called "shadow AI." Eighty-two percent worry that employees are building AI tools faster than the IT department can manage.

This goes well beyond an individual quietly using ChatGPT for work. Entire departments are building their own AI tools or bolting on outside services, and the IT department can't see the full picture. While 87% said AI agents are already being used in core business functions, only 25% said they have real-time visibility into every AI system in use.

Organizations don't know where AI is being used, how, or on whose authority. No wonder 89% said this situation could create serious technical debt.

This isn't a technology problem

Step back, and what this report describes isn't a failure of technology selection. It's the absence of a framework for judgment.

The criteria for deciding which AI to adopt were never clear, the metrics for measuring results after adoption were never defined, and there was no structure for reversing course when something went wrong. Technology was abundant, but the capacity to evaluate it, choose it, and own the consequences was lacking.

This isn't unique to AI. It's a pattern that repeats every time a new technology arrives. The same thing happened when cloud first emerged, and when companies rolled out ERP. The more powerful the tool, the more the judgment of the people operating it matters.

You can break down the capabilities an organization needs in the AI era into three: skill, knowledge, and attitude. AI is rapidly replacing the first two. Skills like coding, data analysis, and document writing. Knowledge like market research and competitive analysis. In these areas, AI is already faster and more accurate than people.

But what the 74% in this report regret isn't a matter of skill or knowledge. It's the realm of attitude—judging whether to adopt this tool now, what standard to measure results against, and how to reverse course if something goes wrong. The impatience to adopt in a hurry, the herd instinct of doing it because everyone else is, the complacency of just getting started without performance metrics. The causes of failure all lie in attitude.

In 2026, AI puts leadership to the test

The report frames 2026 as "the year AI begins testing executive leadership in earnest." Explainability, AI agent accountability, tech-stack flexibility, governance, and proving ROI—these five decisions, it says, will determine whether AI becomes an asset or a liability.

Seventy-three percent of CIOs believe that if the AI bubble bursts, it will cause major upheaval for their companies. Fifty-seven percent said their company's very survival could be at risk. This is neither optimism nor pessimism, but a clear-eyed reading of reality.

In the end, what separates the organizations and individuals who survive the AI era isn't which AI they use. It's the judgment they exercise in front of it. AI fills in the skill and the knowledge. What it can't fill in is the attitude that judges. By what criteria will you choose? How will you reverse course when you fail? How will you define results? The person who can answer those questions is the one who uses AI as a tool. The person who can't becomes the one AI drags along.