Council Post: The AI Judgment Gap
Juan Santiago | CEO at Santex and Technology with Purpose | Cofounder of Incutex | Partner at Kalei Ventures.
getty
In sports, video-assisted refereeing (VAR) never lies. That's always been the point. The technology tracks motion at a microscopic level, surfaces data no human eye could catch and executes without hesitation. The hardware works perfectly.
When there's a failure, what fails, every time, is the human judgment layer on top of it. Pristine data fed into an interpretive void doesn't resolve conflict. It speeds it up.
AI works the same way inside your organization. Most companies have invested heavily in the cameras. Almost none have figured out who gets to make the call.
The enterprises pulling ahead right now aren't sitting on the largest models. They're defining which decisions need amplification, then building governance to stay in control of what follows.
Every company is a tech company, whether it knows it or not.
If your business manufactures sneakers, grows soybeans or runs a hospital network, here's what's true: You are already a technology company. The product just happens to be physical. Code has been consuming every industry for decades. The executives who don’t understand this aren't just falling behind. They're losing the ability to compete at all.
The winners went deep on data before they went shopping for models. Organizations that spent the past five years collecting, cleaning and actually learning from their internal data hold a structural advantage that compounds every quarter. The ones still seeing IT as overhead are financing their competitors' business.
The numbers show the scale of what's at stake. Global AI spending is projected to hit $2.5 trillion this year, a 44% jump year over year. Private sector investment is now within range of global defense spending. Capital markets have placed a very large, very concentrated bet on where value will accrue next. Observation mode is not a strategy.
Step away from the fear machine.
An entire industry runs on your anxiety.
Legacy consultancies, platform vendors and the tech media cycle push two contradictory feelings at once: that you're falling behind (buy now) and that the risks are catastrophic (buy this governance product, too). The result is reactive capital allocation and deployments nobody thought all the way through.
Gartner projects that by 2027, 40% of enterprise autonomous AI agents will be decommissioned or rolled back, not necessarily because the technology underperformed but because companies deployed without governance and discovered the damage only after agents had already acted. Recently, Hugging Face disclosed that an autonomous agent breached its own production infrastructure in what researchers are calling one of the first documented end-to-end AI-led cyberattacks. The same category of systems is being aggressively pitched to your teams right now.
Pulling back makes it worse. Stop rushing to buy a front-row seat for a show that hasn't finished rehearsals.
There's a real difference between using AI and adopting it. Renting a Ferrari to drive at 25 mph through city traffic looks impressive but accomplishes little. Adoption means putting that car on an open track with a driver who knows the telemetry, the braking thresholds, the weight distribution. Same vehicle. Completely different outcome. Those two scenarios share the same budget. Preparation is the only variable.
Exit the spectator role.
The 2026 PwC Global CEO Survey is blunt: 56% of CEOs report zero measurable return from their AI investments. No revenue lift, no cost reduction. Zero.
The 12% actually getting results share one thing: They didn't lead with tool procurement. They mapped their decision bottlenecks first, built a formal operational architecture and deployed enterprise-wide instead of hiding outcomes inside tidy innovation labs where accountability goes to die.
Strategic clarity before deployment. Every time, this beats model selection.
Here's what that means in practice: Automating a broken process doesn't fix it. The breaking just gets faster. AI amplifies whatever culture and decision logic already exist inside your organization. Ambiguous decisions become faster and more consistently wrong at scale. Bad data produces hallucinated confidence. The tool faithfully accelerates whatever it finds.
Build governance into your architecture from day one. Cloudflare put it precisely: Your content, your rules. The agentic internet is being built around owner control. Your organization needs to establish that authority internally before an external platform does it for you.
The knife is on the table.
An AI model is a sharp instrument. It can prepare a meal or cause serious harm. The tool has zero agency about which one happens. That's entirely on the person holding it.
With frontier models now largely democratized, owning one is the entry fee, not the advantage. Sustainable edge comes down to execution: which routine decisions you hand to automation, which critical ones stay human and where you draw the line between the two.
Sitting this out has a price. Every quarter in observation mode transfers relative advantage to whoever is moving with intent.
Filmmaker Lucrecia Martel described this moment as an era of transparent clumsiness. She's right. The noise is loud, the imitation is everywhere and the hype shows no signs of clearing. Leaders who stop to execute with genuine rigor have a window. It won't stay open indefinitely.
That work can't happen in isolation. The AI programs that deliver results are team efforts: people who genuinely understand what's changing, leaders who take ownership of the decisions their systems make, ecosystems built around clear accountability at every node. Skip the vendor dependency. Ditch the isolation.
The VAR system won't make the final call. You will. The only question is whether you're directing the game or still watching from the stands.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?