We’re forgetting the most critical system in the AI loop: the human brain

The question I am asked most frequently today is no longer “which AI tools should we deploy?” but “why are our people not performing at the level our technology investment should be enabling?”

The numbers tell a story that should concern every C-suite leader and CIO investing in artificial intelligence right now. According to a 2025 MIT study, 95% of enterprise AI pilot programs are failing to deliver measurable financial returns. BCG research puts it plainly: 74% of companies struggle to achieve and scale value from AI. And yet, as an Orgvue analysis from 2026 reveals, 57% of organizations deploying AI are doing so primarily because their competitors are, not because they have a strategy.

Read that again. More than half of the organizations racing to deploy AI are running a technology adoption program without a clear reason for it. That is not transformation. That is theater.

Investment is accelerating regardless. Deloitte’s 2025 survey of over 1,800 senior executives found that 85% of organizations increased AI investment in the past year, and 91% plan to increase it again. Meanwhile, only 6% of those organizations qualify as genuine high performers, those seeing a 5% or more impact on earnings. The gap between what organizations is spending and what they are getting back is not a technology problem. It never was.

The real reason AI initiatives are underperforming

The most forward-thinking technology leaders I’ve encountered have already made this shift. They understand that the most significant risk to their technology investment is not a system failure. It’s a human one. AI amplifies what already exists in an organization. If the underlying human system is undisciplined, reactive or misaligned, AI will accelerate those same weaknesses.

I’ve seen this pattern repeat consistently. Companies make significant capital investments into AI infrastructure, but invest almost nothing in redesigning how people work, make decisions or learn from failure. The result is sophisticated capabilities layered on top of outdated operating models, broken decision-making cultures and teams that have never been taught to think critically about outputs.

Three human behaviors, in particular, derail good technology more than any other.

The first is confirmation bias. People use AI to validate what they already believe rather than to genuinely challenge their assumptions. The model becomes a sophisticated mirror rather than a thinking partner.

The second is risk aversion disguised as due diligence. Organizations create endless review cycles and approval layers that slow execution to the point where the insight the AI generated is no longer relevant by the time a decision is made.

The third, and perhaps the most damaging, is the absence of a failure-forward learning culture. AI systems improve through iteration and feedback. But many organizations still treat failure as something to be managed politically rather than mined for intelligence. When people are incentivized to hide errors rather than learn from them, the feedback loop that makes both humans and AI systems smarter simply breaks down.

What does good human infrastructure look like?

Here is the question I find most important, and most consistently ignored at the leadership level. AI systems require clean data, disciplined governance and structured feedback loops. What is the human equivalent of that infrastructure?

Clean data in a human system is the quality of thinking people bring to their roles. Their ability to observe accurately, reason clearly and separate signal from noise. Most organizations invest nothing in this.

Governance on the human side is behavioral discipline. The standards, habits and decision-making frameworks that determine how people operate under pressure, not just in ideal conditions.

And the feedback loop equivalent is what I call failure-forward learning. The organizational capacity to extract structured insight from what goes wrong and feed it back into how people and teams operate.

When these three elements are absent, you do not have a human operating model. You have a group of individuals hoping that good intentions and expensive software will be enough. Newsflash, they won’t be.

The organizations seeing the strongest returns from AI are not necessarily those with the most advanced models. They’re the ones who invested in the human infrastructure around the technology first.

Five actions leaders must take now

1. Stop treating AI adoption as a technology deployment. AI is a business transformation that happens to involve technology. The moment a leader frames it as an IT initiative, they have constrained its potential before it has even begun. Framing determines resourcing, sponsorship and accountability. Get it wrong, and everything that follows is compromised.

2. Build structured reflection into the operating rhythm. High-performing teams use short, focused weekly review cycles, rather than lengthy retrospectives. They ask the right questions. What did we learn? What did we assume that proved incorrect? What do we need to adjust? Without this, people repeat the same cognitive errors at increasing speed.

3. Establish decision hygiene. Teams need explicit frameworks for how decisions are made. Make it clear who holds accountability, what information is required and how outcomes are tracked against the reasoning that produced them. AI surfaces decisions faster than ever before. Without decision hygiene, that acceleration becomes a liability.

4. Measure outcomes, not inputs. Too many leadership teams celebrate the number of tools deployed, licenses purchased or training hours completed. None of that is operational value. Operational value is measured in decision quality, execution speed, cost efficiency and competitive differentiation. Hold yourself and your organization to outcome metrics from day one.

5. Model the behaviors you want to see. Intellectual curiosity, comfort with iteration and the willingness to be publicly wrong and publicly learning are not soft cultural aspirations. They are hard operational requirements. A team operating in cognitive overload or fear will corrupt even the most sophisticated AI initiative from the inside.

The most critical system in the AI loop

In the next three to five years, AI itself will be largely commoditized. The competitive advantage won’t come from simply having the technology, but from the quality of the human system wrapped around it.

Every serious organization investing in AI today has people responsible for model performance: Tracking what the system gets wrong, retraining on new data, auditing outputs continuously. That rigor is appropriate. However, in most organizations, the same level of discipline is almost entirely absent when it comes to evaluating and improving the ‘human system’ surrounding the technology.

When did you last audit the quality of thinking in your senior team with the same discipline you audit your data? When did you last build a genuine feedback loop around human performance, not an annual review, but a real mechanism for learning and adjusting? For most leaders, the honest answer is rarely, or never.

The organizations that will define excellent AI-augmented performance will be those whose leaders understood, at a foundational level, that they were always running two systems simultaneously, and that the performance of one was always a ceiling on the performance of the other. AI will continue to advance at a pace that is difficult to overstate. But the human brain, its capacity for judgment, creativity, ethical reasoning and genuine connection, will remain the most critical system in the loop.

Treat it accordingly.

This article is published as part of the Foundry Expert Contributor Network.
Want to join?