The Real Reason Your AI Rollout Isn’t Working Has Nothing to Do with Technology
MIT spent three years inside 20 companies deploying generative AI. The organizations that failed had one thing in common — and it wasn't the software. Here's what the research actually found, and what it means for your team.
Field Notes from ArtWorks — July 2026
In early 2026, MIT's Working Group on Generative AI and the Work of the Future published the results of a three-year study. They had spent that time inside more than twenty companies — large, established, industry-leading organizations in healthcare, retail, finance, insurance, and manufacturing — watching what actually happened when generative AI tools were deployed at scale.
Not what was promised in the pitch deck. What actually happened.
The findings are worth reading slowly. Because buried inside a report about technology is one of the most important management documents of the decade — and most leaders deploying AI right now are going to miss it entirely.
What MIT found
The researchers identified three common problems that organizations were trying to solve with generative AI.
The bottleneck problem: skilled people buried under near-routine tasks that eat time and prevent higher-value work. AI handles the volume. The human reviews the output.
The cafeteria problem: getting the right answer requires tracking down the right expert across the organization. AI synthesizes what those experts would likely say. The human decides whether it's right.
The learning curve problem: developing expertise in complex domains takes years. AI accelerates the ramp. The human still has to know enough to catch the machine when it's wrong.
These are real problems. The solutions are genuinely useful — in the right conditions.
Here is the sentence that should concern every leader reading this:
"Management practices matter for how generative AI will shape jobs."
That's MIT. Not a consultant. Not a vendor. MIT — after three years, fifty organizations, and dozens of interviews with the leaders, managers, and workers on the front lines of this experiment.
The technology is not the variable. The organization is the variable.
Why some rollouts stick and others get quietly shelved
MIT documented what separates the AI applications that scaled from the ones that didn't. Three factors appeared consistently in the successes.
First: the problem being solved was already well-defined and recognized as valuable before AI entered the picture. The organizations that succeeded weren't experimenting with AI looking for a problem to solve. They had a problem. They found a tool that addressed it.
Second: the solution required a combination of technologies and a human in the loop — not AI alone. The best applications kept people meaningfully engaged with the work, not just rubber-stamping outputs.
Third: persistence. The teams that succeeded kept iterating when early results were disappointing. They had organizational support to fail and try again.
Notice what all three of these factors have in common. They are not technology factors. They are organizational culture factors. They describe environments where people are clear on the mission, supported when things don't work, and engaged enough to keep going.
Now ask yourself honestly: does that describe your organization right now?
The hidden cost nobody is measuring
MIT's report surfaces something that most AI adoption conversations completely ignore: what happens to the humans inside organizations where these tools are deployed badly.
The researchers found evidence of mental offloading — workers using AI to complete tasks without retaining the knowledge required to do those tasks independently. In short: doing without learning. The productivity numbers look fine. The skill development is quietly eroding.
They found evidence that AI tools designed to address the cafeteria problem were associated with less collaboration and less mentorship. The informal processes of gathering expert input, which were also social processes that built relationships and shared understanding, were being eliminated. Teams were becoming more capable on paper and less cohesive in practice.
They found that trust — in the tools, in the organization, in the direction things were heading — was a critical predictor of whether workers used AI effectively or used it as a liability shield.
None of these are technology problems.
They are the consequences of deploying technology into organizations that are not measuring the human conditions that determine whether that technology creates value or quietly degrades it.
What the best-run organizations understand
MIT's ten recommendations for employers close the report. Read them carefully.
Gather evidence before scaling. One size does not fit all. Learn when to trust. Minimize drudgery. Promote learning. Preserve teamwork. Design the interface for human awareness. Maintain domain expertise. Build in accountability. Create new work.
Eight of these ten recommendations are fundamentally about people conditions — not software configuration. They describe organizations that understand what their people are actually experiencing, what's driving them and what's blocking them, and how to respond before the signal becomes a crisis.
The organizations that are winning with AI are not necessarily the ones with the best tools. They are the ones with the clearest picture of their human environment — and the discipline to act on what they see.
Most organizations have no reliable way to see that picture in real time. They have last year's engagement survey. They have gut instinct. They have attrition data that arrives after the damage is done.
That is not enough anymore.
The measurement gap
This is where we come in.
At ArtWorks, we work with leadership teams, HR leaders, and organizations at exactly this inflection point — deploying new tools, navigating significant change, and discovering that the hardest part was never the technology.
Interplay™ is our human flourishing diagnostic. Built on Harvard's Human Flourishing Framework and validated across eight institutions with approximately 1,200 participants, it measures the conditions that determine whether people can do their best work — stability, connection, meaning, energy, guidance, belonging — and generates immediate, actionable next steps using Creative Problem Solving methodology.
It doesn't tell you what happened last year. It tells you what's happening right now. And it tells you what to do about it before your AI investment becomes another case study in what not to do.
MIT spent three years documenting the gap between what organizations assumed about their people and what was actually true. Interplay™ closes that gap.
The question worth Asking
If you are a leader deploying AI tools into your organization — or planning to — the MIT research leaves you with one question worth sitting with:
Do you actually know what your people are experiencing right now?
Not what they told you in the last all-hands. Not what your managers reported. What is actually happening, for the humans inside your organization, as the nature of their work changes beneath them?
If the answer is anything other than a confident yes — that's where we start.
