Two years ago, the goal was to get the human out of the way. Build the AI agent, point it at the work, let it run. The promise was speed: tasks that used to take a team a week, done in minutes, untouched by human hands.
Enterprises tried it. Then they did something the autonomy story didn’t predict.
According to KPMG’s AI Quarterly Pulse Survey for the first quarter of 2026, 63% of enterprise leaders now require a human to validate the outputs of their AI agents. A year earlier, that figure was 22%. The requirement nearly tripled in twelve months. KPMG surveyed hundreds of U.S. executives at billion-dollar companies, part of a global sample of more than 2,100 leaders across 20 countries.
These are the firms with the most capital and the most reason to automate. They had every incentive to remove the human. Instead they put the human back, and they did it fast.
The lesson inside that number is one I keep returning to in my own work. The automation was the easy part. The structure you have to build around it is the hard part.
I call that structure the Accountability Surface: the layer of observability, audit trails, and human sign-off that surrounds an AI system and makes its decisions explainable and defensible.
The Accountability Surface is what the market has started to pay for. The feature that generates an output is now cheap. The apparatus that lets a company stand behind that output is what carries value.
This is augmentation over automation, the position I’ve argued since 2023, showing up in enterprise budgets. Back then it was a claim about how people ought to work alongside these tools. In 2026 it’s a line item.
Governance turned out to be the job.
The path to 63% isn’t mysterious. Organizations moved fast. They shipped agentic workflows that booked, classified, drafted, and decided. Then a regulator, a client, or a board member asked a simple question: why did the system do that?
Too often, nobody could answer.
You can’t defend a decision you can’t explain. Not to an auditor reviewing a denied claim, not to a customer disputing a charge, not to your own leadership asking why the model flagged one account and cleared another. An AI agent that produces an answer with no trace behind it is a liability wearing the costume of efficiency.
So enterprises did what mature industries always do with powerful, opaque processes. They wrapped them in controls.
The KPMG data shows this is the mainstream now. The companies requiring human validation are the billion-dollar firms furthest along the adoption curve. They kept investing in AI. They added structure because they finally understood what scaling it safely costs.
That cost has a shape. It looks like dashboards that surface why a model reached a conclusion, and logs that capture every input, output, and confidence score. It looks like review screens that let a qualified person approve or reject in seconds, so oversight is real work instead of a rubber stamp.
The model produces the output. Everything wrapped around it to make that output trustworthy is the Accountability Surface, and building it is where the durable jobs are.
I’ll admit this got concrete for me recently. I run a small set of AI agents that do research for me overnight, pulling data, tracking the developments I follow, and flagging what changed while I slept. It’s useful, and a little unnerving the first morning you wake up to a stack of findings you didn’t watch get assembled.
When I gave those agents the ability to spawn their own helper agents, my instinct surprised me. Instead of stepping back, I fenced them in.
I locked down what they were allowed to touch. I kept every irreversible action, anything that sends or changes a record, behind my own review the next morning. I added a separate pass whose only job is to check the first agents’ work before I rely on any of it.
I’d built an Accountability Surface for my own small operation before I let it scale. Nobody told me to. The work demanded it.
That instinct maps onto a framework I’ve used for years to describe who an organization needs in an AI economy. Three profiles do the work.
AI Producers build the systems. At the core are the engineers and developers who design the agents and the observability and audit layers beneath them. That circle is widening fast. No-code and AI-assisted tools now let analysts and business owners stand up working applications without writing much code, so producing is no longer the engineers’ monopoly.
AI Practitioners run the tools every day. The tariff analyst checking an LLM’s classification, the claims reviewer who can tell when a flagged file is wrong, the logistics planner validating a model’s route. They pair deep knowledge of how the business runs with enough understanding of the model to know when to trust it and when to overrule it.
AI-Aware Individuals are the executives and policymakers who set direction and sign off without operating the tools themselves.
Here’s what the KPMG number does to that picture. When 63% of firms require human validation, demand for AI Practitioners stops being a forecast and turns into a hiring plan. Validation is their job. The Accountability Surface gets built and staffed by exactly the people the framework named.
That second profile is where the most valuable hire in a company now sits. For a decade the prize was the person who could produce: build the model and ship the dashboard. AI made production cheap. Comprehension stayed expensive. The person who understands how the business actually runs and how the model actually works, both at once, is rare, and rare is what companies pay for.
That person is the spine of the Accountability Surface. Done well, validation is a senior judgment call: can this answer be defended, and would the reviewer catch the error and explain it? A company that treats the job as a clerical checkbox staffs it with people who can’t tell when the model is wrong, which defeats the purpose. The firms that win at this pay for comprehension. The clicking was never the job.
It also raises the bar on the AI-Aware. Being aware is no longer enough. A leader signing off on an agent’s decision can’t vouch for what they don’t understand, and an executive running an AI overhaul she’s never tested with her own hands is making a bet she can’t see. “The system told me it was fine” is not a defense a regulator accepts. Awareness has to become readiness, the shift from knowing AI matters to knowing enough to govern it.
The World Economic Forum’s Future of Jobs report sharpens the stakes. It estimates 39% of workers’ core skills will need to change by 2030, expects AI and information processing to affect 86% of businesses, and projects 170 million new jobs over the same stretch. A real share of those jobs live on the Accountability Surface.
One clarification, because the headline number invites a misread. Requiring human validation isn’t the same as reviewing every keystroke. Most firms apply oversight where the risk is highest and let routine actions run. The Accountability Surface is structured oversight, calibrated to consequence, and companies are designing it on purpose.
If governance is the job, the question for everyone becomes where they fit and what to build. Each profile has a move.
For AI Producers and the firms that employ them: treat the Accountability Surface as a product. The observability layer, the audit trail, the review interface, these are features customers will pay for, because they’re what let customers sleep at night. A company that ships an agent with a clean explanation of every decision has a real edge over one that ships a faster black box.
For AI Practitioners: the skill that’s appreciating in value is judgment fused with fluency. Knowing a domain well enough to catch a confident, wrong answer, and knowing the tool well enough to see how it got there. That pairing is harder to automate than the underlying task, which is exactly why the stable work is heading toward it.
For AI-Aware leaders: make governance a requirement at the design stage. Ask how a system will be audited before you ask how fast it runs. Decide which calls an agent may never make on its own. The firms that wrote those rules early are the ones in the KPMG survey scaling with confidence today instead of scrambling after a bad headline.
There’s a through-line under all three. The old survival line, the one quoted in business until it’s threadbare, still holds: the survivors won’t be the strongest or the smartest. They’ll be the ones that adapt best to change. In an AI economy, the best adapters are the best governors.
The workers and companies who understand that are the ones building the structure that lets everyone else trust the technology.
The autonomy story got the power of these tools right. It got the location of the value wrong. The market has now settled that question with a number: 63%, up from 22%, in a single year.
Structure is how you scale trust. A company can buy the most capable model on the market, but with no way to explain and defend what it does, that capability stays stuck in a pilot, too risky to turn loose.
The next wave of advantage will go to whoever builds the Accountability Surface first and staffs it with people who can do the judging.
I built a modest one for a handful of overnight agents before I’d trust them with my own errands. The billion-dollar firms are doing the same thing, at scale. The data finally caught up to the argument.
Editor’s Note: The above commentary was penned by Dr. Daniel Covarrubias, director of the Texas Center for Border Economic and Enterprise Development at the A.R. Sanchez, Jr. School of Business at Texas A&M International University. He writes on trade, logistics, and the economics of exponential technologies. The commentary appears in the RGG Business Journal with the permission of the author.
