Artificial intelligence scales. Accountability does not.
That line, pulled from a joint report by Accenture and the Wharton School, sums up the moment we are living in with surgical precision. Autonomous AI agents are no longer a distant promise or a science fiction concept. They are here, operating in real time, making decisions, managing processes, and transforming entire sectors of the global economy — often without anyone around knowing exactly what is happening behind the scenes.
And that is precisely where the problem lies. 🤔
What is at stake with autonomous AI agents
The report, titled The Age of Co-Intelligence: How Humans, AI Agents, and Robots Are Redefining Value, did not come to rehash the usual talking points about automation and unemployment. The message is different this time, and far more urgent. We are not just talking about which jobs will disappear or emerge. We are talking about who, or what, is actually in control of the decisions that drive companies, markets, and entire lives. And the answer to that question, right now, is still way too murky for anyone to feel comfortable.
- The smarter AI becomes, the more it demands from the humans managing it
- Productivity gains only translate into real growth when there is a conscious redirection of people and resources
- Governance of these tools is still crawling while the agents are already sprinting
- Accountability for what these systems do remains, almost always, nobody’s responsibility
More than 50% of work hours in the American economy are already being impacted by AI agents, according to O*NET and Bureau of Labor Statistics data analyzed by the researchers. That is more than 120 million workers affected directly or indirectly, across 18 different sectors. In banking and capital markets, the share of hours impacted by digital agents alone already exceeds 45%. The numbers are too big to ignore, and the conversation we need to have has barely even started.
The central point of the study is not technological — it is human. The big question is not whether autonomous agents work. They do, and they work really well. The question is whether organizations are prepared to deal with the consequences of what these systems decide, execute, and eventually get wrong. Because mistakes happen. And when an autonomous agent makes a mistake at scale, the impact is not small.
As the report itself states: intelligence is scalable, but accountability is not. That asymmetry is at the heart of everything the study discusses. As AI removes the limits on how much reasoning and analysis can be done, humans still need to decide what matters, set the strategy, and above all, own the responsibility for the outcomes.
The lesson from Charlie Chaplin and Lucille Ball for the age of AI agents
Professor Eric Bradlow, who chairs the marketing department at Wharton and co-authored the report, made an analogy worth knowing about. He compared the current moment to classic scenes from film and TV. The first reference is the 1936 movie Modern Times, in which Charlie Chaplin plays a factory worker trying to keep up with an ever-faster assembly line — until he is literally swallowed by the machine.
The second reference is the Job Switching episode of I Love Lucy, where Lucille Ball ends up stuffing chocolates in her mouth because the factory conveyor belt moves too fast for her to wrap everything.
Nearly 90 years later, those images remain disturbingly relevant. The machines are smarter now. The stakes are higher. And according to the report, the humans operating these machines are falling behind in a way that should concern every boardroom in the country.
Bradlow explained the problem pretty directly: if one person in a 20-step process adopts AI and triples their productivity while the person at the next step is still running everything in Excel, the bottleneck does not disappear. It just moves. And that mismatch is going to become obvious very quickly. The image of Chaplin being sucked into the conveyor belt captured the early days of industrial capitalism in the 20th century. Today, the same metaphor applies to the age of autonomous agents: either you master the machine, or it can grind you up in the gears.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Productivity without direction is just noise
One of the most important points in the report is that the productivity gains generated by artificial intelligence do not automatically convert into real growth for companies or for the economy. It seems obvious when you read it like that, but in practice a lot of people still believe that all you need to do is deploy agents and wait for the results to show up on the balance sheet. It does not work that way.
The study modeled the numbers for a real company with $60 billion in revenue. The researchers estimated that full implementation of agentic AI could generate approximately $6 billion in annual revenue growth, plus $1.7 billion in productivity gains. But here is the important detail: by 2028, roughly one-third of those productivity gains would not show up as direct cost savings. They would manifest as freed-up capacity — hours that need to be deliberately redirected toward higher-value work, or they simply evaporate without generating any return.
The report is emphatic: productivity only transforms into real growth through intentional redistribution. If leaders do not channel that freed-up capacity into higher-value activities, the gains stall at efficiency and never translate into growth.
Think of it this way: if an autonomous agent takes over tasks that previously occupied three analysts for eight hours a day, what happens with those three people and that time? If the answer is nothing planned, the productivity gain vanishes. It becomes operational efficiency without strategic purpose, and that does not move any company forward in a sustainable way.
James Crowley, global president of industrial product practices at Accenture and co-author of the report, told Fortune that the big trap is not deploying too many agents, but not thinking about them as a coherent workforce. Everybody is building an agent here, another one there, sometimes thousands. What the research tried to do was take a step back and look at what the agentic landscape will look like at the enterprise level as a whole.
Crowley also pointed out that most companies have been overly focused on efficiency and productivity, when the revenue potential is significantly greater. The gains on the revenue side, he said, will eventually dwarf the gains on the efficiency side. And now is the right time to turn that conversation into a story that includes both sides.
Governance: the missing link
If there is one theme the report hammers home, it is governance. And that makes sense, because it is exactly the area where most organizations are still most vulnerable. Autonomous AI agents are already operating in critical environments, making decisions about credit, healthcare, logistics, communications, and much more, while the frameworks for control and oversight scramble to keep up.
The report highlights that AI agents are already spreading rapidly across enterprise value chains, frequently ahead of any formal strategy or governance. Nearly three-quarters of knowledge workers already use AI, often through unauthorized tools, in a phenomenon that has become known as shadow AI. By 2028, roughly one-third of enterprise applications are expected to have agentic capabilities built in. And yet, the governance architecture has not kept pace.
Governance of artificial intelligence systems involves far more than creating an internal policy or appointing an ethics committee. It involves clearly defining which decisions an agent can make fully on its own, which require human validation, how decision logs are recorded and audited, and what happens when something goes wrong. It also involves creating real accountability mechanisms, because without them any failure becomes a blame game where nobody takes responsibility and the problem simply repeats itself.
The report notes that companies that are already more mature in this area share a few characteristics. They treat governance as part of the system design, not as a layer added after everything is already up and running. They invest in internal transparency, documenting how agents make decisions and making that process understandable for the teams that depend on those systems. And they create clear channels so that any employee can question a decision made by an agent, without excessive red tape and without fear of looking old-fashioned for questioning the machine. 🛡️
The study goes further and suggests that organizations may need an entirely new executive role: a chief agentic resources officer, someone dedicated exclusively to managing and overseeing the workforce composed of autonomous agents.
Sales: the most promising and most risky area at the same time
The report identified that sales represents both the biggest revenue opportunity and the biggest governance risk for AI agents. It is a function that combines a massive volume of decisions, high suitability for digital agents, and elevated commercial risk — customer interactions, pricing, commercial judgment. Sales is simultaneously the top candidate for early agent deployment and, as the report puts it, a critical governance domain where trust, accountability, and human oversight need to be deliberately planned.
The view from those already deploying agents
Andrey Khusid, CEO of Miro, the productivity startup valued at $17.5 billion that gained attention for deciding to leave Russia after the start of the war in Ukraine, shared his perspective on the current moment with Fortune. Miro’s core product, a collaborative productivity software that has been around for over a decade, is now incorporating AI.
According to Khusid, for nearly 15 years the work on the platform was human to human. But then the dynamic shifted. Now, a lot of collaboration happens between humans and agents together. By bringing agents onto the platform, the company is enabling users to deliver work in an agentic way. It is more complex than human-to-human work, he explained, but it is much more powerful and with a much faster time-to-value. Before, you would need a human with a certain expertise or another expertise. With agents, you can have an entire team working alongside you with different specialties.
Still, Khusid acknowledged that agents can be prone to errors, just like humans, and that much of the current agentic work is a black box. Miro is working to unpack that opacity, so that it becomes possible to correct agents when they get things wrong. Recognizing that it feels like an agentic revolution, he tempered expectations: we are at the beginning.
Errors at scale: the risk of agents that hallucinate
Bradlow and Crowley openly acknowledged that agents can be prone to errors and even hallucinations, and at massive scale, this could lead to widespread failures. Imagine a scenario where an agent hallucinates an inventory number and that causes downstream agents to place absurdly inflated stock orders. Or a customer service agent that tells a consumer the issue is resolved when it actually is not, and no human steps in to correct it.
Crowley was blunt: the philosophy should be humans in the lead, not just humans in the loop. If humans are not consciously leading, the errors multiply at scale.
Bradlow brought a valuable technical perspective, drawing on his experience as a mathematician and data scientist. He explained that agents are built on the premise of reinforcement learning, which means they seek good outcomes as programmed by the human who defines the objective function. When agents get bad results, they change their approach. They adapt. It is not so obvious that humans learn in the same way. When an agent makes a mistake and you indicate what to reinforce, it should not repeat that mistake. Which makes Khusid’s point about opening the black box even more important.
Bradlow also referenced the British TV show Weakest Link, one of the BBC’s biggest hits, where the host would eliminate contestants by coldly declaring: you are the weakest link, goodbye. According to him, agentic AI is going to expose the weakest link in every organization. 👀
Accountability: whose responsibility is it?
This is perhaps the most uncomfortable question in the entire debate. When an autonomous AI agent makes a decision that causes harm — whether financial, operational, or human — who answers for it? The company that deployed it? The vendor that developed it? The manager who approved its use? The analyst who configured the parameters? In most cases we see today, the honest answer is that there is no clear definition, and that ambiguity is a massive risk that many organizations still do not take seriously enough.
Accountability in artificial intelligence systems needs to be thought of structurally, not merely as a reactive response to crises. This means organizations need to document, before putting any agent into operation, what the boundaries of that system’s actions are, what criteria it uses to make decisions, and who, specifically, is responsible for monitoring and answering for that agent’s behavior on a daily basis.
In a poorly designed agentic enterprise, a single human can suddenly find themselves responsible for an exponential cascade of consequences they never saw coming. Agents can reason, execute, and coordinate. What they cannot do is take responsibility for the final outcome. That is a fundamental distinction that needs to be at the center of any implementation strategy.
What the Accenture and Wharton School report makes clear is that the absence of accountability is not just an ethical problem — it is a business problem. Companies operating with agents without a clear accountability structure are creating liabilities that can turn into serious crises — regulatory, reputational, and financial. And as AI-specific regulations gain momentum around the world, organizations that do not build this structure now will pay a much higher price down the road. ⚠️
Real growth requires vision beyond efficiency
Bradlow agreed with Crowley that the revenue potential is one of the biggest findings in the study. He cited comments he heard at an executive breakfast that Wharton and Accenture co-hosted at Nvidia’s GTC conference in March. According to him, companies are managing to do things they simply could not do before, launching new types of products that were previously unimaginable.
But that growth premium comes with a human cost. The more intelligence you scale, the more accountable — and irreplaceable — your human leaders become. You cannot delegate ultimate responsibility to a system that, no matter how sophisticated, operates within parameters defined by humans and lacks the ability to handle the ethical, social, and strategic implications of its own decisions.
The word deliberately appears throughout the report’s 40 pages like a constant refrain. Leaders cannot simply enable agents and expect value to emerge on its own. They need to set explicit P&L targets, build human-led operating models, and assign clear decision rights before agents go into operation.
What changes from here
The scenario described by the report is not catastrophic, and it is important to make that clear. The growing presence of autonomous AI agents in the economy does not have to be a problem, as long as organizations, governments, and the people working with these technologies take seriously the responsibilities that come with this power. The real productivity gains, the ability to solve complex problems at scale, and the execution speed these systems offer are genuine opportunities, not threats disguised as progress.
But for all of this to convert into real, lasting value, governance needs to grow at the same pace as adoption. Accountability structures need to be built before the need for them shows up in the form of a crisis. And the people working alongside these systems need to be prepared — not just technically, but also to play the role of critical oversight that no autonomous agent can perform on its own.
The phrase modern times still means exactly what it meant in Chaplin’s era: you need to master the machine, or you risk being ground up in the gears. Artificial intelligence scales fast. The challenge now is making sure accountability scales with it. 🚀
