Employees are becoming bot bosses — without boss pay

You entered your profession because you wanted to teach, code, write, create, or design. But thanks to AI — and the immense pressure to use it — you now spend less time practicing your craft and more time managing the technology that’s doing the work you wanted to do.

For many professionals, that means supervising agents: tools designed to perform tasks autonomously but that, in practice, require workers to set expectations, delegate assignments, evaluate performance, demand revisions, and intervene when things go wrong. Others manage a steady pipeline of AI-generated work, reviewing the output and taking responsibility for what ultimately ships.

Congratulations! You’re a manager now. You just won’t be receiving the title or the raise.

The idea that AI will turn everyone into a manager has been floating around for a while. Now companies are building around that premise. Salesforce unveiled earlier this month “job-ready agents” designed to do work in sales, customer service, e-commerce, and internal operations, while giving human employees tools to teach agents about their companies, coordinate them, and improve their performance. Meanwhile, Wipro, an Indian IT services company, has increased productivity through AI by the equivalent of 20,000 workers, according to chief technology officer Sandhya Arun, who envisions “the same engineer managing a bunch of agents.” At Daytona, a 30-person AI startup, each of the company’s 16 engineers manages an average of five AI agents, CEO Ivan Burazin recently told Business Insider. It’s to the point where he says no one at the company writes code anymore.

For some workers, managing a digital workforce has become a new kind of workplace flex. I’m not being replaced by AI; I rule over AI. But plenty of others never asked to become the boss of the bots. That includes software engineers who pursued the principal or staff track because it offered career advancement without becoming a supervisor, only to find themselves becoming de facto managers of machines anyway.

Sinda Khenine became a software engineer because she wanted to build things. Now, as a software and AI engineer at appliance maker Electrolux who is simultaneously building her own AI business, she spends much of her time managing the systems that do the building.

Agents have allowed Khenine to produce far more code and build a company by herself — something she says would have been “10 times harder” in the past. Now she spends less time doing the engineering work that drew her to the profession. Instead, she chooses models, supplies context, sets boundaries, reviews the work, and coaxes the output of multiple agents into a finished product.

“The effort is more focused on the coordination itself rather than the engineering problem that we are solving in the first place,” Khenine says.

Khenine, who also leads a human team at Electrolux, says managing agents can be more complicated than managing people. They require a lot of setup, and the oversight doesn’t stop once they begin working.

“It’s like a loop,” she says. “You need to monitor the results. You need to see the progress. You need to decide if you need to rerun or finish the job.”

Khenine considers that managerial work part of being a software engineer these days and doesn’t believe it entitles her to a raise. But she says people outside the AI industry may overlook how much labor agent orchestration requires.

Mari Nnanna, a freelance technical writer and backend developer, estimates that she now spends 20% to 30% of her working time managing agents.

“It’s suddenly becoming a skill you need to have,” Nnanna says.

Agents can help her work faster, particularly by searching documentation or explaining unfamiliar material, but they need a lot of handholding. Otherwise you’ll end up with “so many mistakes.”

Although Nnanna would prefer to remain an individual contributor, supervising AI has become part of her work — and is unavoidable when she uses it if she doesn’t want to sacrifice accuracy.

The shift extends beyond techies.

Patricia Diaz-Hymes, a product marketing consultant, estimates that she now spends about 30% of her time supervising AI.

“It does feel like management,” she says. “I give my agents names. I have their configurations. I try and talk to my agents in a very human way.”

I’m trying to avoid my judgment from living just in my head and instead — at least partially — for it to live in my agents.

That means giving them defined jobs, familiarizing them with her business, and explaining why their work falls short rather than simply asking them to try again. Recently, her strategy agent, “Atlas,” drafted questions for an initial conversation with a prospective client. The questions sought agreement rather than a diagnosis of the client’s problems, so Diaz-Hymes rejected them and explained why. Atlas hasn’t made the same mistake since.

“I’m trying to avoid my judgment from living just in my head and instead — at least partially — for it to live in my agents,” she says.

Diaz-Hymes builds similar knowledge systems for her consulting clients. She thinks marketers’ larger role will be defining what good work looks like and building that judgment into AI systems. “Supervising is just this funny middle we’re in right now,” she says.

Not everyone taking on this management work has chosen it — or feels equipped to do it.

An American professor teaching at a university in China tells me she has been made responsible for managing AI in her classroom, despite having no special training in the technology and being deeply skeptical of it. She asked not to be named for fear of repercussions from her employer.

“I’m being asked to train my students to use it,” she says. “I’m also asked to make sure they don’t use it for assignments. A fun contradiction.”

She is also expected to teach students to use AI “ethically,” though she says no one has clearly defined what that means. Telling students that using AI may itself be unethical risks making her seem insufficiently “technology forward,” she says. Focusing too much on its environmental impact can similarly be treated as ignoring the technology’s inevitability.

At the same time, incoming students are being told AI is “the future” and that they will learn about it in their classes — leaving professors like her to figure out what that instruction should actually look like.

“So then that’s dumped on me to manage because no one knows how to use it, why it’s useful, or what it’s actually going to do,” she says.

These examples reflect a much broader change in how work is being divided between people and machines. In a global survey of nearly 12,000 employees, 47% say they spend more time managing and directing AI than doing the work themselves, according to Boston Consulting Group. Sixty-seven percent say AI has taken over simpler tasks, leaving them with more complex work.

In other words, AI may be doing more of the execution, but it is leaving humans with more of the judgment, responsibility, and oversight.

David Martin, leader of the People & Organization practice at BCG and coauthor of the report, says the managerial shift is particularly pronounced among certain career types like software engineers and marketers, who can face an overwhelming amount of agent oversight, leading to AI “brain fry.”

These are “people who historically weren’t managers, who are now feeling like they have to manage,” Martin says. They must tell agents what to do, judge their work, and own the outcome. “That is a new muscle that historically they might not have had to build.”

The new responsibilities are also arriving faster than the training required to handle them. In the BCG survey, 72% of respondents say AI has considerably changed the skills expected of them, but only 36% believe they have received adequate upskilling — a figure Martin says has remained unchanged across BCG’s successive surveys, even as more workers have begun using AI.

If you are a manager of people, you have the financial advantage. If you are managing agents, nothing changes for you.

Martin says much of the training companies provide is classroom-oriented and generic. Employees instead need to use AI on real assignments, compare what worked with colleagues, and develop the judgment to evaluate its output.

And while workers are taking on more managerial responsibilities, their titles and pay often aren’t keeping pace.

According to data from an upcoming survey from the workforce analytics company Visier, 54% of full-time US employees surveyed say their roles or careers have significantly changed due to AI in the past two years. Of those whose roles changed, only 17% have been promoted.

Andrea Derler, principal researcher at Visier, says individual contributors are increasingly doing “managerial activity” without the formal recognition that traditionally comes with management.

Previous research by the Kellogg School of Management based on Visier’s data found that individual contributors earn about 33% less on average than team leaders at the same firm.

“If you are a manager of people, you have the financial advantage,” Derler says. “If you are managing agents, nothing changes for you. You still have the same role.”

Maria Colacurcio, CEO of compensation software company Syndio, says she is “100%” seeing individual contributors take on what she calls “management light” responsibilities governing agents.

“If I’m responsible for assigning work to five agents, evaluating their performance, correcting them, being accountable for what they produce, what exactly makes me an individual contributor anymore?” Colacurcio says. Workers may benefit from the additional output their agents produce. But that doesn’t always translate into greater recognition: Managers can devalue employees’ contributions when they know AI was involved.

Companies are largely layering agents onto existing organizational structures and expecting workers to absorb the additional responsibility without much preparation or recognition, she says. After recently asking compensation leaders at large companies how they are addressing the shift, Colacurcio heard responses across the spectrum: Some are exploring ways to reward employees who manage agents effectively while others have barely considered the question. One adjacent effort: Ernst & Young is giving out $100 million in bonuses for employees who demonstrate the human skills needed to make AI work.

If companies begin rewarding agent management, Colacurcio expects the money to come through bonuses and other incentives rather than higher base pay.

“Managing an agent is its own skill set,” Colacurcio says. “Companies aren’t training for it or paying for it yet. And I think they should be.”

For now, workers are getting more responsibility — and little else.