AI is coming for managers’ headcount

I was chairing a panel discussion about AI the other day when someone in the audience asked about something I had never thought of before.

The question went like this: the way your average manager is graded and paid often depends on how many people they manage. The higher the headcount, the higher the pay and seniority. But what happens when a manager’s human underlings are replaced by AI agents?

These agents are not standard ChatGPT, or generative AI, but systems that can beaver about doing stuff unaided, like check an invoice, browse the web or send an email.

They are already being used widely in 40 per cent of companies with $1bn or more in revenue, McKinsey says, up from 27 per cent last year. Growth is flat at smaller outfits but still a chunky 22 per cent.

So do bosses deserve as much pay for managing a bot as a flesh-and-blood human?

Having put this question to a number of HR executives and specialists last week, I can say it is a live issue at some organisations, and a tricky one, even for tech companies whose business model is based on AI.

“We are going to have to rethink definitively how people are compensated,” says Kit Krugman, chief people officer at the AI-driven supply chain platform Altana, which has nearly 300 employees.

You could argue that, if everyone in a company is managing an agent workforce, they all have management responsibilities and should be paid accordingly, she told me.

After all, some aspects of overseeing humans apply to agents as well. Both need feedback and direction, though there are obvious differences in their emotional responses to this supervision. “Humans are much more highly volatile than agents, at least so far,” said Krugman.

This is a good point. When I think back to the happily brief moments in my career when I had to manage other people, I would have said that being in charge of a robot would be a comparative cakewalk.

But the more I spoke to people about what is involved in running a team of agents, the more I realised I was wrong.

For one thing, there is what researchers call “AI brain fry”, the mental overload of trying to keep tabs on subordinates that move at warp speed and never rest.

“I’ve heard people saying engineers should be using 10 agents at once,” says Brian Elliott, a former executive at tech groups including Google and Slack who now advises on the future of work.

“That’s a massive load cognitively,” he told me, adding he has learnt to run no more than two or three agents at once. Any more is bedlam.

But the remorseless energy of agents may be the least of their management challenges, considering their propensity to go on wild hacking sprees in testing sessions. Or to hack into gyms to get their user a spot in a pilates class by kicking other people off waiting lists, as happened in Australia this year.

No wonder what is politely called AI governance is a ballooning issue, especially in the world of finance. Postings for banking jobs involving “responsible AI” have jumped almost 660 per cent from last year, according to the hiring data firm Draup.

The thing is, even well-behaved agents have problems. Instead of crashing, or displaying big error messages like other software systems, they muck up silently without ever letting you know.

That is not to say agents are useless. One executive I spoke to spelled out in startling detail how well his company’s agents were doing at tasks that ranged from forecasting demand for thousands of products, to keeping an eye on stock. The humans who had always done these tasks were still on the books, for now. But bosses were understandably keen to talk about whether their roles would be around forever.

Still, everyone I spoke to agreed on one thing. There is no way existing managers would have their pay cut if they halved their human headcount and deployed several dozen agents.

Apart from everything else, they would have saved their company a lot of money, as long as their agents did not then head off and run up massive token bills.

This is a relief, of sorts. So is the rising recognition of the complexity of managing agents. But like everything else in today’s AI world, you have to wonder how long it will last.