We are used to thinking of people as the ones who tell machines what to do. Now that we are safely ensconced in the Intelligence Age , that dynamic is starting to flip: AI is now assigning tasks for people to carry out. Author Cory Doctorow calls this arrangement a reverse centaur . But first, we need to define the underlying concept. Research from MIT Sloan finds that “centaurs maintain structured and controlled interactions with AI, harnessing it as a tool for targeted efficiency.”

For centaurs, the delegation arrow points one way: humans tell AI what to do, usually via prompts. We direct machines to work for us, whether that means coding a website, summarizing a meeting or completing a task. Now that we are several years past ChatGPT’s arrival which kicked off widespread artificial intelligence adoption, the arrow is starting to reverse course.

Introducing Ambient Delegation

To help make sense of the shift, I sat down with Chaser CEO Josh Martow. Formerly director of business intelligence at Thriver, he now leads a task management platform built directly into Slack that reports more than one million users. “The assignment of work is going to become ambient ,” he told me. “Now that AI agents can perceive enough of what’s going on in the background of a company they can connect dots to suggest what must be done.”

To appreciate the implications, imagine the following situation at a rental-car company. A customer reports a problem: they were given a vehicle with a faulty transmission that broke down several miles from the airport. In the ambient-delegation scenario, an AI agent would receive the alert first. Then, using a Slack channel or another task-management service, it would inform a call-center employee of the next steps for getting the customer a replacement so they can quickly get back on the road.

Anticipating this sea change in the future of delegation, an MIT Technology Review Insights article produced in association with the Deloitte Microsoft Technology Practice suggests “enabling agent-first process redesign” in company workflows. It also argues for embracing new ways of working so organizations are not left vulnerable to competitors that may already be several steps ahead in AI adoption. “Unlike static, rules-based systems, AI agents can learn, adapt and optimize processes dynamically. As they interact with data, systems, people and other agents in real time, AI agents can execute entire workflows autonomously.”

A recent Anthropic release speaks to the utility of collaborating with AI in a way that makes it more of a partner in workflows, if not yet the delegator. “Claude Tag is a new way for teams to work with Claude,” the company’s release states. “… Grant Claude access to selected channels, and connect it to whichever tools, data—and even codebases—you choose. Then, anyone in the channel can tag @Claude in, and delegate tasks to it while they focus on other work. Claude builds context by remembering relevant information from the channels it’s in and can plan out tasks to complete in the future.”

Programming the AI Manager

Chaser’s Model Context Protocol integration speaks to this idea. Beyond creating formal tasks within Slack, it enables AI agents to act in a project-managerial role: assigning owners and tasks, setting deadlines, and documenting each assignment for visibility and follow-up. There is a major difference between this behavior and a chatbot merely offering a recommendation in a chat box.

As Martow sees it, AI is becoming an integral part of workflows, specifically directing human attention. Moving forward, we can expect major vocational and even cultural shifts as the nature of business operations transforms. “At the end of the day, what you're effectively doing is programming a manager at your company,” he explains. “You're deciding how a synthetic manager should operate and giving it the ability to make real strategic decisions.”

That reality is not without challenges, as Martow also points out. “If you're an employee, you don't have control over how the CEO set that AI manager's direction. You're now in a situation where an AI is effectively telling you what to do and you're hoping the black box operating above you has your best interests, and the company's, at heart.”

Founder and CEO Flo Crivello has been following this subject closely. His company, Lindy , is also in the business of improving how work gets done. Its AI executive assistant can connect to your email inbox on your behalf, supporting you in much the way a human EA might, whether that means drafting messages in your voice, scheduling meetings, summarizing chats or preparing you for calls.

He views today’s developments as a continuation of what he witnessed as Head of Product at Uber Works during his five years with the company. “Staff would talk about the ‘API line,’ he told me via interview. “People above it, such as corporate employees, would tell the algorithm what to do. People below it, such as drivers, were directed by the algorithm.”

The rise of AI agents is an extension of this model, affecting how work gets done across more companies and sectors, a phenomenon Crivello dubs the “AI sandwich.” Under this arrangement, people are still in the driver’s seat, no pun intended. They design the system and create its goals. AI comes next, relying on those instructions to delegate tasks and assignments to other humans farther down the organizational chart.

What About Middle Management?

Fans of the series The Office will recall Michael Scott (played by Steve Carell) as perhaps the archetypal middle manager. As the regional head of Dunder Mifflin’s Scranton branch, his role is to supervise employees all day so that they can produce and sell paper. Based on the developments Martow and Crivello describe, we have to wonder whether the version of work dramatized on this beloved TV show is heading for extinction, much as the dispatch-driven cab company depicted in another television hit, Taxi , now feels dated.

Of course, competent middle managers, not the beloved buffoon Michael Scott, do far more than distribute assignments. They coach employees, resolve conflict, translate strategy and exercise judgment. Still, we must wonder how much of their traditional purview AI may soon be able to assume. Crivello has strong opinions on the matter. “I don’t think middle management should keep existing,” he said. “It served a coordination function that is now better served by AI.”

Whether or not that conclusion proves too sweeping, we must acknowledge that the nature of work is changing rapidly. Even so, Martow takes a more human-centered view. “I do want to push back on the idea of AI being the leader, because it’s the humans that are deciding the strategic direction of any organization and what’s to be done. Even in a delegator capacity, the AI agents are following the instructions of humans.”

Reflecting on where work is going, it is clear that for years people have worried about which tasks AI might perform for us and which jobs might disappear as a result. Perhaps the next question we should consider is different: What will it mean for tomorrow’s workforce when AI starts handing those tasks right back?