Employers cited AI in 101,743 announced U.S. job cuts in the first six months of 2026, roughly 23% of all cuts this year, according to Challenger, Gray & Christmas. That is nearly double the 54,836 AI-attributed cuts recorded across the whole of 2025, and AI has now been the single most cited reason for U.S. layoffs four months running. Challenger calls it a structural shift its data has never recorded before.
That is the scary half of the story, and it is real. The other half is that Gartner cannot find any link between cutting staff and actually making money from AI, Gallup found that only 1% of laid-off workers blame AI for losing their job, and Anthropic’s own economists see limited evidence AI has moved aggregate employment at all. This article gives you both halves, organized by job category, with every number sourced. It is about what AI agents genuinely change about the work you do, and what you can do about it in the next thirty days.
The Key Takeaways
- As of Challenger’s June report, employers cited AI in 101,743 announced U.S. job cuts in H1 2026, nearly double the 54,836 logged in all of 2025. AI has led the stated reasons for four straight months.
- But the money is not there. Gartner found roughly 80% of large enterprises deploying autonomous AI cut staff, with no correlation between those cuts and AI return on investment.
- Only 1% of laid-off U.S. workers cite AI or automation as the primary cause of their job loss (Gallup, June 2026). The layoffs are real; the attribution is looser than the headlines suggest.
- The damage is concentrated at the bottom. Employment for software developers aged 22 to 25 has fallen nearly 20% from 2024, according to the Stanford AI Index 2026. Separately, in a survey of business leaders, 21% said they have already stopped hiring entry-level workers because of AI.
- Gartner now projects AI will create more jobs than it eliminates beginning in 2028. The squeeze is the transition, not the destination.
The Numbers (What the Data Actually Says)
Before getting into specific jobs, here is the scale and the pace. Every number below is sourced, and where a figure is a projection rather than a measurement, it says so.
| Stat | Source |
|---|---|
| 101,743 U.S. job cuts citing AI in the first half of 2026, about 23% of all announced cuts | Challenger, Gray & Christmas, June 2026 |
| 54,836 AI-attributed U.S. cuts across the whole of 2025, against 71,825 cumulative since 2023 | Challenger 2025 year-end report |
| 51% of companies say they will lay off staff in 2026 because AI is consolidating roles; 21% say AI is the sole reason | ResumeTemplates.com survey of 933 business leaders, February 2026 |
| 21% have already stopped hiring entry-level workers because of AI; 36% expect to by the end of 2026 | ResumeTemplates.com, February 2026 |
| Roughly 80% of large enterprises deploying autonomous AI reduced headcount, with no correlation to AI ROI | Gartner, May 2026 |
| 170 million new jobs projected by 2030 against 92 million displaced, a net gain of 78 million | WEF Future of Jobs Report 2025 |
| 40% of enterprise apps will feature task-specific AI agents by the end of 2026, up from under 5% in 2025 | Gartner, August 2025 |
| Over 40% of agentic AI projects will be cancelled by the end of 2027 | Gartner, June 2025 |
The paradox: companies are cutting staff for what AI might do, not for what it has actually delivered. That is the argument of a January 2026 Harvard Business Review piece by Thomas Davenport and Laks Srinivasan, Companies Are Laying Off Workers Because of AI’s Potential, Not Its Performance, built on a survey of 1,006 global executives. Their warning is blunt. The strategy sours employees on AI, breeds cynicism, and ends in embarrassing rehiring retreats.
The technology is running ahead of the implementation, and that gap is your window. The people who learn to work effectively with AI agents now will be the ones still standing when organizations finally work out what actually delivers.
The Case Against the Panic
The layoff numbers are the ones that travel. The numbers that follow rarely do, and they complicate the story considerably. Both sets are true at the same time, which is uncomfortable but important.
The ROI Never Showed Up
Gartner surveyed 350 executives at companies with at least $1 billion in revenue that had deployed autonomous AI, publishing the results in May 2026. Around 80% had cut headcount. The cuts were almost identical between companies seeing strong AI returns and companies seeing negative ones, which is another way of saying the layoffs had nothing to do with whether the AI worked.
Gartner’s Helen Poitevin put it in one line. “Workforce reductions may create budget room, but they do not create return.” If you are being told your role is going because AI has made it redundant, the evidence says the AI probably has not, yet.
Gartner also expects the direction to reverse. Its HR research, published in May 2026, projects that AI will create more jobs than it eliminates beginning in 2028, with a broadly neutral net effect through 2026. The same research expects over 32 million jobs a year to be significantly transformed along the way, which is the part that should actually concern you.
Only 1% of Laid-Off Workers Blame AI
Gallup fielded a survey in February 2026 and published it in June, covering 23,717 employed U.S. adults plus a separate group of roughly 660 people who had been laid off. Among that laid-off group, just 1% cited AI or automation as the primary cause of losing their job. Restructuring, cost-cutting and weak demand are what people actually said. Be careful how much weight you put on that 1%, because 1% of 660 people is a handful of respondents answering an open-ended question, and it tells you what workers were told, which is not always what happened.
Gallup did find one pattern worth your attention. Laid-off workers were more likely to be AI non-users (62%) than workers who kept their jobs (50%). In tech specifically, Gallup’s model put the predicted likelihood of being laid off at 18% for workers using AI less than monthly, against 6% for those using it at least monthly, roughly triple the risk. Those are modelled probabilities rather than headcounts, and the pattern is a correlation rather than proof of cause, though it does survive controls for age, education and industry. Read it as a signal about who looks valuable, not a mechanism.
Those two numbers look like they fight each other. They do not. Challenger counts announced cuts, and announcements skew heavily toward large public employers. Measured against the roughly 10 million Americans who actually lost a job in the first half of 2026, 101,743 is itself about 1%. The Challenger figure and the Gallup figure are the same story seen from opposite ends. Employers say “AI” to investors; workers get told “restructuring”.
Anthropic’s economists reach a similar place from a different direction. Their labor market research, published March 2026, finds “limited evidence that AI has affected employment to date” and no systematic rise in unemployment among highly exposed workers since late 2022. They do flag one exception, and it is the one that keeps recurring in this article. Hiring into exposed professions has slowed for workers aged 22 to 25.
How AI Agents Are Changing Specific Jobs
For each category below, here is what agents can genuinely do now, what has already happened, what humans still do better, and what you should do about it. If you are unsure whether the thing your company is deploying is a genuine agent or just a rebranded workflow, we untangle that in agentic AI versus chatbots versus automation.
Knowledge Workers (Analysts, Researchers, Writers)
What agents can do now: research synthesis across hundreds of sources, first-draft report generation, data analysis and visualisation, meeting summarisation, email triage, and competitive intelligence gathering.
What has already happened: Chegg cut 45% of its workforce (388 people) in October 2025, citing the new realities of AI alongside collapsing Google traffic. Combined with an earlier round, it shed more than half its staff in under six months. The stock has lost about 99% of its value from its 2021 peak, and the company only clawed back NYSE minimum-price compliance in June 2026 after a brush with delisting. Students stopped paying for homework help because a chatbot answers for free.
What humans still do better: original analysis that connects ideas nobody has connected. Nuanced judgment about what matters and what is noise. Relationship-based insight, the kind you only get from knowing the people behind the data. Creative vision that produces genuinely new frameworks rather than remixes.
Practical advice: use agents to handle the first 80% of information gathering. Your value moves from gathering information to interpreting it, deciding what it means and persuading people to act on it. If most of your day is collecting and organising, that part of your role is shrinking. If most of it is judgment calls and persuasion, you are in a far stronger position than the headlines suggest.
Developers and Engineers
What agents can do now: write, debug, test and refactor code, and increasingly ship it unsupervised. SemiAnalysis reported in February 2026 that Claude Code was authoring about 4% of all public GitHub commits, and projected that figure would pass 20% by the end of 2026. Note the qualifier, because it gets dropped constantly. That is 4% of public commits, not all code written everywhere.
The prediction, and what actually happened: in March 2025, Anthropic CEO Dario Amodei said AI would be writing 90% of code within three to six months. It was not, industry-wide. But inside Anthropic the trajectory is steep and documented. Anthropic’s own figure is that more than 80% of the code merged into its codebase was authored by Claude as of May 2026, up from low single digits before Claude Code launched. An independent check by Redwood Research in October 2025 was more sceptical, putting merged lines of code closer to 50%. Those are two different measures from two different counters, so read them as dated snapshots rather than a clean growth curve. The direction, though, is not in dispute.
Amodei has not softened on code. At Davos in January 2026 he suggested AI could do most, maybe all, of what software engineers currently do within six to twelve months. He has softened on jobs. Fortune reported in May 2026 that both he and Sam Altman were walking back their jobs-apocalypse rhetoric ahead of expected IPOs, with Amodei reframing automation as a multiplier. “If you automate 90% of the job, then everyone does the 10% of the job, and the 10% kind of expands to be 100% of what people do.” Keep those two claims separate, because conflating them is how most coverage gets this wrong.
Boris Cherny, who created Claude Code, told Fortune in June 2026 that he had not written a line of code by hand in about eight months. Earlier that year he posted that he had shipped 22 pull requests one day and 27 the day before, each entirely written by Claude. Those were two specific days, not a sustained rate, but they tell you where the ceiling now sits.
The cost is landing on juniors. The Stanford AI Index 2026 found that employment for software developers aged 22 to 25 has fallen nearly 20% from 2024. Senior developers are not seeing the same decline. The job is not disappearing; the entry ramp is.
Practical advice: learn to review and direct AI-generated code. Writing clear specifications, evaluating output quality and catching the subtle errors agents miss is becoming its own discipline, and tools like Claude Cowork and ChatGPT Work are built around exactly that loop. Spend a week using an agent for real work, not experiments. The developers who thrive will be the best evaluators and architects, not the fastest typists.
Customer Service and Support
What agents can do now: handle a genuinely large share of inbound volume. Klarna’s AI assistant handles about two-thirds of all customer inquiries and does work the company equates to 853 full-time staff.
What actually happened, in full: Klarna’s headcount roughly halved, from about 7,400 in 2022 to around 3,000. Most of that came through a hiring freeze and attrition rather than AI layoffs, which is a distinction almost every retelling drops. Then, in May 2025, CEO Sebastian Siemiatkowski started hiring human agents back. His words: “As cost unfortunately seems to have been a too predominant evaluation factor when organizing this, what you end up having is lower quality.” Klarna’s customer service and operations costs then rose to $50 million in Q3 2025, up from $42 million a year earlier, even as the company booked around $60 million in AI-driven savings. The AI genuinely worked. Leaning on it too hard still cost them.
Salesforce went further and has not walked it back, though it disputes the framing, saying it rebalanced and redeployed support staff into sales rather than simply cutting them. It cut 4,000 customer support roles, taking the team from 9,000 to about 5,000, with Marc Benioff saying plainly that he needed fewer heads. On its May 2026 earnings call the company said Agentforce had autonomously handled 4 million customer inquiries in 15 months, roughly double the volume its human reps handled on the same channel.
What humans still do better: empathy in escalations. Genuinely angry or distressed customers. Edge cases with no script. High-stakes trust, meaning financial disputes, medical worries, legal problems. Klarna is the proof. Agents absorb volume well; quality and nuance stayed human, and the company paid to learn it.
Practical advice: move toward training and supervising agents, owning the escalations they cannot handle, and managing ongoing relationships. The person who can teach an agent to handle a difficult scenario, and catch it when it gets that scenario wrong, is worth considerably more than the person processing routine tickets by hand.
Marketing and Creative Professionals
What agents can do now: generate content drafts at scale, analyse campaign performance, build decks, produce design variations for testing, draft social posts and ad copy, and synthesise competitor research.
What humans still do better: brand voice, the tonal consistency that makes a brand feel like a person rather than a committee. Creative direction, meaning knowing which idea to back and which to kill. Emotional resonance, the gap between technically correct and genuinely moving. Strategic positioning.
Practical advice: become the creative director who directs agents, not the producer hand-executing every deliverable. Your taste and judgment appreciate in value as production costs fall toward zero. The marketer who generates 50 ad variations and picks the three that will land beats both the one who hand-made a single ad and the one who shipped all 50 without curating.
Middle Managers
What agents can do now: status reporting, scheduling, data aggregation, progress tracking, meeting summaries, KPI dashboards, workflow coordination. A large share of what middle management does is collecting updates, synthesising them and passing them upward, and agents do that faster and more consistently.
The clearest real-world case: Cloudflare cut 1,100 people, about 20% of its staff, in May 2026, the largest reduction in its history, in a quarter when revenue grew 34%. CEO Matthew Prince said the vast majority of those laid off were what he called “measurers”, which he defined as middle management, finance, legal, internal auditing and revenue recognition. His reasoning was that AI can now measure an organization with a precision no human layer can match.
Gartner predicted back in October 2024 that through 2026, 20% of organizations would use AI to flatten their structure, eliminating more than half of current middle management positions. Treat that as a forecast on the clock rather than a description of the present. Gartner has not renewed the prediction, its window closes within months, and nothing close to half of middle management has actually gone. The direction looks right. The magnitude does not, at least not yet.
What humans still do better: actually developing talent. Resolving conflict. Supporting people through a crisis. Cross-team coordination that runs on political skill and relationship capital. Weighing competing human interests when there is no clean answer.
Practical advice: this is, frankly, the most exposed category in the article. If your primary value is aggregating information from your team and relaying it upward, that function is being automated in front of you. Shift into coaching and relationship management, or move into the agent manager role described below, where your grasp of the business process is exactly the qualification.
Entry-Level and Early-Career Workers
The data: this is where the evidence is strongest and least ambiguous. The Stanford AI Index 2026 found employment for software developers aged 22 to 25 down nearly 20% from 2024. Separately, 21% of companies have already stopped hiring entry-level workers because of AI, and 36% expect to have stopped by the end of 2026, per ResumeTemplates.com’s February 2026 survey.
The Federal Reserve Bank of Dallas adds precision, and the precision matters. Among young labour-market entrants aged 20 to 24, the rate of finding work in the most AI-exposed occupations has fallen more than 3 percentage points from its November 2023 peak, while holding steady in low-exposure jobs. Its most-exposed group is retail supervisors, admin assistants and customer service reps rather than developers, so this is a separate signal pointing the same way, not the same finding counted twice. Note what it is and is not. Young people already unemployed are not finding it harder than anyone else. The Dallas Fed is explicit that layoffs do not appear to be the cause. The door is not pushing people out; it is opening less often.
The honest caveat: none of this is a clean AI experiment. Junior hiring also fell because the 2021 tech over-hiring boom unwound and rates rose, and the decline runs from a late-2022 peak, before agents were deployed anywhere. Stanford says as much itself. AI is not the only thing pressing on the entry ramp. It is just the newest thing pressing on it.
The paradox: agents eliminate exactly the routine tasks that used to be how juniors learned the job. A first-year analyst who once spent months hand-building financial models, and learned how models work in the process, now generates one in minutes. The work gets done. The learning does not. Organizations are buying short-term efficiency with their own talent pipeline.
What to do: build the skills that complement agents instead of racing them. Judgment, meaning the ability to tell when an output is wrong. Communication, meaning presenting and defending a recommendation. Ambiguous problems with no clean answer. Push to be in the loop as a reviewer, not just a task executor, and be explicit with your manager that you want the why behind the work, not only the what. This group needs the most practical guidance and reliably gets the least.
Freelancers and Independent Workers
The double edge: freelancers can multiply their output with agents, genuinely doing the work of two or three people, or be cut out entirely by clients who use agents directly. A company that used to pay a freelance writer $500 an article can now generate drafts itself and pay an editor $150 to fix them. Or skip the freelancer altogether.
The shift: from selling hours and deliverables to selling judgment, quality control and domain expertise. The designer who delivers 20 agent-generated logo options plus a clear rationale for the top three is selling something different, and arguably more valuable, than the one who hand-crafts a single option.
Practical advice: use agents to raise both output and quality, then position yourself as the human who guarantees the result meets a professional standard, because most raw AI output still does not. Price on the value of the outcome rather than hours spent. And understand your clients’ businesses deeply enough that an agent without that context cannot replace you.
The New Job: AI Agent Manager
In February 2026, Harvard Business Review published To Thrive in the AI Era, Companies Need Agent Managers, by Suraj Srinivasan of Harvard Business School and Vivienne Wei, COO of the Agentforce platform at Salesforce. It names a role that has been forming for about a year. Worth noting the co-author works at Salesforce, so read the Salesforce examples with that in mind, but the underlying role is showing up across the industry regardless.
What agent managers actually do: define tasks and performance metrics for agents, review and evaluate what they produce, handle the exceptions where the agent fails, adjust configuration to improve workflows, and watch for bias, accuracy drift and policy violations. Somebody has to be accountable for what the agents do.
The skills are less exotic than the title suggests. You need to write precise instructions that produce consistent output. You need to break a business workflow into agent-sized tasks with clear success criteria. After that you need to evaluate output critically and recognise failure patterns. Then you need to judge where an agent’s mistake is cheap (a rough email draft) and where it is expensive (approving a payment). And you need enough domain expertise to know whether the work is actually any good.
Who is best positioned: project managers, team leads, operations specialists and quality analysts, anyone who already manages a process. HBR is explicit that domain expertise matters more than AI expertise here. The best agent managers will come from the roles that already understand the process being automated, which is precisely why exposed middle managers should be reading this section twice.
The analogy: “social media manager” did not exist in 2005. By 2015 it was standard at every mid-sized company. Agent manager is on the same curve, compressed. We are somewhere around the 2008 equivalent, where early adopters are hiring and the mainstream is about to.
The AI Burnout Problem Nobody Is Talking About
While everyone argues about displacement, a quieter problem is landing on the people who keep their jobs. AI is not making work lighter. It is making it denser.
Berkeley researchers Aruna Ranganathan and Xingqi Maggie Ye spent eight months, from April to December 2025, embedded inside a 200-person tech company, running more than 40 interviews alongside on-site observation, and published the results in HBR as AI Doesn’t Reduce Work, It Intensifies It. What they found was task expansion. Workers moved faster, took on a wider scope of tasks, and pushed work into more hours of the day. Nobody instructed them to. It happened on its own, and it produced cognitive fatigue and weaker decision-making.
The mechanism is simple and you have probably felt it. If you can now draft a report in two hours instead of eight, the expectation quietly resets to four reports rather than one. The tool did not reduce your workload; it raised the baseline. Add the pressure to keep learning new tools that change every few months, on top of everything you already owe, and the phrase “AI won’t replace you, but someone using AI will” stops sounding like motivation and starts sounding like a threat.
This is one company, and it should be read as a signal rather than a law. But it points at the risk the layoff headlines miss entirely. Displacement is not the only way this goes badly. Keeping your job and finding it has quietly become worse, more intense, more pressured, more exhausting, is the far more likely outcome for most people reading this.
What you can actually do: agents should be cutting your busywork, not inflating your total load. If your employer uses AI to demand double the output for the same pay, that is a management decision, not a technological inevitability. Push back on workload creep and be explicit about how the productivity gain gets shared. If you manage people, understand that using AI to justify a relentless pace is the fastest way to lose the ones you cannot replace.
5 Practical Steps to Take This Month
Not “upskill” platitudes. Five specific things you can do in the next thirty days.
1. Use an AI Agent for Your Actual Work This Week
Not playing around, and not generating a funny poem. Take one real task from your job, researching a topic, building a report, cleaning a dataset, drafting customer communications, and hand it to an agent end to end. Then grade the result honestly. What was good? What needed fixing? Where did it save you time and where did it waste it? Our roundup of the best AI agents is a reasonable starting point if you do not know which to try.
This matters because experience is the only real preparation. Reading about agents is not the same as using them. You need an instinct for where they are strong and where they quietly fail, and that instinct only comes from real work.
2. Learn to Write Good Instructions
Prompting is the new professional literacy. Give an agent a genuinely complex, multi-step task, something like researching five competitors, comparing their pricing and drafting a one-page recommendation, then refine your instructions until the output is actually usable. Notice what makes the difference: specificity about format, examples of what good looks like, explicit success criteria.
The gap between a vague prompt and a precise one is often the gap between useless output and three hours saved. This skill is also the core of the agent manager role above, which is the most direct upgrade path available to most people reading this.
3. Identify the 20% of Your Job Agents Cannot Do
Look honestly at your last month of work. Which tasks could an agent do with light supervision? Which ones genuinely need your judgment, your relationships, your creative input? That second bucket is where your career value is migrating, and it is where you should be spending your development time.
The common answers are complex negotiations, relationship management, ethical judgment calls, creative strategy, mentoring, and the ambiguous situations where there is no clean right answer. If your honest audit finds that bucket is nearly empty, that is not a reason to panic, but it is a reason to act this quarter rather than next year.
4. Become the Human in the Loop
Position yourself as the quality layer between agent output and final delivery. This is structurally strong because it needs both domain expertise, so you know what good work looks like in your field, and AI literacy, so you know how agents fail and what to check. You are needed precisely because agents are not yet good enough to run unsupervised, and the Gartner ROI data suggests that will remain true for longer than the vendors imply.
5. Talk to Your Team About AI Agents Openly
The organizations handling this well are the ones where workers are part of the conversation rather than ambushed by it. If your company is deploying agents, ask which roles are affected, on what timeline, and what retraining exists. If you manage people, start that conversation before they have to. The uncertainty does more damage than the change itself, and Gallup’s data on who gets laid off suggests the people who engage with AI early are also the people who look indispensable later.
What History Tells Us (The Whole Story, Not the Comforting Half)
You have heard the ATM story. When cash machines rolled out from the 1970s onward, bank tellers did not disappear. ATMs made branches cheaper to run, so banks opened more branches, and the number of tellers per urban branch fell from about 20 to 13 between 1988 and 2004 while total teller employment held up. Tellers shifted from counting cash to selling products and advising customers. Economist James Bessen made this the canonical example of automation transforming a job rather than killing it.
Almost everybody stops the story there, and it is a comforting place to stop. It is also about fifteen years out of date. Teller employment peaked around 2010 and has fallen roughly 30% since. The Bureau of Labor Statistics counted about 347,400 teller jobs in 2024 and projects a further 13% decline through 2034, with effectively zero growth. The reason is not the ATM. It is the smartphone, which removed the reason to walk into a branch at all.
That is the real lesson, and it is a better one. The ATM automated the task, and the job survived for thirty years by shifting to work machines could not do. Then a second wave removed the context in which the job existed, and no amount of reskilling into customer service saved it. Automation rarely kills a role in one blow. It hollows it out, the role adapts, and then something else takes the ground out from under it.
Spreadsheets follow the same shape. They did not eliminate accountants; they eliminated bookkeepers and made accountants more productive. Technology tends to eliminate specific tasks, transform roles, and create categories of work nobody had names for.
The honest caveat: the difference this time is real. Previous automation waves hit physical and routine cognitive work. Agents are coming for non-routine cognitive work, analysis, writing, coding and decision support, which was supposed to be the safe ground. The pace is faster too. The distance from interesting demo to deployed in production is now months, not decades. That does not mean knowledge work disappears. It means the transition will be rougher for white-collar workers than the last few were, and the people who prepare during the ROI gap will come out of it in a much better position than the people who wait for certainty.
FAQ
Will AI agents take my job?
Probably not outright, but the honest answer depends on what your job consists of. If your role is mostly routine information processing, compiling reports, handling standard queries, writing formulaic content, parts of it are already being automated. If it runs on judgment, creativity, relationships or genuinely novel problems, you are in a much stronger position. For most people the realistic outcome is that the job changes substantially rather than disappearing. Worth noting that Gallup found only 1% of laid-off U.S. workers cited AI as the primary cause of losing their job.
How many jobs has AI actually cut?
Employers cited AI in 101,743 announced U.S. job cuts in the first half of 2026, about 23% of all cuts, according to Challenger, Gray & Christmas. That is nearly double the 54,836 recorded across all of 2025, and AI has been the single most cited reason for four consecutive months. Two caveats matter. These are announcements with a self-reported reason, not confirmed separations, and Gartner found no correlation between AI-driven headcount cuts and any actual return on AI investment.
Which jobs are safest from AI agents?
No job is untouched, but the least exposed are those requiring physical presence and dexterity (trades, healthcare), deep human relationships (therapy, sales, management), and genuine creative or strategic vision, the traits behind the most AI-proof jobs. The WEF projects the fastest-growing roles through 2030 include care workers, educators, delivery drivers and farmworkers alongside AI and data specialists. The most exposed right now are entry-level white-collar roles and middle management.
Is it harder to get an entry-level job because of AI?
The evidence says yes, and this is the clearest signal in the whole dataset. Stanford’s AI Index 2026 found employment for software developers aged 22 to 25 has fallen nearly 20% from 2024. ResumeTemplates.com found 21% of companies have already stopped hiring entry-level workers because of AI. The Dallas Fed found young labour-market entrants are finding work in AI-exposed occupations at a rate more than 3 percentage points below its November 2023 peak. Notably, this is a hiring slowdown rather than a layoff wave; the door opens less often rather than pushing people out.
What is an AI agent manager?
A role defined by Harvard Business Review in February 2026. The agent manager defines tasks and success metrics for AI agents, reviews their output, handles the exceptions they cannot resolve, and owns quality. Think project manager, but for AI workers. HBR is explicit that domain expertise in the process being automated matters more than technical AI expertise, which makes it the most realistic upgrade path for experienced middle managers and operations people.
What skills do I need to work with AI agents?
Five that keep recurring. Clear written communication, because agents follow written instructions. Critical evaluation, because you have to judge whether the output is accurate. Process thinking, because you have to break work into steps an agent can execute. Domain expertise, because you have to know what good looks like. And adaptability, because the specific tools will change every few months while the underlying skill of directing agents will not.
Should I be worried about AI burnout?
It is the risk most people are not tracking. Berkeley researchers spent eight months inside one tech company and found AI drives task expansion rather than time savings. People worked faster, took on more scope, and stretched work into more hours of the day, without anyone telling them to. If you can draft a report in two hours instead of eight, the expectation quietly becomes four reports rather than one. Agents should cut your busywork, not raise your baseline, and it is worth being explicit with your manager about which is happening.
