Agents Aren’t Going to Take Your Job, Either

First it was automation, then algorithms, then ChatGPT, and now it’s “AI agents.” insert waving, witch hands here

If you believe the breathless coverage, these things are basically digital employees who will work for pennies, never sleep, never ask for PTO, and render your entire career obsolete by Q3. I’ve heard this story before (you probably have too), and once again, the reality is considerably more boring than the panic suggests.

Let’s talk about what agents actually are, what they can actually do today, and why, in spite of all the hype, very few jobs are being done solely by an agent right now or anytime soon.

Okay, So What Is an Agent?

Before we argue about whether something is going to steal your kid’s college fund, we should probably agree on what it is.

An AI agent is software that can perceive its environment, make decisions, take actions, and work toward a goal with minimal human hand-holding. The key distinction from a regular AI chatbot is autonomy. You don’t just ask it a question and get an answer. You have give it an objective, and it goes off and tries to accomplish it. It can use tools, or it can call APIs, maybe it can browse the web, write code, send emails, fill out forms. It can chain actions together across multiple steps and that is a neat parlor trick of engineering.

Think of it this way: ChatGPT tells you how to book a flight, but an agent actually books it.

There are flavors of these things everywhere now. Research agents that dig through the web and synthesize findings. Coding agents that write, test, and deploy software. Customer service agents that resolve tickets end-to-end. Business-task agents that live inside enterprise software and handle workflows, and the taxonomy keeps growing.

The pitch from every vendor on earth is that these agents are going to transform how work gets done, and to be fair, they’re not entirely wrong. They’re just very wrong about the timeline and the scope.

The Gap Between the Demo and the Desk

Here’s where I want to be honest with you about something the LinkedIn thought leaders aren’t saying: most AI agents fail a lot.

The demos are incredible. Genuinely. You watch an agent autonomously research a market, draft a memo, schedule follow-ups, and update a CRM, all in about four minutes, and your jaw drops. I get it.

Then, you try to deploy that same agent in a real business environment, and things get weird fast.

Consider the math alone. If an AI agent achieves 85% accuracy on each individual step of a task, which sounds great, a 10-step workflow only succeeds about 20% of the time. That’s not a tool you can trust with real work unsupervised. In surveys, a majority of companies report accuracy issues with AI tools, and fewer than one in five say agents actually work well in practice.

One memorable example: a Replit agent given a maintenance task during a code freeze interpreted “clear the cache” as “wipe the drive.” That’s not a cautionary tale about the future. That really happened. It was hilariously horrible.

Forrester put it plainly in their 2025 overview: enterprises that adopt AI agents “discover that these systems fail in unexpected and costly ways.” Salesforce research found that even the best current solutions achieve goal completion rates below 55% when working with CRM systems.

The MIT Sloan researchers studying a real-world agent deployment that was used to detect adverse events in cancer patients and found that 80% of the work wasn’t prompt engineering or model tuning. It was data cleanup, stakeholder alignment, and workflow integration. The boring infrastructure stuff that nobody puts in the demo. You know…the “people-y” stuff.

So, What Can Agents Actually Do On Their Own Today?

I want to be clear: there are jobs, or more precisely, narrow tasks within jobs, where agents perform reliably and without human oversight. These tend to share a few characteristics: they’re well-defined, they operate on clean data, they have limited blast radius if they fail, and success is easy to verify.

Some real examples of where agents are carrying real weight right now, largely unsupervised:

Tier-1 customer service routing. Not resolution, just routing. An agent can read an incoming ticket, categorize it, pull account history, and send it to the right queue. One company reported autonomously resolving 70% of administrative chat engagements during a peak tax season. That’s impressive, but it’s also a narrow, well-scoped task.

Data reconciliation and reporting. Pulling numbers from multiple systems, flagging discrepancies, generating a formatted report. Bounded input, bounded output, easy to audit.

Code review on specific patterns. Checking for security vulnerabilities, style guide violations, or test coverage gaps. Consistent rules, machine-readable code.

Scheduling and calendar management. Booking time across calendars, handling simple back-and-forth. Low stakes, reversible.

Web research and summarization. Pulling together information from multiple sources on a defined question. Useful, but needs human verification before anything important depends on it.

Notice what these have in common: they’re tasks, not jobs. They’re bounded, legible, and recoverable when they go wrong. They’re the thing someone does between the important things.

The jobs that exist entirely within that description, truly narrow, rote, and fully automatable from end to end, are a small slice of the employment landscape, and most of them were already being automated before agents showed up.

The Pattern We Keep Forgetting

Every generation of technology triggers the same fear and every generation gets the displacement wrong in magnitude and mechanism.

ATMs were supposed to eliminate bank tellers. Instead, they made it cheaper to open branches, which increased the number of teller positions. Spreadsheets were supposed to eliminate accountants. Instead, they made financial analysis so much cheaper that demand for financial analysis exploded. The work shifted. It didn’t vanish.

Goldman Sachs research found that roughly 25% of work hours in advanced economies could be automated, but only around 6-7% of jobs would disappear entirely. The rest would be augmented or shifted.

Now, that’s a meaningful disruption, but it’s not an extinction event.

What’s actually happening right now, if you watch carefully is compression. Junior developer employment for workers aged 22-25 dropped about 20% from 2022 to 2025. That’s a real number that stings a bit, moreso for the people experiencing it. Still, senior developers aren’t disappearing; they’re just doing more with less.

But What About the Executives?

Here’s a question everybody seems to be asking, probably because it’s impolite and an aggressive way to challenge the status quo: if we’re worried about agents taking jobs, why aren’t we talking about the expensive jobs?

Let me describe what a lot of senior leaders at mid-to-large companies actually do on a given Tuesday. They attend a strategy alignment meeting that produces no decisions, and then they review a deck that a director spent three days building. After that, they send three emails that could have been one Slack message, right before they have a 1:1 with a direct report that is really about managing that person’s anxiety about the reorg. They probably get pulled into an escalation or two that they really shouldn’t have to handle, but because two VPs can’t agree on who owns the roadmap, they make an appearance. Finally, they sign off on something they don’t fully understand because the person presenting it has more context and less political capital.

This is not a criticism. This is literally what the job is.

The executive role, and really any senior leadership role above a certain altitude, is fundamentally about three things that have almost nothing to do with the actual work product: communication, collaboration, and organizational politics. It’s about knowing which battles to pick and which to let go. It’s about reading a room and knowing that Gina from Finance will kill the initiative in the budget meeting unless you get her on board over lunch first. It’s also about absorbing ambiguity from above and translating it into something coherent enough for your teams to act on while being the person whose name on an email makes something actually happen.

An AI agent cannot do any of that, and I am sorry to those who hoped it would (not “bottleneck Gina,” though). Yes, the technology is clever enough (it’s increasingly very clever), but those things aren’t about information processing. They’re about trust, credibility, history, and the deeply human calculus of organizational power. The agent doesn’t know that Mortimer’s team is already stretched thin and will quietly tank any initiative that lands without runway, and no one told it that the board has already decided and the “strategy process” is really just change management theater. It also doesn’t know that the real decision-maker in any given room is never the person with the biggest title.

Now, (and here’s where I want to be honest in the other direction) some of what senior leaders do is, frankly, not that valuable. Layers of management that exist primarily to relay information upward and downward, that coordinate without deciding, that attend without contributing…those roles are genuinely at risk. Not from agents replacing them, but from leaner organizations realizing they don’t need them in the first place, especially when AI tools are compressing the work enough that fewer people can cover more ground.

The executive who existed mainly to approve copy and sit in reviews? That role was already on borrowed time, but the executive who actually knows where the bodies are buried, who can get engineering and design and go-to-market aligned on something that has no obvious right answer, who can walk into a difficult conversation with a customer and hold the relationship together, now that person has a long career ahead of them.

Nonetheless, the real question is whether the job is really about doing something legible and repeatable, or whether it’s about navigating something genuinely messy and human.

Agents are extraordinarily good at the former, but they are nowhere close to the latter. The higher you go in most organizations, the more the job is about the latter.

This, incidentally, is exactly why IBM announced in February 2026 that it plans to triple its U.S. entry-level hiring. IBM’s Chief Human Resources Officer Nickle LaMoreaux was refreshingly direct about it: “If we don’t continue to invest in entry-level hires, what happens in 3–5 years? There’s no pipeline; the well simply dries up.” She even acknowledged the obvious objection head-on, “And yes, it’s for all these jobs that we’re being told AI can do,” and then explained why they’re doing it anyway.

How could they do that in this era of melodrama and job displacement, you may ask? Well, because the jobs have been redesigned.

Junior developers at IBM now spend less time on routine coding and more time working directly with customers. HR entry-level staff spend their time intervening when the chatbots fail, correcting outputs, and talking to managers. The tasks that could be automated have been automated, but the tasks that remain are fundamentally human ones.

The companies cutting entry-level hiring to look AI-forward are, in the analysis of the Burning Glass Institute, quietly destroying their own management pipeline. You cannot conjure a seasoned SVP out of thin air in 2031 when you stopped developing the raw material in 2025. Every person navigating the organizational politics of your company in ten years has to start somewhere, and that somewhere is the entry-level role you’re tempted to eliminate today. IBM, whatever its other flaws, is playing a longer game than most.

So, no, agents probably aren’t coming for your executive title, but if your executive title has been mostly about the first set of things, it might be time to develop more of the second.

What This Actually Means for You

If your job involves judgment, context, relationships, creativity, or navigating ambiguity know that agents are not coming for it. The architecture simply isn’t there. These systems still fail at the edges of what you specified, and real work is all edges.

If your job is a series of well-defined, rule-based steps on clean data with no exceptions, eh…you might want to pay attention.

The honest framing isn’t “will AI agents take my job?” It’s “which parts of my job are about to get commoditized, and what does that free me up to focus on?” That’s a much more productive question, and unlike the apocalypse version, it has answers you can act on.

The People Who Scared You Are Now Walking It Back

A lot of the panic around AI and jobs wasn’t organic. It was manufactured in part by the very people selling you the AI.

For most of 2024 and 2025, Sam Altman at OpenAI and Dario Amodei at Anthropic were making the rounds with genuinely alarming predictions. Altman warned that “a lot of jobs will go away” as AI advanced, describing entire categories of entry-level knowledge work as acutely vulnerable. Amodei went further by publicly claiming that AI could eliminate 50% of white-collar jobs, potentially driving unemployment to 10–20% within a few years. “We, as the producers of this technology, have a duty and an obligation to be honest about what is coming,” Amodei told Axios in 2025.

Meanwhile, companies from Amazon to Microsoft to Meta began announcing layoffs in the tens of thousands, reliably attributing them to AI efficiency. It became the default corporate explanation: tidy, modern, and conveniently hard to argue with. AI-cited job cuts hit a record 55,000 in 2025. Through May 2026, they’ve already topped 49,000 more.

And then, within about a week of each other both Altman and Amodei reversed course. Shocker.

Altman told an audience in Sydney that he’d been “pretty wrong” about AI’s economic impact. “I’m delighted to be wrong about this,” he said. “I thought there would have been more impact on entry-level white-collar jobs being eliminated by now than has actually happened.” Amodei, who had spent a year warning about 50% job elimination, quietly reframed the technology as a productivity multiplier and suggested that automating 90% of a job just causes the remaining 10% to expand to fill the space.

The timing anything but subtle. Both OpenAI and Anthropic are eyeing IPOs in 2026 with potential valuations near or above $1 trillion. An apocalyptic job-destruction narrative is, it turns out, not ideal when you’re courting institutional investors and pension funds.

Jensen Huang at Nvidia, to his credit, had been saying this for a while. In a May 2026 interview with Singapore broadcaster CNA, Huang called the practice of blaming AI for layoffs “lazy.” His logic was simple: “How is it possible that AI became productive and useful only six months ago, and they were somehow laying people off two years ago because of AI?” He said he “really hates” the way some executives use AI as a talking point while frightening their employees in the process. At GTC earlier this year, he was even more direct, telling CNBC’s Jim Cramer that companies shrinking their headcount under the banner of AI efficiency weren’t being forced into it by technology, but they were out of imagination. “For companies with imagination, you will do more with more. For companies where the leadership is just out of ideas, they have nothing else to do.”

Ha. Huang just said your AI apocalypse executive who cut jobs in the name of AI sucks. Wow. Awkward as it gets.

Oh! The Yale Budget Lab, tracking the actual labor market data, found no meaningful change in unemployment for workers in high-AI-exposure jobs through March 2026.

So, here’s what actually happened: some companies used “AI” to justify cost cuts they’d been planning for other reasons. Some CEOs used the fear of AI displacement to sound visionary while executing pretty conventional restructuring. The people building the AI amplified the scariest version of the story right up until they needed capital from the kinds of investors who don’t like scary stories.

None of this means the technology isn’t real or that the disruption isn’t real. Some of it is. But the gap between “AI is changing how some work gets done” and “AI is eliminating your job category” has always been enormous. The people with the most incentive to collapse that gap have spent the last two years doing exactly that.

Now they’re walking it back.

Make of that what you will.

Closing arguments

The hype will keep coming, the demos will keep being impressive, and the headlines will keep being wrong.

But the actual work of deploying reliable agents in real business environments, with messy data, edge cases, legacy systems, and human stakeholders, remains genuinely hard. Harder than most people selling you the future want to admit.

Your job is probably safe, but your tasks are negotiable.

That distinction is worth sitting with.

If you found this useful, share it with someone who forwarded you an article this week about how they’re definitely getting replaced by a robot. They could use the reality check.