Vertex Tech Hub

Most of what gets called “AI replacing jobs” in 2026 is something narrower than that phrase suggests. A company automates one task inside a role. It doesn’t eliminate the role. That distinction gets lost in headlines constantly, and it’s the difference between a real trend and a claim that outran the evidence behind it.

This piece works from data that’s already been published and dated: Challenger, Gray & Christmas job-cut reports, Stanford’s ongoing tracking of entry-level hiring, the World Economic Forum’s five-year labor projection, and Anthropic’s own numbers on how people use Claude at work. None of it settles whether AI is good or bad for employment overall. What it does is separate what’s already happened from what’s still a forecast, and name the specific tools where the evidence for real substitution holds up. For the technical layer underneath these tools, our explainer on Retrieval-Augmented Generation covers how they actually retrieve and generate answers.

Five Different Things People Mean When They Say “AI Is Replacing Jobs”

A lot of the confusion here comes from treating five different situations as one. They get lumped together constantly, and that’s usually how a pilot program in one department turns into a headline about an entire profession disappearing.

• Task automation. AI handles one specific piece of a job, drafting a contract summary or writing boilerplate code, while the rest of the role stays with a person.
• Worker augmentation. A person stays fully in charge, using AI to move faster or catch more, without the job itself getting smaller.
• Reduced hiring need. A team covers the same workload, or a growing one, without adding as many new people. This is where most of the real 2026 effect actually shows up.
• Job transformation. The role survives, but what the person spends their day doing shifts, usually toward oversight and judgment calls instead of hands-on execution.
• Full role replacement. An entire position disappears because AI does the whole job end to end. This is the rarest of the five, and the one people reach for first even though it’s the hardest to prove.

Almost everything documented so far in 2026 falls into the first three. Full role replacement at scale, without a reversal a year later, is uncommon enough that the same handful of cases keep getting cited over and over. There simply aren’t that many others to point to.

What the Numbers Say, and Where They Stop Meaning What People Think

The World Economic Forum’s Future of Jobs Report 2025 surveyed more than 1,000 companies across 55 economies and produced a five-year projection: 92 million jobs displaced by 2030, 170 million created, a net gain of 78 million. People quote the 92 million figure as though it describes jobs already lost. It doesn’t. It’s a forecast for the rest of the decade.

The closer-to-real-time picture comes from Challenger, Gray & Christmas, which has tracked AI as its own stated reason for layoffs since 2023. Their monthly reports through 2026 show AI as the single leading cause of job cuts for five months running, March through July:

Month (2026)	AI-cited job cuts	Share of that month’s total cuts
March	15,341	25%
April	27,645 (cumulative YTD)	13% (YTD)
May	38,579	Led all reasons
June	14,029	31%
July	10,970	33%

Through July, employers had cited AI in 112,713 job-cut announcements for the year, about 24% of all layoffs tracked. The running total since 2023 sits around 184,538. Two things are worth keeping in mind before treating that number as a body count. Companies self-report the reason for a layoff, and “AI” gets used alongside “restructuring” and “cost-cutting” in ways that are genuinely hard to untangle. And overall layoffs in 2026 have been running well below 2025 levels, so AI is the top reason within a shrinking pool of cuts, not evidence that employment overall is falling apart.

Goldman Sachs adds a wider estimate on top of this. Roughly 300 million jobs worldwide show some exposure to generative AI, but only about 2.5% of US employment is judged to be at direct risk of elimination. Exposure and elimination aren’t the same measurement. A lot of coverage on this topic treats them like they are.

The Tools Doing Real Work Right Now

Pull back from the macro numbers and a narrower pattern shows up. A small number of tool categories account for nearly everything that looks like genuine substitution in 2026.

Coding Tools Have the Clearest Case for Real Automation

Coding is the category with the least ambiguity attached to it. About 84% of developers report using or planning to use an AI coding assistant, with roughly 51% on one daily, and GitHub puts Copilot usage at around 90% of Fortune 100 companies. Controlled studies keep landing in the same range: developers finish scoped, well-defined tasks 30% to 55% faster with tools like GitHub Copilot, Cursor, and Claude Code. One study that tracked commit history around Cursor adoption found commit counts up roughly 36% and lines of code added up 76% to 77% over the following six months.

What these tools take off a developer’s plate is real. Boilerplate, debugging, test scaffolding, documentation, a good chunk of routine feature work. What they haven’t done is remove the developer. Senior engineers still make the architectural calls, review what the model produces for correctness, and handle the parts of a system that need context a model can’t infer from the code alone. Where the effect shows up instead is hiring. Several studies tie AI adoption at a firm to a documented slowdown in entry-level engineering hiring, which lines up with the broader early-career pattern covered later in this piece.

Klarna’s Customer Service Bet, and What It Taught Everyone Else

Klarna is still the example people reach for first when they want to argue AI is replacing white-collar workers, and the full timeline tells a more useful story than the headline version. In February 2024, the fintech company said its OpenAI-built assistant was doing the equivalent work of about 700 full-time agents, handling 2.3 million conversations in a single month and cutting average resolution time from 11 minutes down to under 2. Headcount fell from around 5,500 to roughly 3,000 over the period that followed.

By May 2025, CEO Sebastian Siemiatkowski told Bloomberg the company had pushed too far, and that chasing cost had come at the expense of quality. Klarna started rehiring, describing it early on as building a flexible remote workforce rather than admitting a reversal outright. By 2026 the company had landed on a hybrid setup. AI takes the high-volume, routine queries. Human agents handle the complicated cases, positioned publicly as something close to a premium tier of service.

Worth sitting with: the “700 agents” figure was always a productivity-equivalence claim describing how much work got handled, not a specific headcount decision made in a single moment. AI genuinely absorbed the high-volume, low-complexity end of support work. It didn’t hold up once you got into the tier that needed judgment, empathy, or something the system hadn’t seen before. Unwinding a full-replacement bet, rehiring, retraining, rebuilding institutional knowledge that walked out the door, cost more than the original savings projection ever accounted for.

Contracts and Case Law: Automation With a Lawyer Still Signing Off

Legal work follows a similar shape. Heavy task automation, very little role replacement. Spellbook, Kira Systems, Harvey, Luminance, and Casetext’s CoCounsel are now standard tools for reviewing contracts, flagging missing clauses, and pulling relevant case law. One midsize litigation group reported cutting contract review time by about 60% using an assistant that summarizes terms and checks documents against preferred language. Anthropic’s own usage data classifies legal work as one of the most augmentation-heavy categories it tracks. AI drafts and researches. The licensed professional still decides. That division of labor, not replacement, is what’s showing up across firms adopting these tools in 2026.

Research and Content Tools Are Making People Faster, Not Replacing Them

Anthropic’s Economic Index, which classifies how people actually use Claude, found augmentation, the collaborative, back-and-forth kind of use, still narrowly ahead of automation across consumer traffic, at roughly 52% to 45% in its most recent releases. Enterprise API traffic leans further toward automation for back-office work like invoice processing and email sorting. In marketing specifically, these tools are changing how the work gets done more than they’re eliminating marketing jobs outright. Our AI marketing stack guide breaks down which tools are actually cutting recurring work off a team’s plate versus adding a subscription that duplicates something the team already owns. The same pattern shows up in our AI stack guide for freelancers, where research tools like NotebookLM, covered in our full NotebookLM review, speed up what a single operator can produce rather than replacing a role that operator would otherwise have hired for.

The Clearest Pattern in the Data: Entry-Level Hiring Is Shrinking First

The single most consistent, best-documented AI employment effect in 2026 isn’t a wave of layoffs. It’s a slowdown in hiring for entry-level workers, specifically in the occupations most exposed to AI.

The Stanford Digital Economy Lab has been tracking this since August 2025, led by economist Erik Brynjolfsson and built on ADP payroll data covering roughly one in six American workers. The first report found employment for workers aged 22 to 25 in the most AI-exposed occupations, software engineering, customer service, marketing, down about 13% relative to their less-exposed peers. By the team’s June 2026 update, that gap had widened to around 19%. Employment for that age group in the two most-exposed occupation groups fell roughly 11% since November 2022, while the same age group in the least-exposed occupations grew about 10% over the same stretch.

Two parts of this research matter as much as the headline percentage. The effect runs almost entirely through reduced hiring, not through firing people already in the job. Companies aren’t cutting young staff so much as declining to backfill open roles or expand entry-level headcount. And the decline is concentrated specifically where AI substitutes for tasks rather than works alongside a person. In fields where the use skews more collaborative, employment for the same age group holds flat or keeps growing. Workers aged 35 to 40 in the exact same exposed occupations show no comparable drop, something the researchers tie to tacit, hard-to-write-down knowledge that AI hasn’t caught up to yet.

What AI Still Can’t Do Well

Even the more optimistic technical estimates stop well short of claiming most work is automated today. McKinsey Global Institute’s November 2025 analysis put 57% of US work hours as technically automatable with current technology, split roughly between 44% through AI agents and 13% through robotics. McKinsey is careful to call this a technical-potential number, not a jobs forecast, and the gap between what’s technically possible and what companies have actually rolled out remains wide.

The roles holding up best share a few traits. Someone has to be accountable for a decision. The situation requires judgment that hasn’t come up in quite that form before. It needs a physical presence, or a relationship built on trust rather than throughput. Challenger’s data through mid-2026 reflects this by sector: warehousing cuts down 58% year over year, retail down 84%, even as technology-sector cuts rose 67%. The disruption is concentrated in knowledge work far more than it’s spread evenly across the economy.

Full Automation vs. Hybrid Models: What Companies Are Learning the Hard Way

Klarna’s reversal has become the case executives now get asked to explain around, not because it proves AI can’t handle customer service, but because it shows what happens when a company automates a whole role before figuring out where the judgment-heavy edge cases actually sit.

Not every large employer is on the same path. JPMorgan Chase CEO Jamie Dimon confirmed in February 2026 that the bank has already displaced workers through AI and has significant redeployment plans in motion. Total headcount stayed roughly flat at about 318,512, but operations and support roles fell 4% and 2% while client-facing roles grew 4%, a shift the flat total number hides completely. Real displacement inside specific functions, without an overall headcount collapse, tracks closer to what the broader data actually supports than either “AI is taking every job” or “AI barely matters.”

If You’re Trying to Figure Out What This Means for You

For individual workers, the honest takeaway is narrower than the headlines suggest, but still worth acting on. Entry-level roles in software engineering, customer service, and similar AI-exposed fields are the segment that’s actually shrinking right now, mostly through reduced hiring rather than layoffs of people already there. Building the judgment, escalation-handling, and institutional knowledge that experienced workers in the same fields aren’t losing ground on is the response the evidence points toward, more than trying to guess which specific tool gets adopted next.

For teams deciding whether to automate a function, Klarna and JPMorgan point at the same underlying rule. Full automation works best on high-volume, well-documented, low-ambiguity work. Hybrid models beat full replacement almost everywhere judgment or a relationship is involved. Modeling the cost of reversing a bet, not just the projected savings from making it, would have changed Klarna’s 2024 decision before it happened.

Where the Evidence Lands, and Where It Doesn’t

What the evidence supports	What the evidence doesn’t support
Real productivity gains on scoped, well-defined tasks in coding, contract review, and tier-one support	Widespread elimination of entire professions
A documented slowdown in entry-level hiring in AI-exposed occupations	Layoffs of experienced workers at a comparable scale
Net job creation projected at the macro level (WEF: +78 million by 2030)	That projection describing conditions that already exist today
A few specific, well-evidenced tool categories doing real substitution	Every AI tool making broad automation claims performing as advertised
Company-cited “AI layoffs” data as a directional signal	That data as a clean, uncontaminated count of AI-caused job losses

At a Glance

AI is genuinely automating discrete, well-defined tasks in coding, contract review, and routine customer support in 2026. That part is real, not speculative. Full role elimination at scale is still rare, and the most prominent attempt at it, Klarna’s customer service overhaul, got significantly walked back once it reached the tier of work that needed judgment. The clearest, most consistently documented employment effect isn’t a wave of layoffs among experienced workers. It’s a contraction in entry-level hiring inside occupations already exposed to AI, happening through reduced hiring rather than terminations. Company-cited “AI layoffs” numbers are worth watching as a directional signal, but they blend real task substitution with ordinary cost-cutting wearing an AI label. Read them with that in mind rather than as a precise count.