Jobs Already Being Replaced by AI — and the Jobs Being Displaced Now

What AI changed

Neural machine translation turned much of the work from translating-from-scratch into post-editing an AI draft, and the pay structure followed the tasks down. High-volume, lower-stakes content is increasingly handled machine-first with a human checking the output.

Evidence

A January 2024 Society of Authors survey of 787 authors found that 36% of translators said they had already lost work to generative AI. A 2025 study by Carl Benedikt Frey and Pedro Llanos-Paredes, summarised on VoxEU, found that in U.S. regions with faster Google Translate adoption, each one-point rise in machine-translation use was associated with roughly 0.7 points slower translator employment growth. The shift is now reaching publishers directly: in August 2026, on its fiscal-year results call, HarperCollins' chief executive said the company aims to automate its overseas translations, particularly in India. In a September 2026 speech, Federal Reserve Governor Lisa Cook cited coding jobs and simultaneous translation as sectors where there is some evidence that AI may be decreasing labour demand. The BLS still projects slow growth for interpreters and translators from 2025 to 2035, while rating the occupation's AI exposure "very high".

What still needs people

Literary, legal, medical, and diplomatic translation still depend on people who carry nuance, cultural context, and legal responsibility. Live interpretation under pressure, where a mistranslation has real consequences, remains a human-led task.

What AI changed

AI chat and voice assistants now resolve a large share of routine, repetitive enquiries end to end, compressing the volume of tickets that ever reach a person and shrinking the case for large first-line teams.

Evidence

In February 2024 Klarna said its OpenAI-powered assistant was doing the equivalent work of 700 full-time agents; in 2025 it began hiring human agents again after its CEO conceded that cost had weighed too heavily at the expense of quality — a useful caution that full automation can overshoot. The pressure elsewhere is documented, though: in July 2026 Uber cut 10% of its customer-service jobs, the first time it had tied layoffs to its efforts to embrace AI, and Bloomberg reported that Microsoft, Commonwealth Bank of Australia and Hyatt are using automated chat and phone systems for work that once needed people — although the companies differ on how much of their job cuts they attribute to AI. The BLS now projects customer service representatives to shrink between 2025 and 2035, one of the larger declines on this page.

What still needs people

Complex complaints, emotionally charged situations, edge cases, and anything requiring real authority to make an exception still route to people. The role is shifting from high-volume first response toward supervising AI and handling what it cannot.

What AI changed

Document capture, data extraction, and routine back-office processing are exactly the kind of structured, rule-bound work that automation handles well, so the same output now needs fewer hands.

Evidence

The World Economic Forum's Future of Jobs Report 2025, drawn from a survey of employers, lists clerical and secretarial roles — data-entry clerks, administrative and executive secretaries, and bank tellers — among those expected to decline most by 2030, with data-entry clerks projected to fall around 26%. The BLS's 2025–35 projections, published in August 2026, point the same way: it projects a steep decline for data entry keyers and a decline for office and administrative support as a whole, with the BLS citing automation tools including AI. Employers have also started naming AI as a reason for cuts — Challenger, Gray & Christmas counted 116,175 announced U.S. job cuts attributed to AI in January–August 2026, more than a fifth of all announced cuts and the leading stated reason year to date — though AI fell to fourth place among reasons in August, and Challenger records the reasons employers give without verifying them.

What still needs people

Coordination that spans teams, judgement about unusual cases, and the relationship-management side of administrative work are far harder to standardize and tend to stay with people even as the keystroke-heavy tasks fall away.

What AI changed

Generative models produce serviceable first drafts of product descriptions, ad copy, and routine marketing content in seconds, pushing demand away from volume writing and toward editing, strategy, and brand judgement.

Evidence

A peer-reviewed study of a large freelance platform (Demirci, Hannane and Zhu, Management Science, 2025) found that within eight months of ChatGPT's release, job postings for automation-prone writing and coding work fell about 21% relative to manual-intensive work, with writing among the steepest declines. The official statistics tell a quieter story about salaried jobs: the BLS projects writers and authors to be essentially flat and technical writers to grow slightly from 2025 to 2035, even while rating both "very high" AI exposure. The clearest damage so far is in freelance, high-volume writing rather than in staff positions.

What still needs people

Original reporting, a distinctive brand voice, persuasive work that depends on understanding a specific audience, and anything where accuracy and reputation are on the line still reward skilled human writers — increasingly as editors and directors of AI output.

What AI changed

Software now reconciles transactions, categorizes entries, and flags anomalies automatically, so a single person supported by automation covers what once took a small team of clerks.

Evidence

The U.S. Bureau of Labor Statistics projects employment of bookkeeping, accounting, and auditing clerks to decline from 2025 to 2035, saying software has automated many of their tasks, and its new exposure table rates the occupation "very high". This is the clerk-level role specifically: licensed accountants and auditors, whose work centres on judgement and assurance, have a different, more stable outlook.

What still needs people

Interpreting unusual transactions, advising on decisions, and taking responsibility for what the numbers mean stay with people. Automation removes the data-shuffling, not the accountability.

What AI changed

AI grammar and editing tools catch a large share of mechanical errors instantly, suppressing demand for routine, first-pass proofreading even where final editorial judgement still matters.

Evidence

The BLS's 2025–35 edition now projects employment of editors to fall — a reversal from the slight growth it projected a year earlier — and proofreaders and copy markers to edge down, rating both "very high" AI exposure. The BLS attributes the editors' decline mainly to print publications losing ground to digital formats rather than to AI, so the AI case here rests on how many routine checking tasks tools now perform, not on a documented collapse in head-count.

What still needs people

Substantive editing — structure, argument, tone, fact-checking, and protecting a publication's credibility — depends on judgement that mechanical correction tools do not provide.

What AI changed

AI voice agents now handle outbound calling, lead qualification, and routine follow-ups, automating the scripted, high-repetition core of the role.

Evidence

The BLS projects telemarketers to decline steeply from 2025 to 2035 — one of the steepest drops of any occupation — and rates them "very high" AI exposure. In February 2024 the U.S. FCC ruled that AI-generated voices in robocalls count as "artificial" under the Telephone Consumer Protection Act, so outbound calls made by voice agents fall under the same consent rules as prerecorded ones.

What still needs people

Complex, consultative selling that depends on reading a person and building trust over time is much harder to automate than scripted outbound dialing.

What AI changed

Text-to-image generation produces usable visuals on demand, undercutting commissions for routine, high-volume illustration and template-style design work.

Evidence

A January 2024 Society of Authors survey found that 26% of illustrators had already lost work to generative AI and that 37% said the income from their work had decreased in value because of it. Salaried design work looks steadier than that suggests: the BLS projects graphic designers to decline from 2025 to 2035 while rating the occupation "very high" AI exposure, and art direction, brand identity, and concept work are the parts of the job least easily replaced by image generation.

What still needs people

Art direction, brand identity, conceptual work, and design that has to solve a specific business problem still reward people who can decide what is worth making — not just generate more options.

What AI changed

Document review, discovery, and first-draft legal research — the high-volume reading work of legal support — are the tasks that legal-AI vendors now target with tools that sort and summarise large sets of documents, which changes how the work is done even though headcount has so far barely moved.

Evidence

Legal-AI tools such as Harvey and CoCounsel are marketed for document review and first-draft research, but the labour data has not followed so far: the BLS projects paralegals and legal assistants to be essentially flat from 2025 to 2035, rating their AI exposure "high" rather than "very high". A steeper projected decline is among legal secretaries. The role is being reshaped task by task, not eliminated outright.

What still needs people

Client interaction, supervising and verifying AI output, courtroom and filing logistics, and the parts of legal work that carry professional responsibility remain firmly human.

What AI changed

AI code generation now writes routine implementation, boilerplate, and tests, which hits entry-level programming hardest — the very tasks junior developers used to be hired to do.

Evidence

The August 2026 update of Stanford's "Canaries in the Coal Mine" study, using payroll data through June 2026, finds employment of 22–25-year-olds in the most AI-exposed occupations about 19% below where it would be had it kept pace with similarly aged workers in less-exposed occupations — up from about 15% in the July 2025 update — with the adjustment operating mainly through reduced hiring rather than layoffs. The authors describe these as descriptive patterns, not causal estimates. The BLS splits the profession in two: computer programmers are projected to shrink from 2025 to 2035 while software developers are projected to grow. In September 2026 Federal Reserve Governor Lisa Cook cited coding jobs in the software industry as one area where there is some evidence that AI may be decreasing labour demand, and noted that many recent graduates are having more difficulty finding first jobs.

What still needs people

System design, judgement about what to build and what to refuse, debugging under uncertainty, and accountability for production systems are where experienced engineers keep their value — and senior demand has held up better than junior.

What AI changed

Generative AI now drafts research summaries, standardizes reporting, and pulls together comparisons, automating the routine synthesis layer of analyst work — but so far this shows up as changed tasks, not lost jobs.

Evidence

This is the most tentative entry on the page and belongs on a watch-list. The BLS rates market research analysts and marketing specialists "very high" AI exposure, yet still projects the occupation to grow from 2025 to 2035, and we found no documented, AI-attributed head-count cuts. Whether the automation of routine synthesis turns into fewer jobs remains an open question.

What still needs people

Framing the right question, designing sound methodology, and judging which findings actually matter to a decision are harder to hand to a model than the write-up of results.

What AI changed

AI tools now assemble pitch materials, build first-pass models, and summarize filings, compressing the hours of routine analysis that junior finance roles were built around.

Evidence

AI tools are increasingly used for first-pass modelling, pitch materials and filing summaries — the routine analysis that junior finance roles were built around. The labour data is mixed rather than dramatic: Bloomberg reported in July 2026 that payrolls in finance and information combined had been falling by about 28,000 a month so far in 2026, though economists quoted said it was too early to attribute that broadly to AI. The BLS still projects financial and investment analysts to grow from 2025 to 2035, while expecting declines for tellers, credit analysts and insurance underwriters. If the change comes, it is likelier to show up in how many juniors a desk needs than in mass layoffs.

What still needs people

Judgement under uncertainty, client relationships, deal negotiation, and ownership of recommendations stay with people; the model accelerates the analysis, it does not sign off on it.

What AI changed

AI agents now source candidates, screen applications, and schedule interviews, automating large parts of the coordination-heavy core of recruiting.

Evidence

Vendors now market AI agents that source candidates, screen applications, and schedule interviews. Head-count has not followed so far: the BLS projects human resources specialists, the category that includes recruiters, to grow from 2025 to 2035 despite a "very high" exposure rating, so the evidence to date is of task-level automation rather than documented displacement.

What still needs people

Persuading scarce candidates, assessing fit and judgement, closing offers, and the relationship and advisory side of hiring remain people-led — recruiters shift toward strategy as the admin is automated.

What AI changed

AI-generated answers at the top of search results are absorbing clicks that used to flow to websites, eroding the search-traffic foundation that much SEO and content-marketing work was built on.

Evidence

Search traffic to publishers has fallen sharply, with AI-generated results one of the pressures. Chartbeat's 2026 Publisher Playbook found Google Search referrals to publishers in its network fell 40.2% between July 2025 and July 2026 — against a 21.9% fall the year before — while total pageviews barely moved, and it named AI-generated search results as one pressure among several. The effect shows up first as falling traffic that undercuts demand, and the field is reorienting toward visibility inside AI answers.

What still needs people

Strategy, brand positioning, original research and data, and earning genuine authority are harder to automate than keyword-driven content production.

What AI changed

Some outlets now publish AI-assisted or AI-generated articles, while AI search answers pull readers away from news sites, squeezing the advertising economics that fund newsrooms.

Evidence

Newsroom job cuts continued through 2026, and AI is named as the cause in only some of them. In August E.W. Scripps announced 268 job cuts as part of a pivot to streaming-first local news, with its chief executive saying the company was "leaning into AI, automation, technology". McClatchy's newsroom cuts, which the union put at more than 90 journalists, were explained in an internal memo by a 41% fall in consumer revenue rather than by AI — a figure the NewsGuild said needed to be verified. The larger pressure is economic: Google Search referrals to publishers in Chartbeat's network fell 40.2% over the year to July 2026. The BLS projects news analysts, reporters and journalists to decline from 2025 to 2035.

What still needs people

Original reporting, source relationships, verification, and accountability for what gets published are the core of journalism and are not something a model can take responsibility for.

What AI changed

AI voice synthesis now narrates audiobooks, ads, and explainer content, taking on routine narration work that used to be recorded by people.

Evidence

AI voice synthesis now narrates ads, explainers, and a growing share of audiobooks. The argument is moving toward consent and licensing: in July 2025 Audible's ACX opened an invitation-only U.S. beta that lets narrators create AI replicas of their own voices for projects they choose to take, with replica-narrated titles labelled as such. The BLS draws the line in the same place: it rates actors only "moderate" AI exposure, but broadcast announcers and radio DJs "very high", with a projected decline from 2025 to 2035.

What still needs people

Character-rich performance, emotional range, and work where a specific human voice is part of the value still belong to performers — and consent and licensing of one's own voice are becoming central issues.