Browse the site

Where the change puts work

Automation does not only remove work; it creates work, and it fails in specific places. This page shows both, from the same evidence base as the rest of the site. It is deliberately not a list of startup ideas — see below for what we cannot tell you.

Tasks automation created

across 30 roles

Deployments that stalled

across 20 roles

Core work people keep

What this page cannot tell you

Market size, who is already building it, whether anyone will pay, and whether the need is real rather than merely visible. We have none of that data, and generating it from what we do have would produce confident-sounding guesses. What is below is narrower and checkable: work that exists because automation created it, and places where a deployment was tried and did not hold.

Work automation created

These tasks exist because something was automated. They are new duties inside existing roles, and nobody is assigned to them by default.

Post-editing machine output✓ Evidence-backedcore

Fixing what the machine got wrong, and knowing where to look for it.

In Translator / Interpreter — It is real work, but it is usually priced per word at a fraction of translation rates. More of the industry's hours, less of its income.

Reviewing generated code✓ Evidence-backedcore

Reading plausible code carefully enough to catch what is confidently wrong.

In Junior software developer — Reviewing well requires having written the thing being reviewed. If the writing work that trained that judgement is the work being automated, this task has a supply problem within a few years.

Supervising the automation itself✓ Evidence-backedsignificant

Checking what the tools produced, catching silent errors, and owning the result when a model got it wrong.

In Accountant / Bookkeeper — New work is not the same as new headcount, and this task is usually absorbed by people already there rather than hired for. It also requires the judgement built by doing the work that is disappearing.

Reviewing what the bot said✓ Evidence-backedsignificant

Auditing automated conversations for wrong answers, bad tone and promises the company cannot keep.

In Customer service representative — One reviewer can audit the output of many bots, which is precisely why this task does not replace the headcount that the bots displaced. It is a smaller, more senior job.

Owning the system, not the artefact≈ Platform inferencesignificant

Defining the rules, components and constraints that keep output coherent when other people (and tools) produce it.

In Graphic designer — Design systems are per-organisation, not per-designer. One system owner serves everyone who produces, so this role scales with the number of brands, not with the volume of output.

Checking what the tools produced✓ Evidence-backedsignificant

Confirming that a cited case exists, that a summary matches the source, that a generated clause does not contradict another.

In Paralegal — Whether this task is paid as skilled work or absorbed as unpaid diligence depends on the firm, and that decides whether it is an opportunity or just more load.

Supervising machine-assisted work✓ Evidence-backedsignificant

Deciding what the tools may be used for, and signing for the output as if you had done it yourself.

In Lawyer — This is an added duty, not an added role — it arrives as unbilled responsibility on people who already had a full week. Nobody is hiring a supervisor of machine work; they are extending the existing duty of competence.

Detecting synthetic and manipulated media≈ Platform inferencesignificant

Deciding whether a viral clip, image or audio is real before it is reported as fact.

In Journalist — Specialist desks exist at large outlets and essentially nowhere else. For most journalists this is a skill that makes them more employable, not a job that exists to be applied for.

Defining and policing a voice≈ Platform inferencesignificant

Deciding how the brand sounds, writing that down in a way tools and colleagues can follow, and catching the drift.

In Copywriter — One voice owner per brand is the ceiling. This role is real and it is where experienced copywriters land, but there is exactly one of it per company.

Operating the marketing automation itself✓ Evidence-backedsignificant

Designing the prompts, workflows and guardrails that let tools produce and place content without embarrassing the brand.

In Marketing specialist — The site's own phrasing is that the production hours reappear 'partly' as supervision. Partly is the operative word: supervision of many automated channels is fewer hours than producing for them was.

Owning the definitions the tools rely on≈ Platform inferencesignificant

Maintaining the metric definitions, the documentation and the semantic layer that make natural-language querying produce correct answers.

In Data analyst — Curating definitions is a smaller job than answering questions was, and it is one per organisation. Headcount being 'redirected' into it is a euphemism for a smaller team doing different work.

Redesigning jobs around automation≈ Platform inferencesignificant

Working out which tasks in which roles are changing, what the new roles look like, and how to move people into them.

In HR / recruiter — It is project work that peaks during a transition and subsides. Building a career on helping other people through redundancy has an obvious ceiling.

Clinical services✓ Evidence-backedsignificant

Vaccinations, medication reviews, minor ailment consultations, chronic disease monitoring.

In Pharmacist — This expansion is driven by policy and workforce shortage, both of which can reverse. It is a bet on health systems continuing to be short of doctors.

Designing for non-deterministic products≈ Platform inferencesignificant

Shaping products whose output varies — conversational interfaces, agents, generated content — where the old screen-by-screen craft does not apply.

In Product / UX designer — Scarcity here is temporary by construction — almost no one has done it for long because it is new, and that gap closes as everyone gets the same few years of practice.

Directing generated footage✓ Evidence-backedsignificant

Producing shots that were never filmed — generated b-roll, extensions, replacements — and making them cut seamlessly with what was.

In Video editor — Disclosure rules, client policies and audience trust constrain where generated footage may be used, and the constraints differ sharply between advertising, corporate and journalistic work.

Testing systems with a model inside≈ Platform inferencesignificant

Evaluating software whose output is not deterministic — building evaluation sets, catching regressions in behaviour, testing for harmful outputs.

In Software tester / QA engineer — Few practitioners is a statement about now. The discipline is young enough that its eventual size is unknown, and it may end up inside the model teams rather than in QA.

Picking online orders in store≈ Platform inferencesignificant

Walking the aisles to assemble a customer's online basket for collection or delivery.

In Retail cashier / shop assistant — In-store picking exists because delivery economics have not been solved, not because the work is valuable. Dark stores and automated fulfilment centres are the direction of travel where volume justifies them.

Electrification: EV chargers, heat pumps, solar, storage✓ Evidence-backedsignificant

Installing and integrating the equipment that the shift away from fossil fuels puts into every building.

In Electrician — This demand is created by policy — subsidies, bans, targets — and policy reverses. A trade whose growth depends on a subsidy schedule should read the schedule.

Running the machine that now sells with you≈ Platform inferencesignificant

Checking what the assistant drafted before it goes out, correcting the price it suggested, feeding back what actually closed, and catching the confident mistake before the customer sees it.

In Sales / account manager — New duties are not new headcount. This lands on the reps who are already there, usually without a title change and without coming off their number — which is how a role gets heavier while looking unchanged on the org chart.

Remote supervision of autonomous fleets✓ Evidence-backedperipheral

Monitoring several autonomous trucks from a desk and intervening when one gets stuck.

In Truck driver — The ratio matters and is not yet settled publicly. One supervisor per many trucks is the commercial premise; one per two trucks would not change the industry's labour picture much.

Orchestrating the analytical tooling≈ Platform inferenceperipheral

Setting up the data feeds, model assistants and checks so the team's output is fast and trustworthy.

In Financial analyst — Explicitly peripheral by weight, and typically absorbed by one person on a team as an extra. It is not a job posting.

Orchestrating tutoring tools✓ Evidence-backedperipheral

Deciding which tools students use for what, checking what the tools taught them, and catching what they got wrong.

In School teacher — This arrives as new curriculum with no extra time, which the site's own reasoning says explicitly. New responsibility without new hours is a workload change, not a new role.

Working with clinical decision tools≈ Platform inferenceperipheral

Using and overriding the alerts, triage suggestions and documentation drafts, and reporting when they are wrong.

In Registered nurse — Peripheral by weight and formalised by hospitals as part of the existing role. It adds responsibility to a shift that is already full.

Operating the executive's assistants✓ Evidence-backedperipheral

Configuring, supervising and correcting the software assistants that now handle the routine — and catching what they get wrong.

In Administrative assistant — Peripheral by weight and it lands on whoever already knew the preferences. It is a reason to keep one assistant, not a reason to keep the team.

Teaching customers the digital channels≈ Platform inferenceperipheral

Sitting with someone and getting them onto the app, safely, so they do not need to come back for the routine.

In Bank teller — The task is explicitly about teaching customers to need the branch less. It is stable only for as long as there is a cohort that has not made the transition, and that cohort shrinks every year.

Operating and maintaining the robot fleet✓ Evidence-backedperipheral

Monitoring the fleet, recovering stuck robots, swapping batteries, running the first-line maintenance.

In Warehouse worker — A fleet needs far fewer technicians than it displaces pickers, and the skills are different enough that the displaced are not the ones hired. This is a real job and a poor answer to the previous three.

Programming and maintaining the robots≈ Platform inferenceperipheral

Teaching a robot a new path, adjusting the vision system, running preventive maintenance on the line's automation.

In Assembly line worker — Plants report shortages of these technicians while also not funding the training that would produce them from the operators they already have. The opening is real and the bridge to it is not built.

Working alongside kitchen automation✓ Evidence-backedperipheral

Loading, monitoring and cleaning the automated fryer or assembly unit, and taking over when it fails mid-rush.

In Chef / cook — Peripheral by weight, and it appears in the kitchens that automated — which are the kitchens with fewer cooks. It is a consequence of the change, not a shelter from it.

Loading, recovering and supervising delivery robots✓ Evidence-backedperipheral

Where robots or drones operate, someone loads them, retrieves the stuck one and handles the delivery it could not finish.

In Delivery rider / courier — It exists only where the pilots do, which today is a handful of neighbourhoods. The people-to-robot ratio is a commercial question still being answered, and the answer the operators want is a low one.

Remote assistance and fleet operations≈ Platform inferenceperipheral

Monitoring vehicles, resolving stuck ones, cleaning and repositioning the fleet, handling the passenger the car cannot.

In Ride-hail / taxi driver — Every deployment employs these people today at ratios the operators openly describe as temporary. Planning around a job whose employer has publicly said it intends to need fewer of them is a short plan.

Where deployment stalled

Every record here is a verified failure, withdrawal, regulation or cost that suppressed adoption. An unsolved problem looks like this before anyone solves it.

Constraint2026-07-28Graphic designer

Equinox removed an AI-generated poster from its 2026 'Question Everything But Yourself' campaign and apologised, after months of criticism that the image was dehumanising