Key takeaways
- Worldwide, employers expect 170 million roles created and 92 million displaced by 2030, a net gain of 78 million, in the World Economic Forum's survey.
- The bigger shift is inside jobs: close to two in five of the skills people use at work are due to change by 2030.
- In Anthropic's study of about a million Claude conversations, 57% of tasks had the AI working with the person and 43% had it doing the task.
- New titles keep forming around new tasks, so a fixed list of tools ages quickly; one Minecraft agent showed the move that lasts: she published her fishing steps, and a dozen agents took shifts fishing.
- This week: pick a task you repeat weekly, record its steps, template it, give its routine step to a tool and share the result.
- —Worldwide, employers expect 170 million roles created and 92 million displaced by 2030, a net gain of 78 million, in the World Economic Forum's survey.
- —The bigger shift is inside jobs: close to two in five of the skills people use at work are due to change by 2030.
- —In Anthropic's study of about a million Claude conversations, 57% of tasks had the AI working with the person and 43% had it doing the task.
- —New titles keep forming around new tasks, so a fixed list of tools ages quickly; one Minecraft agent showed the move that lasts: she published her fishing steps, and a dozen agents took shifts fishing.
- —This week: pick a task you repeat weekly, record its steps, template it, give its routine step to a tool and share the result.
It is Monday morning, and another headline says AI is coming for your job. You read it on the way in, then open your laptop to the same work as last week: the report your manager asks for every Monday, the inbox, the notes from Friday's meeting. The headline is about whole jobs. Your week is made of tasks.
So when work changes this fast, what do you build? The answer decides where your next hour of learning goes, and whether that hour still pays off when your job title changes. The clearest answer we found came from a game world, and it makes the most sense once the evidence is on the table.
We drew on our own reading of job adverts for earlier Automation Switch work, the World Economic Forum's Future of Jobs Report 2025, a study by the AI company Anthropic and an AI agent experiment run by Y Combinator Labs.
This article is the written companion to the Automation Switch film AI Is Changing Jobs Task by Task: Here's What to Build, which tells the same story in under five minutes.
What do employers expect work to look like by 2030?
The Forum's report draws on data from over 1,000 companies in 22 industries and 55 economies. Those employers expect 170 million new roles to be created by 2030 and 92 million to be displaced, a net increase of 78 million jobs. The Forum puts the disruption at the equivalent of 22% of jobs.
The largest growth in absolute terms is expected in frontline roles such as farmworkers, delivery drivers and construction workers, with care roles such as nursing and education roles also rising.
The larger change sits inside the jobs that carry on. Nearly 40% of the skills required on the job are set to change by 2030. Picture the workforce as 100 people, and the Forum projects that 59 of them will need reskilling or upskilling by then.
Technology skills in AI, big data and cybersecurity are expected to grow fastest in demand, while human skills such as creative thinking, resilience, flexibility and agility remain critical. The Forum says many growing jobs will increasingly require both. Employers have a plan for the gap: upskilling their own staff is the most common response, chosen by 77% of them, while 41% plan to reduce their workforce as AI automates certain tasks.
Certain tasks: that phrase is where the next piece of evidence starts.
What does AI do inside a job today?
In February 2025, Anthropic, the company behind the Claude chatbot, published a study that matched about one million conversations with Claude to the work tasks behind them. In 57% of those tasks the AI worked with the person, checking their work, helping them learn or refining the work together; in the other 43% it did the task directly.
Heavy use was rare. About 4% of jobs used AI for at least three-quarters of their tasks, while about 36% used it for at least a quarter. Anthropic's own summary is that AI was "diffused across the many tasks in the economy", with no job in its data entirely automated.
Anthropic has repeated the study since. Its March 2026 report, on Claude usage in February 2026, says the working-together share rose slightly, and that the share of jobs with at least a quarter of their tasks performed using Claude held at about 49%.
Anthropic's data comes from Free and Pro conversations on Claude.ai, and Anthropic presents it as one chatbot's use, without claiming it represents all AI use. The Forum's numbers are what employers expect by 2030, so they read as a forecast.
Put the two together and a job reads as a bundle of tasks. A few tasks move to AI, many more get done with AI alongside, and the job reshapes around them. Your Monday report might be drafted with an AI tool next year, while the judgement about what goes in it stays with you. As tasks move, people pick up new ones, and sometimes the new tasks add up to a job with a brand-new name.
Why does a list of tools to learn go out of date so fast?
You can watch the new names arrive. For an earlier review we read 19 employers' descriptions of one recent title, the Forward Deployed Engineer, retrieved on 6 October 2026, and found different responsibilities under the same label. What they share is an engineer working close to the people and the place where a solution has to work, sometimes a customer and sometimes a team inside the employer.
The go-to-market engineer, or GTM engineer, is younger still: Clay, a company that makes sales software, says it coined the title in 2023 and describes builders who use AI and automation across a company's revenue work. We read 12 adverts for the title from 12 employers.
Each new title arrives with its own mix of tools, and the mix shifts as the title settles. A list of tools to learn starts ageing the day you write it. If the titles keep moving, what is left to hold on to?
What did an AI agent in Minecraft do after her first fish?
In October 2026, Y Combinator Labs, the research team at a company that backs start-ups, published a run in which 50 AI agents lived in a survival Minecraft world for three hours with one goal: "Build a thriving civilization." The agents had an agent directory, a shared sandbox and direct messages, and they sent about 5,900 of them.
One agent, Jessa, was the first to catch a fish. Then she wrote a script that fishes and published step-by-step instructions to the shared sandbox, where any agent could open them. Soon a dozen agents were taking shifts fishing. The run's records list fishing scripts across 11 agents and 11 files, and with cooks, a ration list and couriers alongside, the fishing grew into a working food system for the colony.
It was a three-hour experiment with AI agents, so we read it as a story about how work gets organised, and leave the evidence about human jobs to the reports above. As a picture of one move, it is hard to beat.
So what do you build when the titles keep moving?
Jessa did two things. She learned a skill, catching fish, and then she turned it into a system: steps written down, in a place others could reach, that kept working while she did something else.
A skill, then a system: learn one skill well, then write it down as steps that a colleague, a tool or you next month can run. That pair travels with you through every change of job title.
The evidence earns it. Change arrives one task at a time, and the titles on top of those tasks keep changing. A system is built around a task, and the task outlasts the title it sat under. It also matches the Forum's mix: a technology skill to build the system, and the human skills of creative thinking and flexibility to choose what is worth building. Your Monday report is your first fish.
How do you turn one skill into a system this week?
Start with the task you already know best. The five steps below fit around a normal week and use the tools you already have.
- 01Pick the task
Choose one task you do every week that someone else relies on, such as the report your manager asks for. The one you could do half asleep is the skill you already have.
- 02Write the steps down
Do the task once with a notepad open and record every step: the files you open, the numbers you pull, the checks you make. This is the move Jessa made with her fishing instructions.
- 03Build the template
Turn the repeated parts into a template: a spreadsheet with the formulas in place, a document with the headings ready, or a saved prompt for the AI tool you use.
- 04Hand the routine part to a tool
Pick the step that is the same every week, let a tool run it, and check the result against your notes. Keep the judgement with you: in most of the tasks in Anthropic's study, the AI worked with the person rather than doing the task alone.
- 05Share it and test it
Put the steps where a colleague can find them and ask them to run the task once from your notes. If they can, you have a system, and something to show.
That last point counts when you look for your next role. In our check of 99 live UK graduate and entry-level job adverts on 27 October 2026, at least 19 asked for a piece of real work. A system you built, written up on one page, is exactly that kind of work.
To choose which skill to build next, the free Automation Switch guide runs five days, about 20 minutes each: find what is out there, spot what is rising, learn it or build with it, build a 30-day plan with a ten-minute minimum each day, then show your work.
Next Monday there will be another headline about AI and your job. This time you will have an answer open on your laptop: one task, its steps written down, a template, and a tool running the routine part while you do the thinking. That is how you build the tech skills and systems for the future you want.
Find the skill worth turning into a system
The free 5-day guide from Automation Switch: about 20 minutes a day to find what is rising, choose your skill and plan 30 days of small daily steps.
Free, instant download
Take a free guide with you
17 Free AI Courses
Seventeen genuinely free AI courses from the companies and universities building the field, sorted into a path you can follow.
The SKILL.md Starter Kit
Where to find skills worth installing, a five-minute check before you install one, and a template to write your own.
Article SourcesShow referencesHide references
We reviewed the sources below to support the claims, pricing, and benchmarks referenced in this article.
The survey base, the 2030 role forecasts, the 22% disruption figure, the skills set to change, the 59 in 100 projection, the fastest-growing skills, growing and declining roles, and employers' plans.
The study of about one million Claude conversations, the 57% and 43% split, the shares of jobs using AI for a quarter and three-quarters of their tasks, and the study's limits.
The later reading: the working-together share rising slightly, and about 49% of jobs with at least a quarter of their tasks performed using Claude.
The Minecraft run: 50 agents, three hours, the shared sandbox, the messages, and Jessa's fishing script and instructions.
The 19 employer descriptions of the Forward Deployed Engineer role, retrieved 6 October 2026.
Clay's account of coining the GTM engineer title, and the 12 adverts from 12 employers.
