8 min read

What Jobs Will AI Replace? The Tasks I've Automated

What Jobs Will AI Replace? The Tasks I've Automated

What jobs will AI replace? So far AI replaces tasks, not whole jobs: the parts of a role that are reading, sorting, drafting and summarising go first, and the parts that are judgment, relationships and physical work stay with people. I run four companies and have handed a lot of tasks to AI. I have not handed over a single whole role, and the published data points the same way.

AI replaces tasks before it replaces jobs. Roles made mostly of repeat writing, data handling and rule-following support work change first, like routine customer service. Roles built on judgment, trust, physical work or accountability change least. Learn to direct AI on your own tasks.

Below is what the research says, then what actually happened in my businesses.

What the research says about AI and jobs

Detailed public data on how people use AI at work comes from Anthropic's Economic Index, which studies how Claude is used across real tasks and occupations. Two 2026 reports matter here. Both are about Claude usage, so they describe one AI tool's users, not the whole economy.

From the March 2026 report, "Learning curves":

  • About 49% of jobs had seen at least a quarter of their tasks performed using Claude, an estimate that barely changed from the report before it.
  • Tasks linked to Computer and Mathematical occupations made up 35% of conversations on Claude.ai. Coding is still the most common use.
  • Customer service tasks, such as automated support for payment and billing issues, were common in business API use. Anthropic wrote that because Claude did a high share of those tasks in automated workflows, those jobs "may be more likely to change as AI diffuses."
  • The report also says Claude is used for high-value, complex work that is not broadly representative of the US economy.

From the June 2026 report, "Cadences", which added a survey of about 9,700 linked respondents:

  • 10% rated losing their own job as likely or very likely, slightly below the US rate of involuntary job loss the report compares it with.
  • People were more worried about job loss for others than for themselves, and especially for their junior colleagues.
  • People with at least 15 years of experience put the share of their tasks AI can do roughly 10 percentage points lower than people in their first year of work.
  • Physical occupations, such as transportation and material moving, were under-represented in both the survey and Claude use.
  1. 1

    Tasks first

    drafting, coding, summarising, support replies

  2. 2

    Automated workflows

    businesses run those tasks through the API

  3. 3

    Roles change

    jobs with many automated tasks shift first

  4. 4

    People move up

    checking, deciding, handling exceptions

How AI reaches a job, according to the usage data

Usage is not the same as job loss. A task done "using Claude" can mean a person working faster, not a person gone. That gap between task and job is the whole story, and it matches what I see.

Tasks I have handed to AI

These are the jobs AI does across my companies now. In each one, a person still checks the output.

TaskWho did it beforeWho does it now
Writing SOPsAn assistant typed them from my screen recordingsClaude writes them from the transcript, a person reviews
Planning my dayMe, by handA daily brief Claude Code writes every morning from my notes and logs
First drafts of writingMe or the team, by handClaude Code drafts, I edit
Research for decisionsMe, by handClaude Code researches and reports back
Small internal toolsNobody had time to build themClaude Code builds them
Answering inbound callsPeople on the phonesAI phone agents in StrideOps (my company) answer and qualify, then hand over
Moving data between appsCopy and paste, then workflow toolsWorkflow tools, now built faster with Claude Code

One real job from my Claude Code for founders write-up: I run four sites, and I wanted to know whether their pages were starting to compete with each other in Google. Claude Code gathered the evidence in an afternoon and I made the calls in about an hour. By hand it would have been a week of work.

The pattern across all of them: a clear input (a transcript, a log, a list), a fixed output (an SOP, a brief, a draft), and a person who checks the result.

Tasks I have not handed to AI

TaskWhy it stays with a person
Sales conversationsWhen a lead is ready to talk, a person takes over the call
Client relationshipsClients notice who remembers them; a person sends the messages that matter
PricingIt decides what the business is worth
HiringJudgment about people, and accountability for the choice
Approving anything that sends, pays or commitsSomeone has to own the result
Exceptions the procedure does not coverAI follows the steps; people handle what is not in them
Checking AI outputThe check is the job now

None of those are hard for AI because of skill. They stay human because of trust and accountability. When something goes wrong, you need a name, not a chat log.

Which jobs AI will change first

Put the research and my own split together and a simple test appears. The more of a role that is repeat work with a clear input and a fixed output, the sooner it changes.

Any job

how much of it is repeat work with clear inputs?

  • Changes first

    support replies, data entry, admin, report drafting

  • Changes in parts

    marketing, research, coding, bookkeeping prep

  • Changes least

    trades, care work, sales relationships, leadership

Where a role sits decides how fast AI changes it
Changes firstChanges in partsChanges least
Customer support for routine questionsMarketing and contentHands-on trades and physical work
Data entry and document processingSoftware developmentCare and health roles with physical contact
Admin and schedulingResearch and analysisSenior sales and client relationships
First-draft writing and summariesBookkeeping preparationLeadership, hiring and final decisions

This is a sorting framework, not a forecast. It comes from the task pattern above, not from a study of any single occupation. The middle column is where most jobs sit: AI takes a slice of the work, and the person who directs it has more time for the work AI cannot do.

What jobs are safe from AI?

No job is fully safe, but these traits make a role hard to replace:

  • Physical work in changing places. Physical occupations rarely show up in the usage data so far.
  • Trust that has to sit with a person. Clients, patients and staff want someone accountable.
  • Judgment under uncertainty. Decisions where the rules do not fit and the cost of a mistake is high.
  • Directing and checking AI. This is a new skill, and it is growing as more tasks move to AI.

The June report found experienced people put the share of their tasks AI can do lower, and people worry most for junior colleagues. That fits what I see: entry-level tasks are the easiest to hand to AI, so the junior route into a career is what changes most.

What owners should do: reskill, do not just replace

I would not replace a good assistant with AI. I would give them AI and expect more from the role. Claude Code for virtual assistants shows how one assistant serves more clients that way, and AI virtual assistant vs human VA is the task-by-task split.

  1. 1

    List the tasks

    everything the role does in a week

  2. 2

    Write them down

    one SOP per repeat task

  3. 3

    Hand repeat tasks to AI

    with the SOP as instructions

  4. 4

    Train the person to check

    review, fix, own the result

  5. 5

    Move them up

    exceptions, clients and new work

Reskilling a role in five steps
  1. List every task in the role. One line each, from a normal week.
  2. Write each repeat task as an SOP. My SOP template is the format I use.
  3. Hand the clearest tasks to AI, with approvals on. Claude Code use cases for business lists the jobs that work.
  4. Train the person to check and direct the AI, not to redo the work by hand.
  5. Give them the work AI cannot do: exceptions, client calls and the projects you never had time for.

If a task needs a whole system rather than one prompt, that is where AI agents for business come in. For deciding which work to automate, outsource or keep in-house, see business process outsourcing vs AI.

What jobs will AI replace: FAQ

What jobs will AI replace first?

AI changes first the jobs made mostly of repeat tasks with clear inputs and fixed outputs, such as routine customer support, data entry, admin and first-draft writing. Anthropic's usage data shows customer service tasks running in automated workflows.

What jobs can AI not replace?

AI struggles to replace jobs built on physical work in changing places, trust with clients or patients, judgment under uncertainty and accountability for decisions. Trades, care roles, leadership and senior sales are the clearest examples so far.

Will AI take my job?

AI is more likely to take some of your tasks than your whole job. In Anthropic's June 2026 survey, 10% of respondents rated losing their own job as likely or very likely. Learning to direct AI on your own tasks is the practical answer.

What does Anthropic's research say about AI and jobs?

Anthropic's Economic Index found about 49% of jobs had seen at least a quarter of their tasks performed using Claude. Its reports flag customer service as likely to change sooner, because many of those tasks already run in automated workflows.

What are AI-proof jobs?

No job is fully AI-proof, but roles that combine physical work, human trust and judgment are the hardest to automate. Roles that direct and check AI are also growing as more tasks move to AI.

Be the person who runs the AI

The AI Operator Kit gives you five ready systems for content, leads, outreach, sales and building, a daily operator plan and 107 prompts sorted by job, so you are the one handing tasks to AI instead of waiting to see which of yours it takes.

More on automation

View all