What jobs will AI replace?

Nobody can measure a job that has been replaced: it stops being posted. What can be measured is what 369 companies using AI are hiring for today. Across 26,929 live postings, 772 (2.9%) are for building AI, 8,176 (30.4%) expect daily AI tool use, and 17,981 (66.8%) barely mention it.

The number this article rests on: 8,176 of 26,929 live postings expect you to work with AI tools without building them (Level 2 and 3), against 772 that hire AI builders. Measured September 2026.

Most answers to this question are lists of professions ranked by how automatable they look on paper. This one is built differently. Level reads every live posting for the AI tools and tasks it names, then rates it from Level 1 (AI absent or mentioned in passing) to Level 4 (building AI is the job). Postings are decisions with a budget behind them. When a company replaces a task with AI, the next posting for that role either disappears or changes level. Both are visible here, and both are re-measured monthly.

What jobs will be gone by 2030?

No job in this catalogue is gone, and this page will not pretend to know 2030. What the 26,929 live postings show is a shift inside jobs rather than a disappearance of jobs: 8,176 roles now list AI tools as part of the daily work. The task changes first; the title changes later, if at all. Measured September 2026.

The clearest signal is where the Level 2 and Level 3 postings are. These are the jobs where the employer expects AI in the workflow but still hires a person to run it:

1. Software Engineering: 2,686 postings at Level 2 or 3, out of 7,783 live.
2. Sales: 1,488 postings at Level 2 or 3, out of 5,180 live.
3. Data: 510 postings at Level 2 or 3, out of 1,259 live.
4. Product: 467 postings at Level 2 or 3, out of 1,185 live.
5. Operations: 383 postings at Level 2 or 3, out of 1,433 live.

The tools those postings name are not exotic. Across 4,357 postings that cite any tool, the most common are OpenAI (1,062), Claude (989), Cursor (745), Clay (534), and PyTorch (514). A job that asks for OpenAI is not being replaced by it; it is being redefined around it.

What this data cannot show is the posting that never happened. If a support team of ten is rehired as a team of six with an AI queue, the catalogue sees six postings, not the four missing. That is why this page is measured monthly: a category whose live count falls while its Level 2 and 3 share rises is the pattern to watch. Today, customer support has 651 live postings, 23.3% of them at Level 2 or 3.

What jobs are safest from AI?

By the share of postings where AI is not the job, the safest categories are Healthcare, Content, Legal, and HR. In each, at least 77% of live postings are at Level 1. These companies use AI, sometimes heavily, and keep hiring people whose work stays human. Measured September 2026.

1. Healthcare: 284 live jobs, 93.0% at Level 1, 73.9% remote, median maximum $242,500 on 58 USD postings.
2. Content: 173 live jobs, 91.9% at Level 1, 83.8% remote, median maximum $164,850 on 24 USD postings.
3. Legal: 325 live jobs, 78.8% at Level 1, 49.5% remote, median maximum $265,000 on 165 USD postings.
4. HR: 450 live jobs, 77.1% at Level 1, 47.3% remote, median maximum $191,500 on 180 USD postings.

Three traits these categories share. The work carries a relationship, a liability, or a decision that someone must own. AI can draft the contract; it cannot be the lawyer of record. It can qualify a lead; it cannot close a $230,000 account and be accountable for it. And it can summarize the meeting; someone still has to run the company.

The full list of 17,981 Level 1 postings is on the Level 1 page. The companion article AI-proof jobs ranks these categories by pay and remote share.

Which jobs will AI create?

The jobs AI creates are the ones building it, and they are a small, well-paid minority. 772 live postings are at Level 4, 2.9% of the catalogue. The categories where that share is highest are Data (10.7%), Research (10.0%), Software Engineering (4.8%), and Security (2.3%). Median maximum salary at Level 4 is $246,750 on 385 USD postings. Measured September 2026.

By title, 221 postings say AI engineer, 206 say machine learning engineer, and 63 say LLM engineer. Prompt engineer appears in 5. The job market did not create a new profession for talking to models; it added the skill to existing ones.

Software engineering remains the largest source of new AI work, with 373 Level 4 postings out of 7,783 live in the category (4.8%). The rest of the category, 7,410 postings, is software work that now happens with AI tools rather than on them. That ratio is the whole story of this page in one category: AI replaces tasks inside jobs far more often than it replaces the job.

Which jobs is AI changing the fastest?

Not the ones on the automation lists. Ranked by the share of postings that name AI tools as daily work without building them (Level 2 and 3), the fastest-changing categories are Research (69.8%), Data (40.5%), and Product (39.4%). These are jobs being rewritten around AI while the headcount stays. Measured September 2026.

1. Research: 69.8% of 430 live postings at Level 2 or 3, 43 at Level 4.
2. Data: 40.5% of 1,259 live postings at Level 2 or 3, 135 at Level 4.
3. Product: 39.4% of 1,185 live postings at Level 2 or 3, 21 at Level 4.
4. Security: 37.7% of 525 live postings at Level 2 or 3, 12 at Level 4.
5. Software Engineering: 34.5% of 7,783 live postings at Level 2 or 3, 373 at Level 4.

The list skews technical, and that is the expected shape: the people closest to the models adopt them first. The more telling number sits further down. Sales is at 28.7% Level 2 or 3 on 5,180 live postings, and the tools it names are not engineering ones: Clay (534), Gong (232), Decagon (173), and Sierra (131) are sales and support products. A posting that names them is hiring a salesperson or a support lead who works with an AI layer. The count of those postings, month over month, is the best measure this catalogue has of change inside a job, not loss of it.

What jobs will no longer exist in 5 years?

The postings cannot answer for 2031, but they show which rung of the ladder is thinnest today. Of 19,442 postings with a known seniority, 960 are intern or junior (4.9%), against 12,056 senior or lead. The entry-level rung is the one to watch: it is the work AI tools absorb first. Measured September 2026.

Three cautions before reading that as a verdict. Seniority is unknown on 7,487 postings, so the entry-level count is a floor. Companies that publish on job boards skew senior in any year, AI or not. And a thin junior rung today says nothing about the titles that will exist in five years; it says only where the pressure is. This page is re-measured monthly, and the number to compare is the entry-level share, not the total.

The AI jobs questions pages hold the shorter answers on entry-level hiring, degrees, and pay, each with the same monthly measurement.

The jobs right now

Generated from the live catalogue: postings at Level 2 and 3, where the employer expects you to work with AI tools every day. Newest first, refreshed every six hours.

See the Level 3 page for the full list, or the Level 4 page for the roles building the tools.

Method

• Scope: live postings in the Level catalogue, rated Level 1 to 4 from the tools and tasks named in the text. Categories shown have at least 100 live jobs; the other category is excluded.
• Sample: 26,929 live jobs, of which 11,953 publish a salary and 10,606 publish it in USD.
• Measured: September 2026, from the live catalogue at jobsbylevel.com, refreshed every six hours.
• Currency: USD only, no conversions. Medians are on the posted maximum unless stated.
• Not measured: postings that were never written; seniority is unknown on 7,487 postings; no forecast for 2030 or any other year is made.