Computer Programmers
In short: today's best AI models could do about 44% of this job's working time (projected from METR's April 2026 measurement and Epoch's November 2026 index) if the work were set up for them, and about 86% by the end of 2028 if progress keeps its long-run pace. The conversations, in-person and hands-on work, about 11% of the time, stays with people.
Create, modify, and test the code and scripts that allow computer applications to run. Work from specifications drawn up by software and web developers or other individuals. May develop and write computer programs to store, locate, and retrieve specific documents, data, and information.
92,230 people do this job in the US. $100,390 a year.
Reviewed on October 1, 2026: what each task is made of, and what handing it over would ask.
A typical 40-hour week
- AI could take on3 h
- A person checks AI's work15 h
- Desk work AI can't do yet18 h
- Stays with people4 h
Computer Programmers, now: of a 40-hour week, AI could take on 3 hours, 15 hours become checking its work, 18 hours are desk work it can't do yet, and 4 hours stay with people.
The largest pieces: Write and rewrite programs (about 3 h, a person checks); Run trial program runs (about 3 h, a person checks); Write program documentation (about 2 h).
Estimates for a typical week of this job, if progress keeps its long-run pace, counting work the best models finish four times in five. Every task is below.
of the working time. Now is projected from METR's April 2026 measurement and Epoch's November 2026 index.
- 85%
- 4%
- 9%
- 1%
- 0%
of the work: Anthropic's measure of what people already bring to Claude at work (2025). It counts help with part of a task, so it can be higher than the share within reach.
On the chosen pace and reliability. Estimate.
Share of the working time within reach, now to 2030
Change assumptionsHide assumptionsLong-run pace, four in five
What each choice means: , , , , . Every pace starts from .
At this date and pace, models finish tasks up to 2.4 work days long four times in five.
What handing it over would take, now
The 44% within reach splits by what a model would need. The dates, pace and reliability follow the controls above. Estimates.
of the working time. The material can be handed to a model and the result handed back.
of the working time. The model has to be connected to the system the work lives in.
is conversation short enough by length. This site leaves conversation with people, so it is not in the total.
How far this has come
With GPT-4, measured in March 2023, 0% of this job's working time was within reach at . With Claude 3.7 Sonnet, measured in February 2025, 0%. With Claude Mythos Preview (early), measured in April 2026, 8%. Today's estimate: 44%.
A backcast: today's task data with the best model METR had measured by each date. It shows how fast the models have moved, not what anyone forecast at the time.
Measured use over time
Of the Claude.ai conversations Anthropic sampled and tied to a work task, 1.7% went to this job's tasks in May 2026, the 11th largest of the 447 jobs that showed up at all; 9 of its 17 tasks showed up. Within each of Anthropic's methods, never across a change: 6.2% to 7.6% (released February and March 2025); 5.3% to 3.6% (August 2025 to February 2026); 1.8% to 1.7% (April to May 2026).
Each release
| Conversations | Share | Rank | Tasks seen |
|---|---|---|---|
| Released February 2025 | 6.2% | 1 of 708 | 12 of 17 |
| Released March 2025 | 7.6% | 2 of 703 | 11 of 17 |
| The week of Aug 4, 2025 (released September 2025) | 5.3% | 3 of 655 | 11 of 17 |
| The week of Nov 13, 2025 (released January 2026) | 4.4% | 2 of 672 | 10 of 17 |
| The week of Feb 5, 2026 (released March 2026) | 3.6% | 2 of 677 | 10 of 17 |
| April 2026 (released June 2026) | 1.8% | 10 of 434 | 8 of 17 |
| May 2026 (released June 2026) | 1.7% | 11 of 447 | 9 of 17 |
Measured, not estimated: Anthropic's samples of Claude.ai conversations, each tied by a model to an O*NET task, from the Anthropic Economic Index. It says where Claude's use goes, not how much of this job's working time it covers, so it is not on the same scale as the reach above. The releases differ in their sample, their classifier and the smallest share they report (the June 2026 file rounds each task to a hundredth of a percent, so a smaller task reads as none), which is why a change of method starts a new line. At least 96% of each older release's task-tied use matched this site's tasks, and all of the newest. Claude.ai only, so use through other tools and other companies' models is not in it. Every job ranked.
Robots and the physical work
Anthropic rates almost none of this job's working time as physical work, so robots do not reach it.
Anthropic's rating, not this site's estimate: Claude Opus 5, searching the web, rated each physical O*NET task by the kind of setting in which a robot it could cite (deployed, sold or demonstrated) does most of the task, weighted by its own estimate of each task's share of the time (What work can robots do?, September 30, 2026, data CC BY 4.0; every job ranked). A robot shown somewhere is not a robot in your building: cost, setup and safety decide that. The share within reach on this page still counts no body work. The robot behind each rated task is under the task below.
What changes for the people in this job
From the task-by-task review, by the end of 2028 on the long-run pace at four in five. It says which parts of the work stay with people and which change, not whether anyone keeps a job: that depends on decisions no data set can see.
What stays with people
11% of the working time: the conversation, in-person and hands-on parts of the tasks.
- Train users on programs, the conversation, 2%
- Revise and expand existing programs, the conversation, 2%
- Assign and review programmers' work, the conversation, 1%
- Clarify program intent with the team, the conversation, 1%
- Train junior programmers, the conversation, 1%
- Develop methods with vendors and users, the conversation, 1%
What becomes checking the machine
76% of the working time: desk work a model could do by then, in tasks where a mistake costs something or a licensed person signs. A person checks it. Another 9% is desk work that is small if wrong, which can be handed over with lighter checks.
- Write and rewrite programs, costly if wrong, 12%
- Write and maintain business programs, costly if wrong, 11%
- Maintain systems software, costly if wrong, 10%
- Develop websites, costly if wrong, 9%
- Revise and expand existing programs, costly if wrong, 9%
- Chart the logic and code it, costly if wrong, 9%
What to learn
- Briefing a model and checking its drafts. About 85% of the time is documents and data, the part models reach first: writing a clear brief, giving the right material, and checking what comes back.
- Reviewing work you did not do yourself. The tasks under What becomes checking the machine, where a mistake costs something, stay a person's to check: spotting what a draft gets wrong is a skill of its own.
Your own version of this job: read it against your job description. The headline is free; the full personal report, every task with its timeline and what to learn, is $19 (a sample).
The tasks
Sorted by . 17 tasks from .
| Task | ||||
|---|---|---|---|---|
| Write, analyze, review, and rewrite programs, using workflow chart and diagram, and applying knowledge of computer capabilities, subject matter, and symbolic logic.Writing and reworking code from the charts and the logic of the problem.Done hourly or more by most who do it. AI observed on this task. | 12% | 6.7 h3.3 h to 4.2 work days | Documents 100% | 61%Half by now |
| Perform or direct revision, repair, or expansion of existing programs to increase operating efficiency or adapt to new requirements.Changing existing code, and directing others who do.Done daily by most who do it. AI observed on this task. | 12% | 6.7 h1.7 h to 4.2 work days | Documents 80%, conversation 20% | 32%Desk part half by Jan 2027 |
| Write, update, and maintain computer programs or software packages to handle specific jobs such as tracking inventory, storing or retrieving data, or controlling other equipment.Writing and keeping up the programs that do the organization's jobs.Done several times a day by most who do it. AI observed on this task. | 12% | 2.1 work days3.3 h to 1.7 work weeks | Documents 100% | 15%Half by Oct 2027 |
| Develop Web sites.Building and coding websites.Done yearly or less by most who do it. | 12% | 4.2 work days6.7 h to 3.3 work weeks | Documents 100% | 5%Half by Apr 2028 |
| Perform systems analysis and programming tasks to maintain and control the use of computer systems software as a systems programmer.Analysis and code for the systems software, with work in the systems themselves.Done daily by most who do it. AI observed on this task. | 10% | 1.3 work days3.3 h to 1 work week | Documents 70%, systems 30% | 39%Half by Jan 2027 |
| Prepare detailed workflow charts and diagrams that describe input, output, and logical operation, and convert them into a series of instructions coded in a computer language.Drawing the flow of the program, then writing the code from it.Done more than monthly by most who do it. AI observed on this task. | 9% | 6.7 h1.7 h to 3.3 work days | Documents 100% | 46%Half by Dec 2026 |
| Conduct trial runs of programs and software applications to be sure they will produce the desired information and that the instructions are correct.Running the program on test cases and checking the output.Done more than yearly by most who do it. | 7% | 2.5 h50 min to 6.7 h | Documents 100% | 96%Half by now |
| Compile and write documentation of program development and subsequent revisions, inserting comments in the coded instructions so others can understand the program.Writing the documentation and the comments in the code.Done daily by most who do it. AI observed on this task. | 6% | 3.3 h50 min to 1 work day | Documents 100% | 86%Half by now |
| Correct errors by making appropriate changes and rechecking the program to ensure that the desired results are produced.Changing the code and running it again until the results are right.Done several times a day by most who do it. AI observed on this task. | 6% | 1.7 h25 min to 6.7 h | Documents 100% | 88%Half by now |
| Train users on the use and function of computer programs.Showing users how the program works, in sessions by video or in the room.Done more than monthly by most who do it. | 2% | 2.5 h50 min to 1 work day | Conversation 60%, in person 40% | Stays with people |
| Assign, coordinate, and review work and activities of programming personnel.Handing out work and reviewing it, in conversation and in the code.Done more than weekly by most who do it. | 2% | 2.5 h50 min to 6.7 h | Conversation 60%, documents 40% | 38%Desk part half by now |
| Collaborate with computer manufacturers and other users to develop new programming methods.Working with outside vendors and users on new methods, in calls and in writing.Done more than monthly by most who do it. | 2% | 6.7 h1.7 h to 4.2 work days | Conversation 60%, documents 40% | 16%Desk part half by Jan 2027 |
Within reach, per task: the share of the task's time whose instances are short enough for a model to finish at the chosen date, pace and reliability. Half by: the date half of that time comes within reach. Made of: how the task's time splits among five kinds of doing. Only the documents and systems parts can count as within reach, so a task that is 30% documents can be at most 30% within reach.
Jobs that share skills with this one
From O*NET's related occupations. Within reach now and by the end of 2028, long-run pace, four in five.
Where this job works
The industry groups that employ the most of it (BLS, May 2025).
Where the most of these jobs are
The metro areas with the most people in it (BLS, May 2025).
Questions about this job
- How much of the work of Computer Programmers is within reach of AI now?
- An estimated 44% of the working time is within reach of AI models now and 86% by the end of 2028, on METR's long-run pace at four in five (73% to 89% across the three paces the site shows). Within reach means a model could finish work that long at that reliability. It is not a forecast of jobs lost or of savings.
- Which tasks of Computer Programmers come within reach first?
- Weighted by the time each takes, the tasks with the most within reach now (projected from METR's April 2026 measurement and Epoch's November 2026 index) are: Write and rewrite programs (61% of the task's time now, 100% by the end of 2028); Run trial program runs (96% of the task's time now); and Write program documentation (86% of the task's time now, 100% by the end of 2028).
- Which tasks of Computer Programmers stay with people?
- Tasks that are people work or body work are not counted as within reach at any date. The largest: Train users on programs; and Train junior programmers. Together such tasks are about 4% of the working time, and the people and body parts of the other tasks stay with people too.
- What should Computer Programmers learn as AI takes on more of the work?
- By the job's own task-by-task review, the first things to learn are briefing a model and checking its drafts; and reviewing work you did not do yourself. By the end of 2028 on the long-run pace at four in five, about 11% of the working time stays with people, 76% becomes checking a model's work and 9% could be handed over with lighter checks. It says which parts of the work change, never whether anyone keeps a job.
- Do people already use AI for the work of Computer Programmers?
- In the Claude.ai conversations Anthropic sampled for May 2026 and tied to a work task, 1.7% went to this job's tasks, the 11th largest share of the 447 jobs that showed up; 9 of its 17 tasks appeared. Anthropic's labor market measure puts the share of the job's tasks seen in real Claude use at work at 75%. These are measures of use, not of how much of the work AI could finish, and they cover Claude only.
- Can robots do the physical work of Computer Programmers?
- Anthropic's robot exposure rating (September 2026) finds almost none of the working time is physical work, so robots do not reach it.
- How is this estimated?
- From the 17 tasks O*NET lists for the occupation, each with an estimated length checked against GDPval and a share of the working time, reviewed one task at a time on October 1, 2026 for what each is made of (documents, business systems, conversation, in person, hands-on), and METR's measurements of how long a task the best AI models can finish, carried forward to the date from the latest, Claude Mythos Preview (early) in April 2026. Data checked November 10, 2026.
This job in your state or metro area
Embed these figures on your site
A small card with the share of working time within reach now and at the end of 2027, 2028 and 2030, the date and a link back. Free under CC BY 4.0; the card carries the credit.
Your own version of this job
The page above is the occupation. Your job is your title and what you actually do: give both, and they are read against these tasks in your browser, nothing sent anywhere. The headline is free: how much of your working time is within reach now and by the end of 2028, and how many of the tasks your description names.
The full personal report is $19, to print or save as a PDF: every task with when it comes within reach, which ones your description names, the parts of your description the task list does not cover, what stays with you, what becomes checking the machine and what to learn.