Converging AI and Human Roles

Role · Tech & Engineering

Software DevelopersSOC 15-1252.00

Software developers design, develop and modify software systems, in the definition of O*NET, the US Labor Department’s occupational database (SOC 15-1252.00). O*NET 31.0 lists 17 tasks for the role, from analyzing user needs and software requirements to determine feasibility of design, to testing and validation procedures, to modifying existing software to correct errors or improve performance.1 Bureau of Labor Statistics figures shown on O*NET count 1,717,800 software developers in the US in 2025, at a median wage of $65.38 an hour or $135,980 a year, with employment projected to grow much faster than average (7% or higher) from 2025 to 2035 and 95,300 projected job openings.1

AI reaches the job through coding agents and assistants, covered in do coding agents make developers faster? The firmest operator cases keep an engineer in the loop: Stripe says its Minions agents merge more than 1,000 pull requests a week with no human-written code, which another engineer reviews;2 Google’s CEO said in April 2026 that 75% of new code at Google is AI-generated and approved by engineers;3 and a Microsoft study of its early 2026 rollout of Claude Code and GitHub Copilot CLI reported 24.0% more merged pull requests per engineer per day, an observational finding with a stated confidence interval.4 Earlier, Meta reported that engineers accepted 73% of the unit-test improvements its TestGen-LLM tool recommended for production,5 and Google described AI-assisted code migrations in which 80% of the changed code in one case was AI-authored.6

The weaker cases are the planning and reporting tasks. GitHub says its cloud agent can research a repository to confirm assumptions before planning,7 Claude Code has a plan mode that proposes a change plan and makes no edits until the developer approves it,8 and GitHub documents pull-request summaries that users are told to review carefully;9 all three are vendor documentation, and no operator has published a result for them, so they are graded as vendor claims. Benchmarks such as SWE-bench Verified score models on isolated bug fixes rather than the whole job; see the independent results ranking. MetaRoles’ scorecard counts 1 of the 6 core tasks it grades as done by AI in production, where the AI completes the change and a reviewer approves afterwards.

Illustration: Software Developers, the role whose code-writing, testing and migration tasks AI coding agents now take on under engineers’ review.

Key facts

O*NET-SOC code15-1252.00 Software Developers1
O*NET tasks17 in O*NET 31.01
US employment, 20251,717,800 (BLS, via O*NET)1
Median wage, 2025$65.38 an hour; $135,980 a year (BLS, via O*NET)1
Projected growth, 2025–2035Much faster than average (7% or higher), with 95,300 projected job openings (BLS, via O*NET)1
Share of new code AI-generated75% at Google (company-stated, April 2026)3

AI use cases

As of

AI use cases for Software Developers, as of 2026-10-22
Task AI system Organisation Evidence grade Latest source Source
Fix flaky tests and other bugs and open pull requests with no human-written code Minions one-shot coding agents (Stripe) Stripe In production 2026-02 2
Generate much of the new code from engineers’ instructions, with engineers approving it AI coding tools Google In production 2026-04 3
Write and modify code from the command line across engineering teams Claude Code and GitHub Copilot CLI Microsoft In production 2026-07 4
Improve unit tests for existing test suites; engineers choose which to land TestGen-LLM (Meta) Meta (Instagram and Facebook) In production 2024-02 5
Migrate code between systems with AI-authored changes AI-assisted code migration tools Google In production 2025-01 6
Research a repository and plan a change before editing GitHub Copilot cloud agent GitHub documentation (vendor) Vendor claim 2026-10 7
Propose a change plan with no edits until the developer approves Claude Code plan mode Anthropic documentation (vendor) Vendor claim 2026-10 8
Draft pull-request summaries for reviewers GitHub Copilot pull-request summaries GitHub documentation (vendor) Vendor claim 2026-10 9

Sorted by evidence grade, then by date. Production means a named company says it runs the tool in its own engineering; the figures are company-stated except the Microsoft study, which is an observational study of its own rollout. Vendor-claim rows rest on the vendor’s own documentation, and their “latest source” is the date we retrieved it. How the grades work.

Appears in

Sources

  1. O*NET OnLine (O*NET 31.0; wages and employment from the Bureau of Labor Statistics), 15-1252.00 Software Developers, retrieved 2026-10-22.
  2. Stripe, Minions: Stripe’s one-shot, end-to-end coding agents, February 9, 2026.
  3. Google, Cloud Next ’26: Momentum and innovation at Google scale (CEO remarks), April 22, 2026.
  4. Murphy-Hill E, Butler J, Savelieva A (Microsoft), Adoption and Impact of Command-Line AI Coding Agents: A Study of Microsoft’s Early 2026 Rollout of Claude Code and GitHub Copilot CLI, arXiv, July 1, 2026.
  5. Alshahwan N, et al. (Meta), Automated Unit Test Improvement using Large Language Models at Meta, arXiv, February 14, 2024.
  6. Nikolov S, et al. (Google), How is Google using AI for internal code migrations?, arXiv, January 12, 2025.
  7. GitHub Docs, Research, plan, and iterate on code changes with Copilot cloud agent (vendor documentation), retrieved 2026-10-13.
  8. Anthropic, Claude Code Docs: Common workflows, plan before editing (vendor documentation), retrieved 2026-10-13.
  9. GitHub Docs, Creating a pull request summary with GitHub Copilot (vendor documentation), retrieved 2026-10-13.