AI can write more code, but that does not make software engineering a hands-off job—or turn every engineer into a supervisor. The evidence points to a more qualified shift: AI can help developers complete work, while people still have to direct it, judge its output and fit it into a delivery system. Whether that help improves the finished product depends on the task, the tools and the team around them.
Why do new engineers act like foremen?
The “foreman” metaphor captures a real change in how some engineering work can be done: instead of writing every line themselves, developers can ask an AI assistant to draft code, explain an unfamiliar module or suggest a fix. They then need to decide what to ask for, inspect what comes back, test it and integrate it with the rest of the system.
As an Amazon Associate I earn from qualifying purchases.
That is not the same as supervising a factory that runs itself. AI-generated code still has to meet the requirements, work with existing software and behave correctly in situations the prompt may not have covered. The engineer’s role may include more direction and review, but the evidence does not show that coding has stopped being meaningful work or that engineers have universally become supervisors.
Recommended Free Tools
DORA’s 2025 report offers a useful way to think about the organizational side: it describes AI as an “amplifier,” magnifying an organization’s existing strengths and weaknesses. The implication is that a coding assistant does not, by itself, repair unclear priorities, brittle processes or inadequate testing. The system in which a team works shapes what it gets from the tool.
#1 Best Overall
Does AI make software developers more productive?
It can, in some settings—but the available results do not support one productivity figure that applies to all developers or projects. Studies measure different outcomes, involve different participants and take place in different kinds of work. A higher task count, faster completion of familiar repository issues and a developer’s view of AI are not interchangeable measures.
| Evidence | Participants and setting | What it measured | Reported result |
|---|---|---|---|
| Microsoft Research, June 2025 | Randomized workplace field experiments at Microsoft, Accenture and an anonymous Fortune 100 company; pooled analysis of 4,867 developers. | Completed tasks among developers using an AI coding assistant. | 26.08% increase in completed tasks (SE: 10.3%) in the combined analysis. Less experienced developers had higher adoption and greater productivity gains. |
| METR, July 10, 2025 | 16 experienced developers who had contributed for years to large open-source repositories; 246 real issues. | Time to complete repository tasks when AI use was allowed. | Developers took 19% longer with AI allowed in this trial. |
| DORA / Google, September 2025 | Nearly 5,000 technology professionals surveyed globally. | Respondents’ reported use of and trust in AI. | 90% of software development professionals reported AI adoption; 65% reported heavy reliance on AI for software development; 30% reported little or no trust in AI outputs. |
The Microsoft result is about task counts in workplace experiments, not a universal percentage reduction in coding time. METR’s result concerns experienced contributors working in repositories they knew well; it is not evidence that AI slows every developer. Its authors described the July 2025 result as a snapshot of early-2025 tools and said it does not establish that AI fails to speed up most developers. The METR page also notes that new data was published in February 2026; the July result alone cannot establish how later tools perform.
DORA / Google’s adoption and trust figures are survey responses, not controlled measurements of output, speed or correctness. Read alongside the experiments, they show that AI use is widespread while perceptions and measured results remain different kinds of evidence—not competing scores on one scale.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Why can faster coding fail to improve software delivery?
Finishing a coding task is only one part of delivering software. A change also has to be reviewed, integrated, tested and released without creating new problems for users or other teams. If code arrives faster than a team can check or safely ship it, individual productivity gains need not translate into better delivery.
DORA’s 2024 report describes this tension directly: AI adoption significantly increased individual productivity, flow and job satisfaction, while negatively affecting software delivery stability and throughput. Its guidance emphasizes end-user focus, stable priorities, small batches and robust testing. These practices matter because they connect the local work of producing code to the larger job of delivering dependable changes.
- Keep the user outcome in view. More generated code is not itself proof that a change solves the right problem.
- Maintain stable priorities. Frequent shifts in direction can make it harder to turn quick implementation into coherent delivery.
- Ship in small batches. Smaller changes are easier to inspect and integrate than a large pile of generated work.
- Test robustly. Code that looks plausible still needs checks against expected behavior and possible regressions.
What the “foreman” metaphor gets right—and what it misses
The metaphor is useful when it points to a greater need for direction and inspection: developers may spend less time producing a first draft and more time deciding what to build, evaluating suggestions and making changes safe to merge. But it becomes misleading if it implies that software engineering is now just assigning tasks to AI.
Giving useful direction requires understanding the problem and the existing system. Reviewing a proposed change requires enough technical judgment to spot incorrect assumptions, missing cases and integration risks. The engineer remains accountable for whether the change fits its purpose; delegating a draft does not delegate that judgment.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsNor do the available studies establish a universal change in job titles or daily duties. Microsoft Research found greater adoption and productivity gains among less experienced developers in its field experiments, while METR studied experienced open-source contributors doing work in repositories they knew well. Those results describe particular groups and contexts, not a single career path for every engineer.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Will AI replace software engineers?
The evidence here does not show that AI has emptied the software factory or made engineering expertise unnecessary. It shows that an assistant can help with some work, that results vary by context and measure, and that delivery still depends on human decisions and organizational practices. The practical question is less whether AI can produce code than whether a team can use that output to make useful, reliable software.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
