Sometimes—but the evidence does not support a universal productivity boost. Results vary with the task, developer, tool and measurement. A controlled test of experienced open-source developers found slower completion with early-2025 tools, while a UK government workplace trial reported time savings and a GitHub study found faster completion of a bounded programming task. Those findings measure different things, so the useful question is whether an assistant improves end-to-end results for the work your team actually does.
What the studies found
These studies differ in design, population and outcome. Their figures are informative within those boundaries, but they cannot be averaged into one productivity percentage.
| Study | Setting and design | Reported result | What the result represents |
|---|---|---|---|
| METR, 2025 | Randomized trial; 16 experienced developers with moderate AI experience completed 246 tasks in mature open-source projects where they had an average of five years of prior experience. The tools were those available at the February–June 2025 frontier. | Participants took 19% longer on average with the AI tools in this study. | Measured task completion in this sample and familiar-codebase setting—not a general estimate for all developers or tools. |
| UK Department for Science, Innovation and Technology and Government Digital Service, 2025 | Workplace trial from November 2024 to February 2025; 2,500 licences were made available across central government organisations. | Participants reported an average of 56 minutes saved per working day, including 24 minutes on code creation and analysis. | Self-reported workplace time savings, not a randomized estimate of additional completed work. The licence count is access offered, not the number of daily users. |
| GitHub, 2022 | Vendor-published controlled study of a defined programming task. | Average completion time was 1 hour 11 minutes with Copilot and 2 hours 41 minutes without it. | Performance on that bounded task under study conditions, not proof of the same gain on complex production work. |
| Microsoft Research, 2025 | Three randomized field experiments involving developers at Microsoft, Accenture and an anonymous Fortune 100 company. | The existence and settings of the experiments are reported; there is no single generalized percentage to apply across them. | Workplace evidence whose individual estimates and outcomes should be considered separately. |
Why the results do not agree
The work and codebase matter
METR studied experienced developers working in mature repositories they already knew. That is a particularly relevant setting for maintenance and issue work in established projects, but it does not directly estimate what happens when a novice explores a new codebase or a team builds a greenfield feature. A tool can make a first draft faster while creating more checking, revision or integration work later; only an end-to-end measure captures that trade-off.
The measurement matters
Elapsed time to an accepted solution is not the same as time spent typing, a developer’s sense of speed, or the share of suggestions accepted or committed. The UK trial’s daily savings are useful evidence about workers’ reported experience, but a self-reported saving does not by itself show that the same amount of time translated into more completed work. Likewise, a controlled task result answers a narrower question than delivery across a production team.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
#1 Best Overall
Tools change over time
GitHub’s task study dates to 2022, and METR’s slowdown result concerns tools available in early 2025. Neither should be treated as a direct estimate of every later assistant or agent. In a February 24, 2026 update, METR said wider adoption had created selection effects in its follow-up study and that participants struggled to account for time while agentic systems ran in the background. METR said it was changing the experiment design; the update did not report a completed replacement result.
What the evidence does—and does not—establish
The evidence supports a conditional conclusion: an assistant can help on some tasks, but access to one does not guarantee faster end-to-end development. The reported 19% slowdown is a warning against assuming that experienced developers will automatically move faster in familiar, mature projects. The UK trial and GitHub task study show why it would also be too broad to conclude that assistants never save time.
None of these figures establishes a pooled productivity effect across products, teams or software work. Nor are speed and quality interchangeable: a faster draft is not a productivity gain if it fails tests, needs substantial rework or creates downstream maintenance costs. The cited results should be read as evidence about their own samples and outcomes, not as a forecast for an individual team.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to tell whether an assistant helps your team
Run a small, representative evaluation rather than relying on a headline percentage. Compare similar tasks with and without the assistant, and define the finish line before the trial begins.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →- Select representative work. Include the kinds of tasks the team actually handles—such as maintenance, debugging, feature work or tests—and record task type, codebase familiarity and developer experience.
- Set a consistent comparison. Use a reasonable no-assistant comparison group or baseline for similar tasks. Record the assistant, model, configuration and evaluation dates so a later tool change is not mistaken for a change in developer performance.
- Measure the whole task. Track time from start to accepted completion, including prompting, waiting, review, verification, revisions and follow-up fixes. Decide whether interruptions or background agent time count, and apply that rule consistently.
- Check quality alongside speed. Record whether the result meets the team’s normal acceptance criteria, including tests and review, and whether it causes rework. Keep completion time, acceptance, defects and developer impressions as separate measures.
- Compare like with like. Break results down by task type and experience level instead of collapsing unlike work into one average. A benefit on one category should not be presented as a team-wide effect if other categories show a different result.
This kind of evaluation cannot make a small team trial universal, but it can answer the decision that published studies cannot: whether the assistant improves the work your developers need to deliver under your team’s conditions.
Quick Recap
Best Value
Rank #4
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.
