When AI Coding Help Slows You Down

Introduction

AI coding assistants are often presented as a straightforward productivity boost: describe a task in plain language, receive working code, and finish sooner. The reality is more complicated. Evidence from controlled experiments, workplace studies and open-source development suggests that coding assistants can save substantial time on small, well-defined and easily tested tasks. However, those gains can shrink—or even reverse—when developers must review, verify and correct AI-generated output inside large, familiar codebases. The key issue is not whether the AI can generate code. It is whether the time spent checking that code is lower than the time required to write it directly. Research increasingly shows that the review burden is one of the most important factors determining whether AI coding assistance helps or hinders software development. [arXiv+2arXiv]arxiv.orgThis paper presents results from a controlled experiment with GitHub Copilot, an AI pair…Read more…

Where Coding Assistants Save Time on Testable Work

The strongest evidence for productivity gains comes from tasks with clear success criteria. In an influential controlled experiment, developers using GitHub Copilot completed a programming task roughly 56% faster than a control group. The task involved building a JavaScript HTTP server, where the output could be readily tested and verified. [arXiv]arxiv.orgThis paper presents results from a controlled experiment with GitHub Copilot, an AI pair…Read more…

Later studies have also reported measurable gains. Research involving thousands of developers found productivity improvements of around 26% when developers used Copilot during software development tasks. These results align with many developers’ day-to-day experiences: generating boilerplate code, creating unit tests, converting data formats, writing repetitive functions and explaining error messages are activities where AI often performs well. [InfoQ]infoq.comcopilot developer productivityStudy Shows AI Coding Assistant Improves Developer…24 Sept 2024 — The team conducted three separate randomized controlled trials…

Several characteristics make these tasks especially suitable for AI assistance:

• The desired output is narrow and clearly defined.
• Automated tests can quickly confirm correctness.
• The developer can evaluate suggestions without needing extensive project context.
• Mistakes are relatively easy to detect and repair.

In these situations, the AI acts less like an autonomous programmer and more like an autocomplete system operating at a larger scale. The cost of reviewing the output remains low enough that the generated code produces a net time saving. [arXiv+2arXiv]arxiv.orgThis paper presents results from a controlled experiment with GitHub Copilot, an AI pair…Read more…

Why Familiar Complex Codebases Can Reverse the Gain

The most striking challenge to the “AI always makes programmers faster” narrative comes from research on experienced developers working in projects they already know well.

A 2025 randomised controlled trial by the research organisation METR studied 16 experienced open-source developers performing 246 real tasks in mature projects where they had, on average, five years of prior experience. Before the study, participants expected AI tools to reduce completion times by about 24%. After using the tools, they still believed they had worked faster. Yet the measured outcome showed the opposite: tasks took approximately 19% longer when AI assistance was allowed. [arXiv]arxiv.orgMeasuring the Impact of Early-2025 AI on Experienced Open-Source Developer ProductivityJuly 12, 2025…Published: July 12, 2025

This result appears surprising until the review burden is examined more closely. Developers reported spending substantial time:

Only a minority of generated suggestions were accepted without modification. Many outputs were directionally correct but failed to account for project-specific constraints, established design patterns or undocumented assumptions known to experienced maintainers. [Reuters+2TechRadar]reuters.comDespite initially believing AI would speed up their work by 24%, developers took 19% longer to complete tasks with AI assistance. The res…

Large repositories present a particular challenge because software projects contain far more context than can easily fit into a model’s working window. An AI may understand a function but miss the architectural reason it exists. It may generate code that compiles while violating conventions, introducing duplication or creating maintenance problems for future contributors. In such environments, experienced developers often already know where changes should be made and how systems interact. The AI’s suggestions therefore become something that must be audited rather than trusted. [arXiv+2Reuters]arxiv.orgMeasuring the Impact of Early-2025 AI on Experienced Open-Source Developer ProductivityJuly 12, 2025…Published: July 12, 2025

The result is a productivity trade-off: the time saved writing code may be offset by the time required to determine whether the generated code belongs in the project at all.

The Hidden Cost: Reading Is Often Harder Than Writing

One reason the review burden is easy to underestimate is that developers frequently feel more productive even when objective measurements show otherwise.

The METR study found that participants believed AI had helped them despite recorded slowdowns. Researchers suggested that reviewing generated code can feel easier than creating code from scratch, even when it consumes more total time. Developers experience less “blank page” effort, but they still bear responsibility for understanding and validating every change. [arXiv+2Reuters]arxiv.orgMeasuring the Impact of Early-2025 AI on Experienced Open-Source Developer ProductivityJuly 12, 2025…Published: July 12, 2025

This creates an important distinction between output volume and useful progress. AI can rapidly generate hundreds of lines of code. Yet every line introduces obligations:

If review effort grows faster than development speed, the apparent productivity gain can disappear. This is particularly important in professional environments where software quality matters more than raw code volume. [Tilburg University]tilburguniversity.eduai productivity gains may come expense quality and sustainabilityTilburg UniversityAI productivity gains may come at the expense of quality…5 Dec 2025 — AI tools such as GitHub Copilot can boost the…

When More Code Creates More Work for Experts

The review burden does not affect all developers equally. Emerging evidence suggests that AI-generated code may shift work from code authors to project maintainers.

Research examining open-source projects after the introduction of GitHub Copilot found increased productivity overall, especially among less-experienced contributors. However, the study also reported increased rework requirements and a greater review load for experienced maintainers. Core developers reviewed more code while their own original coding productivity declined. The researchers argued that apparent gains in output could conceal growing maintenance costs imposed on the people responsible for long-term project quality. [arXiv]arxiv.orgOpen source on arxiv.org.

Other analyses have raised similar concerns. Studies of AI-assisted development have reported increases in duplicated code, short-term code churn and patterns associated with future maintenance effort. While these findings do not mean AI-generated code is inherently poor, they suggest that speed gains can be accompanied by quality-management costs that appear later in the software lifecycle. [GitClear]gitclear.comai assistant code quality 2025 researchAI Copilot Code Quality: 2025 Data Suggests 4x Growth in…We observe a spike in the prevalence of duplicate code blocks, along…

For teams, this means productivity cannot be measured solely by how quickly code is produced. It must also include the effort required to review, revise and support that code over time.

How Tests, Documentation and Review Keep Output Useful

The practical lesson is not that coding assistants fail. Rather, their value depends heavily on verification.

Teams that obtain the most benefit from AI-assisted coding typically rely on mechanisms that reduce review costs:

Automated Tests Catch Obvious Errors

Comprehensive unit, integration and regression tests allow developers to validate AI-generated code quickly. The easier it is to test an output, the more likely the AI provides a net productivity gain. [arXiv]arxiv.orgThis paper presents results from a controlled experiment with GitHub Copilot, an AI pair…Read more…

Documentation Supplies Missing Context

AI systems often lack project-specific knowledge. Clear architectural documentation, coding standards and design records help developers evaluate whether generated suggestions fit the broader system rather than merely solving the immediate problem. [arXiv]arxiv.orgAugust 17, 2025…Published: August 17, 2025

Human Review Remains Essential

AI review tools are improving, but current evidence suggests they should complement rather than replace human judgement. Research evaluating AI code-review systems has found limitations in detecting important security vulnerabilities, while security professionals continue to express concern about relying solely on AI-generated or AI-reviewed code. [arXiv]arxiv.orgGitHub's Copilot Code Review: Can AI Spot Security Flaws Before You Commit?September 17, 2025…Published: September 17, 2025

The most reliable workflow treats AI output as a draft. Developers remain responsible for understanding what the code does, why it works and whether it belongs in the codebase.

The Real Question Is Not Speed but Verification Cost

The debate over AI coding assistants is often framed as a question of whether they make developers faster. Evidence suggests a more useful question: how expensive is verification?

When tasks are small, isolated and easy to test, AI-generated code can deliver substantial productivity gains. When work involves large repositories, specialised knowledge and long-term maintenance responsibilities, the review burden can become the dominant cost. In those situations, generating code is no longer the bottleneck; evaluating it is.

Understanding that distinction helps explain why different studies reach different conclusions. AI coding assistants are neither universal productivity machines nor universal slowdowns. Their impact depends on the balance between generation speed and verification effort—and in complex software projects, verification is often the harder part. [arXiv+3arXiv+3Reuters]arxiv.orgThis paper presents results from a controlled experiment with GitHub Copilot, an AI pair…Read more…

Amazon book picks

Further Reading

Books and field guides related to When AI Coding Help Slows You Down. Use these as the next step if you want deeper reading beyond the article.

Book

Co-Intelligence

By Ethan Mollick

Addresses how humans collaborate with AI tools.

See on Amazon

Book

Clean Code

By Robert C. Martin

Highlights maintainability and review costs in coding work.

See on Amazon

Book

The Pragmatic Programmer

By Andrew Hunt, David Thomas

Rating: 4.5/5 from 7 Google Books ratings

Emphasises practical judgment, review and productivity.

See on Amazon

Book

Accelerate

By Nicole Forsgren, PhD, Jez Humble et al.

Focuses on measurable productivity rather than perceived speed.

See on Amazon

Browse more on Amazon: Co Intelligence Clean Code The Pragmatic Programmer

As an Amazon Associate I earn from qualifying purchases.

eBay marketplace picks

Marketplace Samples

Example marketplace items related to this page. Use the search link to explore similar finds on eBay.

Shop location

Using USA USA USAUKCanadaAustraliaIreland

Example eBay listing

Never Give Up Coding Framed Wall Art Poster Canvas Print Picture

Search eBay.co.uk: coding poster

Browse similar on eBay.co.uk

Example eBay listing

Coding Coder Framed Wall Art Poster Canvas Print Picture

Search eBay.co.uk: coding poster

Browse similar on eBay.co.uk

Example eBay listing

Phonetic Alphabet Print International Code Informative Education Poster Referenc

Search eBay.co.uk: coding poster

Browse similar on eBay.co.uk

Example eBay listing

I'd Rather Be Coding Framed Art Pri Framed Wall Art Poster Canvas Print Picture

Search eBay.co.uk: coding poster

Browse similar on eBay.co.uk

Browse more on eBay.co.uk

Example items shown for inspiration; availability and pricing can change. Branchoria may earn a commission if you purchase through outbound eBay links.

Endnotes

1. Source: arxiv.org
   Link: https://arxiv.org/abs/2302.06590
   Source snippet

   This paper presents results from a controlled experiment with GitHub Copilot, an AI pair...Read more...

2. Source: arxiv.org
   Link: https://arxiv.org/pdf/2302.06590
   Source snippet

   The Impact of AI on Developer Productivityby S Peng · 2023 · Cited by 1251 — We present a controlled trial of GitHub Copilot, an AI pair...

3. Source: reuters.com
   Link: https://www.reuters.com/business/ai-slows-down-some-experienced-software-developers-study-finds-2025-07-10/
   Source snippet

   Despite initially believing AI would speed up their work by 24%, developers took 19% longer to complete tasks with AI assistance. The res...

4. Source: infoq.com
   Title: copilot developer productivity
   Link: https://www.infoq.com/news/2024/09/copilot-developer-productivity/
   Source snippet

   Study Shows AI Coding Assistant Improves Developer...24 Sept 2024 — The team conducted three separate randomized controlled trials...

5. Source: github.blog
   Link: https://github.blog/news-insights/research/research-quantifying-github-copilots-impact-on-developer-productivity-and-happiness/
   Source snippet

   The GitHub Blogquantifying GitHub Copilot's impact on developer...7 Sept 2022 — In our research, we saw that GitHub Copilot supports fas...

6. Source: arxiv.org
   Link: https://arxiv.org/abs/2507.09089
   Source snippet

   Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer ProductivityJuly 12, 2025...

   Published: July 12, 2025

7. Source: techradar.com
   Link: https://www.techradar.com/pro/using-ai-might-actually-slow-down-experienced-devs
   Source snippet

   Conducted on 16 seasoned developers working on 246 tasks across familiar open-source projects, the study revealed that while developers i...

8. Source: time.com
   Title: In the Loop: AI Promised Faster Coding
   Link: https://time.com/7302351/ai-software-coding-study/
   Source snippet

   This Study DisagreesA recent METR study challenges the assumption that AI accelerates software development. In tests with 16 experienced...

9. Source: arxiv.org
   Link: https://arxiv.org/abs/2510.10165

10. Source: gitclear.com
    Title: ai assistant code quality 2025 research
    Link: https://www.gitclear.com/ai_assistant_code_quality_2025_research
    Source snippet

    AI Copilot Code Quality: 2025 Data Suggests 4x Growth in...We observe a spike in the prevalence of duplicate code blocks, along...

11. Source: arxiv.org
    Link: https://arxiv.org/abs/2508.12285
    Source snippet

    August 17, 2025...

    Published: August 17, 2025

12. Source: arxiv.org
    Link: https://arxiv.org/abs/2509.13650
    Source snippet

    GitHub's Copilot Code Review: Can AI Spot Security Flaws Before You Commit?September 17, 2025...

    Published: September 17, 2025

13. Source: techradar.com
    Title: Tech Radar Nearly all security bosses are worried about AI safety
    Link: https://www.techradar.com/pro/security/nearly-all-security-bosses-are-worried-about-ai-safety-with-a-third-saying-they-still-rely-on-manually-reviewing-code-before-launch
    Source snippet

    An overwhelming 90% of security leaders report active concerns about AI safety, particularly as AI coding tools become more widespread in...

14. Source: github.blog
    Title: copilot is moving to usage based billing
    Link: https://github.blog/news-insights/company-news/github-copilot-is-moving-to-usage-based-billing/
    Source snippet

    GitHub Copilot is moving to usage-based billing27 Apr 2026 — Copilot [Business]({{ 'business-adoption/' | relative_url }}): $30 in monthly AI Credits; Copilot Enterprise: $70 in mont...

15. Source: github.blog
    Link: https://github.blog/

16. Source: github.com
    Link: https://github.com/features/copilot
    Source snippet

    GitHub Copilot · Your AI pair programmerCode, command, and collaborate · Make your editor your most powerful accelerator · Ship faster wi...

17. Source: github.com
    Link: https://github.com/features/copilot/plans
    Source snippet

    GitHub Copilot · Plans & pricingPremium requests. Chat, agent mode, code review, cloud agent, and Copilot CLI use premium requests, with...

18. Source: github.com
    Title: Git Hub · Change is constant
    Link: https://github.com/
    Source snippet

    GitHub keeps you ahead. · GitHubFrom your first line of code to final deployment, GitHub provides AI and automation tools to help you bui...

19. Source: docs.github.com
    Title: usage based billing for individuals
    Link: https://docs.github.com/copilot/concepts/billing/usage-based-billing-for-individuals
    Source snippet

    github.comUsage-based billing for individualsGitHub AI Credits allowance by plan; Copilot Pro, $10 USD, 1,000; Copilot Pro+, $39 USD, 3...

20. Source: docs.github.com
    Title: copilot requests
    Link: https://docs.github.com/en/copilot/reference/copilot-billing/request-based-billing-legacy/copilot-requests
    Source snippet

    in GitHub Copilot (legacy)Each time Copilot reviews a pull request (when assigned as a reviewer) or reviews code in your IDE, 13 premium...

21. Source: docs.github.com
    Link: https://docs.github.com/en/copilot/get-started/plans
    Source snippet

    for GitHub CopilotThis free plan includes limited access to select Copilot features, allowing you to try AI-powered coding assistance at...

22. Source: docs.github.com
    Title: using copilot code review
    Link: https://docs.github.com/copilot/using-github-copilot/code-review/using-copilot-code-review
    Source snippet

    GitHub Copilot code reviewGitHub Copilot can review your code and provide feedback. Where possible, Copilot's feedback includes suggested...

23. Source: github.com
    Link: https://github.com/pricing
    Source snippet

    Pricing · Plans for every developerPay only for what you use with compute fees starting at $0.18/hr and storage fees at $0.07/GB per mont...

24. Source: docs.github.com
    Title: code review
    Link: https://docs.github.com/en/copilot/concepts/agents/code-review
    Source snippet

    GitHub Copilot code reviewEach time Copilot reviews a pull request or reviews code in your IDE, the interaction consumes AI credits. The...

25. Source: docs.github.com
    Title: individual plans
    Link: https://docs.github.com/en/copilot/concepts/billing/individual-plans
    Source snippet

    individual GitHub Copilot plans and benefitsPrice, Free, Free, $10 USD per month (free for some users), $39 USD per month; Real-time cod...

26. Source: docs.github.com
    Title: models and pricing
    Link: https://docs.github.com/copilot/reference/copilot-billing/models-and-pricing
    Source snippet

    and pricing for GitHub CopilotEach token is priced based on the model used, and the total is converted into AI credits, where 1 AI credit...

27. Source: github.com
    Link: https://github.com/pricing/calculator
    Source snippet

    Pricing CalculatorThe pricing calculator will help you get cost estimates to create, run, and store codespaces for a specific project wit...

28. Source: metr.org
    Title: 2025 07 10 early 2025 ai experienced os dev study
    Link: https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
    Source snippet

    Measuring the Impact of Early-2025 AI on Experienced...Jul 10, 2025 — We conduct a randomized controlled trial (RCT) to understand how e...

29. Source: metr.org
    Link: https://metr.org/
    Source snippet

    METR (pronounced 'meter') is a research nonprofit that scientifically measures whether and when AI systems might threaten catastrophi...

30. Source: metr.org
    Link: https://metr.org/blog/2026-02-24-uplift-update/
    Source snippet

    We are Changing our Developer Productivity Experiment...24 Feb 2026 — Our early 2025 study found the use of AI causes tasks to take 19%...

31. Source: arxiv.org
    Link: https://arxiv.org/html/2501.13282v1
    Source snippet

    Experience with GitHub Copilot for Developer Productivity...23 Jan 2025 — This paper presents a comprehensive evaluation of GitHub Copil...

32. Source: arxiv.org
    Link: https://arxiv.org/pdf/2507.09089
    Source snippet

    Measuring the Impact of Early-2025 AI on Experienced...by J Becker · 2025 · Cited by 136 — After completing the study, developers estima...

33. Source: github.dev
    Title: Git Hub.dev Setting up your web editor.Read more
    Link: https://github.dev/
    Source snippet

    Setting up your web editor.Read more...

34. Source: tilburguniversity.edu
    Title: ai productivity gains may come expense quality and sustainability
    Link: https://www.tilburguniversity.edu/current/press-releases/ai-productivity-gains-may-come-expense-quality-and-sustainability
    Source snippet

    Tilburg UniversityAI productivity gains may come at the expense of quality...5 Dec 2025 — AI tools such as GitHub Copilot can boost the...

35. Source: Wikipedia
    Link: https://en.wikipedia.org/wiki/METR
    Source snippet

    METRModel Evaluation and Threat Research (METR) (MEE-tər), is a nonprofit research institute, based in Berkeley, California, that eval...

Additional References

1. Source: businessinsider.com
   Link: https://www.businessinsider.com/ai-coding-tools-may-decrease-productivity-experienced-software-engineers-study-2025-7
   Source snippet

   Conducted with 16 seasoned developers familiar with open-source projects they had worked on for years, the study randomly assigned partic...

2. Source: blesssphere.com
   Link: https://blesssphere.com/ai-developer-productivity-github-copilot-results/
   Source snippet

   GitHub Copilot Results Show 55% Faster Development...GitHub/MIT research shows 55% productivity gain with Copilot. See real AI-powered d...

3. Source: reddit.com
   Link: https://www.reddit.com/r/programming/comments/1m25iw2/metr_study_finds_ai_doesnt_make_devs_as/
   Source snippet

   METR study finds AI doesn't make devs as productive as...Idk what point this is meant to make but any developer who is being slowed down...

4. Source: linkedin.com
   Link: https://www.linkedin.com/posts/thestevemitchell_measuring-the-impact-of-early-2025-ai-on-activity-7349786687230644225-OqTH
   Source snippet

   METR study: AI assistants slow down experienced...A study from METR shows that experienced developers working with AI assistants produce...

5. Source: linkedin.com
   Link: https://www.linkedin.com/posts/truetech-itservices-itconsulting_aicoding-digitaltransformation-githubcopilot-activity-7434515408646352896-VPlO
   Source snippet

   GitHub Copilot Boosts Developer Productivity 55% with AI...3 Mar 2026 — GitHub Copilot increases developer productivity 55%. AI transfor...

6. Source: axios.com
   Link: https://www.axios.com/2025/07/15/ai-coding-productivity-study
   Source snippet

   Examining experienced developers working with large open-source projects, the research found that AI often slows down coders rather than...

7. Source: odsc.medium.com
   Link: https://odsc.medium.com/new-study-finds-ai-tools-slow-experienced-developers-in-familiar-codebases-d92695f23665
   Source snippet

   Study Finds AI Tools Slow Experienced Developers in...Contrary to prior findings, the study discovered that experienced developers worki...

8. Source: linkedin.com
   Link: https://www.linkedin.com/posts/zencoderai_a-2025-study-from-metr-found-ai-tools-made-activity-7459611696064720897-2ioL
   Source snippet

   AI Tools Slow Experienced Developers 19% | Zencoder...A 2025 study from METR found AI tools made experienced developers 19% slower. Not...

9. Source: researchgate.net
   Link: https://www.researchgate.net/publication/397516588_Github_Copilot%27s_Impact_on_Developer_Productivity_A_Review_of_Early_Evidence
   Source snippet

   (PDF) Github Copilot's Impact on Developer Productivity28 Dec 2025 — This study evaluates the efficiency gains, areas for improvement, an...

10. Source: linkedin.com
    Link: https://www.linkedin.com/posts/benroyce_measuring-the-impact-of-early-2025-ai-on-activity-7350084067586961409-UIzT
    Source snippet

    AI tools slow down developers, not speed them upBack in July 2025, a study from METR made some noise: “Early-2025 AI coding tools slow ex...

    Published: July 2025