<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/harvard-researchers-find-ai-coding-agents-shifted-the-bottleneck-to-code-review-aeapqg8jp" -->

---
title: Harvard researchers find AI coding agents shifted the...
description: AI coding agents increased code output without increasing finished work, a Harvard working paper finds. Engineers at adopting firms wrote about 30 percent more...
canonical: https://daily.dev/posts/harvard-researchers-find-ai-coding-agents-shifted-the-bottleneck-to-code-review-aeapqg8jp
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Harvard researchers find AI coding agents shifted the bottleneck to code review | daily.dev
og:description: AI coding agents increased code output without increasing finished work, a Harvard working paper finds. Engineers at adopting firms wrote about 30 percent more...
og:url: https://daily.dev/posts/harvard-researchers-find-ai-coding-agents-shifted-the-bottleneck-to-code-review-aeapqg8jp
og:image: https://api.daily.dev/og/posts/AeaPQG8jP.png
og:image:alt: Harvard researchers find AI coding agents shifted the bottleneck to code review
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Harvard researchers find AI coding agents shifted the bottleneck to code review

**[Hacker News](https://daily.dev/sources/hn)** · 15 min read · 0 upvotes · 0 comments

## Summary

AI coding agents increased code output without increasing finished work, a Harvard working paper finds. Engineers at adopting firms wrote about 30 percent more code, but completed Jira issues barely moved, a result statistically indistinguishable from zero. Code review became the bottleneck: time from pull request to merge rose 49 percent, formal change requests nearly doubled from a 13 percent baseline, and review comments rose 35 percent. The study covers 725,938 workers at 718 firms using Jellyfish and Revelio Labs data. AI review tools did not remove the bottleneck, generating only 23.3 percent of review comments even after wide adoption. A separate NBER paper on GitHub developers found a similar pattern, with a 240 percent rise in agent commits shrinking to 30 percent for actual releases.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://complexdiscovery.com/harvard-researchers-find-ai-coding-agents-shifted-the-bottleneck-to-code-review>

## Questions this post answers

### Does using AI coding agents actually increase the amount of finished, deployed work?

Not measurably, according to a Harvard working paper covering 725,938 workers at 718 firms. Lines of code rose about 30 percent per worker-month, but completed Jira issues (deployed to production) moved only 0.12 per worker-month on a 3.67 baseline, a result statistically indistinguishable from zero, with the 95 percent confidence interval ruling out gains above roughly 12 percent.

_Teams weighing AI coding agent adoption can track productivity debates like this one on daily.dev._

### Why did code review take longer after engineers started using AI coding agents?

Pull request merge time rose 49 percent (from a 7.03-day baseline to 3.45 days longer), the Harvard study found, because faster drafting sent more code to a fixed pool of reviewers and because AI-written code appears to need more scrutiny: formal change requests nearly doubled from a 13 percent baseline and review comments per pull request rose 35 percent, while AI review tools generated only 23.3 percent of comments even after wide adoption.

_Anyone budgeting review capacity around AI-generated code can follow this kind of data on daily.dev._

### Did AI coding agents lead to developer layoffs or job cuts, according to research?

No measurable employment decline was found. The Harvard working paper's estimates rule out an overall employment decline larger than 2.9 percent and an engineering employment decline larger than 13.8 percent, though the authors caution these are short-run, partial-equilibrium estimates that could mask effects at competing firms.

_Developers tracking how AI coding tools affect hiring and team structure can follow studies like this on daily.dev._

## Community take

How the wider developer community reacted, aggregated from 1 discussion and 229 comments across hackernews (as of 2026-11-18).

**TL;DR:** Developers broadly confirm the study's finding: reviewing AI-generated code is exhausting, decision fatigue has replaced coding as the daily grind, and many teams are quietly skipping or rubber-stamping reviews under schedule pressure, with some worrying this is eroding both code quality and system understanding.

**Sentiment:** 10% positive · 25% mixed · 65% skeptical

**The case for**

- A few people report real productivity gains by developing strict personal workflows/guidelines that reduce how much they need to review.
- Some argue reviews were never great at building system understanding anyway, so the shift just exposes a pre-existing problem rather than creating a new one.
- One commenter built tooling (local review + MCP, decision/requirement tracking) to make reviewing agent output more manageable.

**The pushback**

- Reviewing AI code all day is described as uniquely exhausting decision fatigue compared to the rhythm of writing code interspersed with review.
- Developers report bypassing review entirely by having the same person prompt, commit, and approve their own agent's PRs.
- Management pressure to not be 'the bottleneck' pushes reviewers toward shallow rubber-stamp approvals, risking major outages.
- Nobody ends up understanding the system deeply since reviews happen at the PR/diff level rather than the architecture level.
- AI agents are seen as producing overengineered, bloated, or brittle code (e.g., duplicated hardcoded logic) that's hard to maintain even when it 'works'.
- Some feel code review has become theater: AI reviews AI-generated code, author's agent 'fixes' the comments, repeat, without real scrutiny.

**By community**

- hackernews (heated): Strong agreement that review has become an exhausting bottleneck, but sharp disagreement over causes (AI itself vs. bad specs/process vs. organizational incentives) and over whether dropping traditional code review is a sane path forward.

**Hottest debate:** Whether the real root cause is AI-generated code quality itself, or pre-existing organizational/specification failures that AI has merely exposed and accelerated.

**Open questions**

- Can human decision-making/review capacity actually scale to match AI-accelerated code output, or is this a hard human limit?
- What should code review even look like going forward — line-by-line PR review, continuous whole-system review, or agent-vs-agent review with humans only spot-checking?
- How do teams keep security and correctness guarantees once nobody fully understands the system anymore?

**Highlights**

> > how little anyone understands the code Everyone's doing code reviews--well, not even code reviews; just PR reviews--so how are they supposed to gain an understanding of the system. I review a PR, I see that one new endpoint was added to the API. It looks correct. I say "LGTM" in a GitHub comment. The sprint's almost over, everyone's in a rush to do code reviews. I look at another PR. Somebody added a button to the UI. The button looks correct, it's connect to some callbacks that look correct. They get a thumbs-up emoji from me. I bet they feel satisfied. I look at another PR. Somebody changed a timeout in the settings because some people complained about dropped connections. It's a really small change, so, approved. At what point am I supposed to understand the system? I looked at a single API endpoint, in isolation, and then a button, and then a harmless looking settings change. I was pressured to hurry. Am I supposed to somehow get an understanding of the big picture through this? Imagine if instead there was time to review large logical sections of the code together. Imagine if we focused our reviews on the overall code instead of on the lines in one random PR at a time.
> — [Buttons840 on hackernews · 2 comments](https://news.ycombinator.com/item?id=50541860)

> The error in the position is that your definition of mediocre was grounded in the pre AI era. You are then applying failure to do review properly as a standard to identify mediocre firms, and now all firms are mediocre. This is a No True Scotsman argument. The prescription you suggest is not dependent on its inspiration, so it can still be discussed. All firms right now, even those with better than average cultures, are dealing with AI code workflows. Previously the volume of code generated over time was lower, even mediocre code review was useful enough over all the code that people generated. Today, the ratio of verification capacity to generation capacity has shifted drastically. Even with automated tooling, the amount of verification isn’t enough to stay on top of the generated code. This changes only when leadership values reviewer capacity, or builds a triage/priority list for code which must be understood.
> — [intended on hackernews](https://news.ycombinator.com/item?id=50542146)

> In my company, people found out that if they use an agent that commits and do the pull/merge request, they can approve it themselves and bypass a second set of eyes. Since they are the one doing the prompting, they quickly got lazy of actually reviewing the output from the LLM which means that nobody is reviewing anything. It hasn't happened yet but I guess that is how you end up with major outages.
> — [prmoustache on hackernews · 2 comments](https://news.ycombinator.com/item?id=50539720)

> It's decision fatigue. The decisions used to broken up by stretches of coding, now it's the whole job. It's exhausting.
> — [0x696C6961 on hackernews · 6 comments](https://news.ycombinator.com/item?id=50541202)

> > Your goal should be to automate and remove yourself out of the process as much as possible while maintaining quality, while also remaining confident that you will still be valuable. I'm working on a setup which will allow non-software developers to develop software in a manner that can be pushed safely to production on our architecture. Not because it's a good idea but because it's what top management wants to happen. We've seen a massive influx of tools developed inside our office by 100% vibe code and no reasonable security or scalability, but because they run on someones computer and they do geniunly useful stuff, management wants to use it. With how AI is evolving in enterprise you can do more and more for these non-developers because you can set their agents up to write code toward arcitecture without their person knowing anything about tech. I'm quite sure it'll eventually lead to us letting go of any real software engineers, because they are slow. I'm also sure it'll fail spectacularly in a few years. It's already sort of failing. You can have someone spend €10000 in a month writing a heap of shit that does some good stuff, and then €80000 having consultants actually getting it to run decently in your cloud architecture. Which is stupid, but none of those expenses are in a budget because it's not considered a recurring cost. Unlike someone's salary. I'm quite sure society on a large scale has no idea the storm that's going to come. Because this isn't just happening in software engineering. It's happening in any white collar profession.
> — [Quothling on hackernews](https://news.ycombinator.com/item?id=50542453)

**Source threads**

- [hackernews](https://news.ycombinator.com/item?id=50538303) · 161 points · 229 comments

## Similar posts on daily.dev

- [Agentic Code Review](https://daily.dev/posts/agentic-code-review-0ue8fcsa1) · Addy Osmani · 148 upvotes · 14 comments
- [Study Finds AI-generated Code Fails to Boost Productivity Due to Downstream Bottlenecks](https://daily.dev/posts/study-finds-ai-generated-code-fails-to-boost-productivity-due-to-downstream-bottlenecks-zk3tkxmhn) · InfoQ · 2 upvotes · 0 comments

---

Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#productivity](https://daily.dev/tags/productivity), [#code-review](https://daily.dev/tags/code-review), [#legal](https://daily.dev/tags/legal)

[View this post on daily.dev](https://daily.dev/posts/harvard-researchers-find-ai-coding-agents-shifted-the-bottleneck-to-code-review-aeapqg8jp)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Harvard researchers find AI coding agents shifted the bottleneck to code review","url":"https://daily.dev/posts/harvard-researchers-find-ai-coding-agents-shifted-the-bottleneck-to-code-review-aeapqg8jp","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/harvard-researchers-find-ai-coding-agents-shifted-the-bottleneck-to-code-review-aeapqg8jp"},"datePublished":"2026-11-18T04:14:42.532Z","dateModified":"2026-11-18T15:54:25.451Z","description":"AI coding agents increased code output without increasing finished work, a Harvard working paper finds. Engineers at adopting firms wrote about 30 percent more...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/88791705a56118bb0763b077a4ae8dcb?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/88791705a56118bb0763b077a4ae8dcb?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hacker News","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Hacker News","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/hn","url":"https://daily.dev/sources/hn"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/harvard-researchers-find-ai-coding-agents-shifted-the-bottleneck-to-code-review-aeapqg8jp","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,productivity,code-review,legal","timeRequired":"PT15M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hacker News","item":"https://daily.dev/sources/hn"},{"@type":"ListItem","position":3,"name":"Harvard researchers find AI coding agents shifted the bottleneck to code review"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/harvard-researchers-find-ai-coding-agents-shifted-the-bottleneck-to-code-review-aeapqg8jp#faq","mainEntity":[{"@type":"Question","name":"Does using AI coding agents actually increase the amount of finished, deployed work?","acceptedAnswer":{"@type":"Answer","text":"Not measurably, according to a Harvard working paper covering 725,938 workers at 718 firms. Lines of code rose about 30 percent per worker-month, but completed Jira issues (deployed to production) moved only 0.12 per worker-month on a 3.67 baseline, a result statistically indistinguishable from zero, with the 95 percent confidence interval ruling out gains above roughly 12 percent. Teams weighing AI coding agent adoption can track productivity debates like this one on daily.dev."}},{"@type":"Question","name":"Why did code review take longer after engineers started using AI coding agents?","acceptedAnswer":{"@type":"Answer","text":"Pull request merge time rose 49 percent (from a 7.03-day baseline to 3.45 days longer), the Harvard study found, because faster drafting sent more code to a fixed pool of reviewers and because AI-written code appears to need more scrutiny: formal change requests nearly doubled from a 13 percent baseline and review comments per pull request rose 35 percent, while AI review tools generated only 23.3 percent of comments even after wide adoption. Anyone budgeting review capacity around AI-generated code can follow this kind of data on daily.dev."}},{"@type":"Question","name":"Did AI coding agents lead to developer layoffs or job cuts, according to research?","acceptedAnswer":{"@type":"Answer","text":"No measurable employment decline was found. The Harvard working paper's estimates rule out an overall employment decline larger than 2.9 percent and an engineering employment decline larger than 13.8 percent, though the authors caution these are short-run, partial-equilibrium estimates that could mask effects at competing firms. Developers tracking how AI coding tools affect hiring and team structure can follow studies like this on daily.dev."}}]}
```

