Why code review is the new bottleneck
Fiona Chen and James Stratton of Harvard University studied engineering data from Jellyfish in their paper "Artificial Intelligence in the Firm: Bottlenecks in Software Production" (current version November 14, 2026). The data covers 445 million work events from GitHub, Jira and Google Calendar for 720,542 workers at 655 firms, from January 2021 to August 2026, with usage of GitHub Copilot, Cursor and Claude Code.
Twelve months after firms adopted AI agents, the paper measures 44.0% more lines of code, 36.0% more commits and 35.7% more pull requests, on top of the gains from AI assistants. Issues resolved rose 13.7%, a borderline significant result. Review slowed down: time from review submission to approval rose 56.1%, time waiting in the backlog 59.9%, active review time 58.5%, and comments per change 38.6%.
The abstract sums it up: "Code review becomes the bottleneck." The paper also finds that 79.1% of firms adopted AI code review tools but still largely keep humans in the loop. It is a working paper. Press coverage from October cites an earlier version with different numbers; the figures here are from the November version.
AI code review tools compared
Prices are in US dollars as listed on each vendor's pricing page or docs. "Not stated" means the official page we read doesn't say. Model usage is extra for the open-source tools.
| Tool | Open source | Platforms | Price | Free option | How you trigger it |
|---|---|---|---|---|---|
| Claude Code | No; the GitHub Action is MIT | GitHub; GitLab merge requests via /code-review --comment; GitHub Actions or GitLab CI/CD; local CLI | /code-review: counts toward plan usage. Ultrareview: typically $5 to $25 per review. Code Review (Team, Enterprise): token usage, no flat per-review price | 3 free ultrareview runs on Pro, Max and Ultra | /code-review in a session; @claude review on a PR; automatic after PR creation or every push |
| Open Code Review | Yes, Apache-2.0 | Local CLI (Windows, macOS, Linux); CI docs for GitHub Actions, GitLab CI, GitFlic CI, Gerrit | Free; you pay your model provider or host agent | Yes | ocr review, ocr scan, or as a skill inside your agent |
| Plannotator | Yes, Apache-2.0 or MIT | Local; GitHub PRs, GitLab MRs, Git, GitButler, Jujutsu, Perforce | Not stated | Open source | /plannotator-review in your agent |
| GitHub Copilot code review | No | GitHub, Azure DevOps, GitHub CLI, GitHub Mobile, VS Code, Visual Studio, Xcode, JetBrains | Pro $10/user/month, Pro+ $39, Max $100; Business $19 and Enterprise $39 per granted seat per month. GitHub estimates about $0.25 to $10 per review in AI credits | No; not in Copilot Free | Add Copilot as a reviewer; automatic via settings or rulesets |
| CodeRabbit | No | GitHub, GitLab, Azure DevOps, Bitbucket; CLI; VS Code, Cursor, Windsurf, JetBrains (beta) | Per user per month: Essentials $30 monthly or $24 billed annually; Team $60 or $48; Advanced $90 or $72 | Free for public repositories | Installed on the repo, it reviews PRs; CLI and IDE reviews |
| Cursor Bugbot | No | GitHub, GitLab, Bitbucket, Azure DevOps Services, Origin | Usage-based; Cursor says the average review costs $1.00 to $1.50. Listed on Cursor plans from $30/month; Teams $30+/user/month | No | Automatic on every PR update; comment cursor review or bugbot run |
| OpenAI Codex | Cloud review: no | GitHub, GitLab (beta), local CLI | ChatGPT Plus $20/month; Pro from $100/month; Business $20/user/month billed annually or $25 monthly | GitLab review (beta) on all ChatGPT plans | Comment @codex review; automatic in Codex settings; /review in the CLI |
| Graphite Agent | No | GitHub | Starter $20/user/month, Team $40/user/month, both billed annually | Hobby: personal repos, limited AI reviews | Runs on every PR |
| Greptile | No | Enterprise lists GitHub, GitLab Self-Managed, Bitbucket Data Center, Gitea, Perforce | Pro: $1 per credit, prepaid; a review costs 1, 3 or 10 credits | Starter: 50 credits/month, 1 developer; free for qualified open-source projects | Not stated |
| Qodo | Not stated (PR-Agent is separate) | GitHub, GitLab, Bitbucket, Azure DevOps; Gerrit on Enterprise | Pro Team $30, billed monthly; review credits $0.015 each | 14-day trial; open-source projects can apply | Not stated |
| PR-Agent | Yes, MIT | GitHub, GitLab, Bitbucket, Azure DevOps, Gitea | Free; you connect your own model | Yes | /review, /describe, /improve, /ask via GitHub Action, CLI, Docker or webhooks |
| Gemini Code Assist on GitHub | Not stated | GitHub; Enterprise Cloud and Server in the enterprise version (Preview) | Not stated on Google's review docs | Not stated | /gemini in PR comments |
| Sourcery | Not stated | GitHub, GitLab.com, self-managed GitLab; Enterprise adds GitHub Enterprise Server | Per developer per month: Pro $12 annual or $15 monthly; Team $24 annual or $30 monthly | Open Source plan, free for public repos | Not stated on the pricing page |
Open-source AI code review tools
If you want a free AI code review tool you can read and run yourself, four projects stand out. All are free software; you pay only for the model they call. Star counts are as of November 18, 2026.
- Open Code Review (Apache-2.0, 53.8k stars): Alibaba's
ocrCLI and Agent Skill. It reviews a Git diff with a model you configure, or runs inside Claude Code, Codex, Cursor, OpenCode and other agents using their model. - PR-Agent (MIT, 16.3k stars): a pull request reviewer that works on GitHub, GitLab, Bitbucket, Azure DevOps and Gitea, with OpenAI, Claude, Gemini, DeepSeek, Mistral or any LiteLLM model, including Ollama for local models.
- Plannotator (Apache-2.0 or MIT, 13.1k stars): a local browser view where you review your agent's plans and diffs and send annotations back to it.
- claude-code-action (MIT, 10.0k stars): Anthropic's GitHub Action that runs Claude in your own CI, with the Anthropic API, Amazon Bedrock, Google Vertex AI or Microsoft Foundry.
- Free plans for open source, not open-source tools: CodeRabbit reviews public repositories free, Sourcery has a free Open Source plan for public repos, and Greptile and Qodo let open-source projects apply for free access.
Claude Code: /code-review, ultrareview and Code Review
Anthropic offers four ways to get a review from Claude. /code-review (alias /review) is a bundled skill that reviews your diff, a PR number, a branch or a path for correctness bugs. Effort runs from low to max; --fix applies findings and --comment posts them on a GitHub PR or GitLab merge request. It runs in your session on any Claude Code plan and counts toward normal usage. /security-review checks your branch diff for vulnerabilities.
/code-review ultra (alias /ultrareview) is a research preview that sends a fleet of reviewer agents to a cloud sandbox. Each finding is independently reproduced and verified, and a run typically takes 5 to 15 minutes. Pro, Max and Ultra get 3 free runs per account; after that a review typically costs $5 to $25 in usage credits, depending on the size of the change.
Code Review is a managed GitHub service for Team and Enterprise. Several agents search in parallel, a verification step filters false positives, and findings are ranked by severity and posted as inline comments. It never approves or blocks a PR. It's billed on token usage with no flat per-review price, and an org-wide monthly spend limit is available. Anthropic says it may miss bugs, and it doesn't build or run your code. For setup and review rules, see Claude Code code review and Claude Code security review.
Open Code Review (Alibaba)
Open Code Review began as Alibaba Group's internal code review assistant. It combines deterministic pipelines with an LLM agent: ocr review reads staged, unstaged or untracked changes, a branch range or a commit, and returns line-level comments; ocr scan reviews whole files. Install it with npm install -g @alibaba-group/open-code-review or as a skill with npx skills add alibaba/open-code-review.
In native mode it calls an OpenAI- or Anthropic-compatible provider you configure. In delegation mode it uses the host agent's model and login, so its cost and limits are your agent's. The README claims higher precision and F1 than Claude Code with about a ninth of the tokens on its own benchmark of 200 PRs, and says its recall is lower than general-purpose agents. It needs Git 2.41 or later.
Plannotator
Plannotator is a local, browser-based review surface for coding agents. When the agent leaves plan mode, its plan opens in the browser to annotate, approve or deny. /plannotator-review opens your uncommitted changes or a PR or MR URL as a diff with a file tree; your comments go back to the agent. Guided reviews sort a diff into chapters, most important first, and it can launch AI reviews that comment on the diff.
It works with Claude Code (through the plugin marketplace), OpenCode, Pi, Codex, Copilot CLI, Gemini CLI, Kiro, Droid, Amp and DeepSeek Harness, using whichever agent you run. Shared links expire after 7 days. Pricing isn't stated; the README mentions a waitlist for team Workspaces. If you're looking for Plannotator alternatives, Open Code Review and /code-review review the diff for you, while Plannotator is built for you to review it.
GitHub Copilot code review
Copilot code review sits in GitHub's normal reviewer flow: add Copilot from the Reviewers menu, or turn on automatic reviews for a user, repository or organization, or through rulesets, with options to review new pushes and drafts. It reviews any language and also works in Azure DevOps, the GitHub CLI and the main IDEs.
It's on Copilot Pro ($10/user/month), Pro+ ($39), Max ($100), Business ($19 per granted seat per month) and Enterprise ($39 per granted seat per month), not on Copilot Free. Usage is measured in GitHub AI Credits (1 credit = $0.01); GitHub estimates about $0.25 per review at Lite, $1.50 at Balanced and $10 at Max, before Actions minutes. You can't switch models. It skips dependency files, logs and SVGs, and GitHub says it's not guaranteed to spot all problems. Related: GitHub Copilot in Slack.
CodeRabbit
CodeRabbit reviews pull requests on GitHub, GitLab, Azure DevOps and Bitbucket, and also reviews in a CLI (which works with Claude Code, Cursor, Codex and Gemini) and in VS Code, Cursor, Windsurf and JetBrains (beta). Every plan includes PR, CLI and IDE reviews, learnings, pre-merge checks and 1-click fixes; Team adds unit test generation, and Advanced adds security prioritization and architecture impact review.
Per user per month: Essentials $30 monthly or $24 billed annually, Team $60 or $48, Advanced $90 or $72, Enterprise by quote. You pay only for developers who open PRs. Plans link 1, 5, 10 or 20 repos, and a fair usage policy limits reviews to 5 to 12 per developer per hour. Overage is $0.25 per reviewed file. Public repositories are reviewed free.
Cursor Bugbot
Bugbot is Cursor's PR reviewer. It runs on every PR update, or when you comment cursor review or bugbot run, and covers GitHub, GitLab (including self-hosted), Bitbucket (including Data Center), Azure DevOps Services and Origin. Autofix starts a Cursor Cloud Agent to fix what it found, billed at Cloud Agent rates; it isn't available on Azure DevOps. Rules live in .cursor/BUGBOT.md.
There's no separate Bugbot subscription: it's usage-based, and Cursor says the average review costs $1.00 to $1.50. Cursor's pricing page lists it on plans from $30/month (and $100, $200), not on Free or the $8/month plan; Teams start at $30+/user/month. Cursor says 70%+ of its flags get resolved before merge. It skips PRs over a default size limit.
OpenAI Codex code review
Codex reviews GitHub pull requests in Codex cloud when you comment @codex review (with optional focus text) or turn on automatic reviews. On GitHub it flags only P0 and P1 issues, and it reads review rules from a ## Code Review Rules section in the nearest AGENTS.md. GitLab merge request review is in beta, on GitLab.com and self-managed or Dedicated 19.0+. In the Codex CLI, /review reviews against a base branch, uncommitted changes or a commit, without modifying your working tree.
ChatGPT Plus ($20/month) lists automatic code review; Pro starts at $100/month; Business is $20/user/month billed annually (2+ users) or $25 monthly. GitLab review is available on all ChatGPT plans. Signing in with an API key gives no cloud review. See Codex code review for setup.
Graphite Agent
Graphite's AI reviews, powered by Graphite Agent, run on every PR with no setup and focus on bugs such as logic errors, edge cases and performance rather than style, with suggested fixes. It's GitHub only (Enterprise adds GitHub Enterprise Server). Hobby is free for personal repos with limited AI reviews; Starter is $20/user/month and Team $40/user/month, both billed annually; Team includes unlimited AI reviews.
Greptile
Greptile prices reviews by credit instead of by seat. Starter is free with 50 credits a month for one active developer. Pro is $1 per credit, prepaid, with no seat fees: a Base review costs 1 credit, Plus 3 and Apex 10. Pro has a 14-day free trial. Enterprise is custom and can be self-hosted, with GitHub Enterprise, GitLab Self-Managed, Bitbucket Data Center, Gitea and Perforce listed. Qualified non-commercial open-source projects can apply for free use.
Qodo
Qodo supports GitHub, GitLab, Bitbucket and Azure DevOps, plus Gerrit on Enterprise. Pro Team is $30, billed monthly with no annual commitment, and the page says it's designed for up to 30 users. Reviews draw from pooled credits at $0.015 each: 2,000 credits is about 14 reviews a month, 4,000 about 29 and 16,000 about 114. There's no permanent free tier; there's a 14-day trial without a card, and open-source projects can apply for free access.
PR-Agent
PR-Agent calls itself the original open-source PR reviewer. Qodo donated it, and it's now community-owned; the repository says it isn't the Qodo product. Comment /review, /describe, /improve or /ask on a pull request. Run it as a GitHub Action (recommended), a CLI, Docker or webhooks, on GitHub, GitLab, Bitbucket, Azure DevOps or Gitea, with the model of your choice through LiteLLM.
Gemini Code Assist on GitHub
Gemini Code Assist is a GitHub app that summarizes pull requests and reviews them; call it with /gemini in a comment. You can set a style guide per repository, or across repositories through Google Cloud. Google lists at least 100 PR reviews per day per installation, and it doesn't review files in .github/workflows. An enterprise version (Preview) supports GitHub Enterprise Cloud and Server. Google's review docs don't state a price.
Sourcery
Sourcery works with GitHub, GitLab.com and self-managed GitLab, with links into VS Code and PyCharm; Enterprise adds GitHub Enterprise Server and self-hosting. Per developer per month: Open Source is free for public repos, Pro is $12 billed annually or $15 monthly, Team $24 annually or $30 monthly, and Enterprise is custom with annual billing.
AI code review on GitLab, Bitbucket and Azure DevOps
Most tools start on GitHub. This is what each vendor's page says about the other hosts. "Not stated" means we found no mention, not that it can't work: a local CLI such as Open Code Review reviews any Git diff.
| Tool | GitHub | GitLab | Bitbucket | Azure DevOps |
|---|---|---|---|---|
| Claude Code | Yes | /code-review --comment on MRs; GitLab CI/CD | Not stated | Not stated |
| Open Code Review | GitHub Actions | GitLab CI | Not stated | Not stated |
| Plannotator | PRs | MRs | Not stated | Not stated |
| GitHub Copilot | Yes | Not stated | Not stated | Yes |
| CodeRabbit | Yes | Yes | Yes | Yes |
| Cursor Bugbot | Yes, incl. Enterprise Server | Yes, incl. self-hosted | Yes, incl. Data Center | Azure DevOps Services (no Autofix) |
| OpenAI Codex | Yes | Beta | Not stated | Not stated |
| Graphite Agent | Yes | Not stated | Not stated | Not stated |
| Greptile | Enterprise lists it | Self-Managed (Enterprise) | Data Center (Enterprise) | Not stated |
| Qodo | Yes | Yes | Yes | Yes |
| PR-Agent | Yes | Yes | Yes | Yes |
| Gemini Code Assist | Yes | Not stated | Not stated | Not stated |
| Sourcery | Yes | GitLab.com and self-managed | Not stated | Not stated |
How to choose an AI code review tool
Start from the agent you already pay for
On a Claude plan,
/code-reviewcosts nothing extra beyond plan usage. On ChatGPT Plus or Pro, Codex reviews GitHub PRs. On Copilot Pro or a business seat, Copilot code review is already in GitHub's reviewer menu. Compare the two agents in Claude Code vs Codex.Want it free or self-hosted: pick an open-source reviewer
Open Code Review runs locally or in CI with any OpenAI- or Anthropic-compatible provider. PR-Agent covers five Git hosts and local models through Ollama. You pay only for model usage.
On GitLab, Bitbucket or Azure DevOps
CodeRabbit, Cursor Bugbot, Qodo and PR-Agent list all four hosts. Copilot adds Azure DevOps; Codex and Claude Code add GitLab merge requests.
Predictable cost per review
Greptile charges $1 to $10 per review on Pro, Cursor states an average of $1.00 to $1.50, and Copilot estimates $0.25 to $10 by effort. Seat-based plans (CodeRabbit, Graphite, Sourcery) cost the same however many PRs you open, within their usage limits.
You review, the agent fixes
If you want to read the diff yourself and send notes back to your agent, Plannotator gives you a browser view of the plan and the changes.
High-stakes changes
For a deep check before a big merge, Claude Code's ultrareview verifies each finding in the cloud (typically $5 to $25 after 3 free runs). For security, run
/security-reviewor Codex's@codex security review(research preview).
Run reviews as a standing job
Every tool above comments when something triggers it. On Crewmark, an AI employee can take review as its job. Each employee is a Claude Code or Codex session on your own computer with a role, a goal, working hours and a heartbeat. It takes issues from a shared board and keeps their status current. See AI code review on every PR for the full setup.
Hire a reviewer
Create an employee with a reviewer or software engineer role, a goal and working hours, and write your review rules in its description.
Let the heartbeat start the work
During working hours the heartbeat wakes it every 15 minutes to 24 hours. Assigning it an issue or @mentioning it also wakes it. A repeating issue, such as a daily "review open PRs", reopens on schedule.
Review with the tools on your machine
It runs
/code-reviewin Claude Code,/reviewin Codex, or an open-source reviewer like Open Code Review, using the Git and GitHub logins on its computer. Crewmark has no built-in GitHub integration.Report on the board
It posts results as comments on the issue. When it needs a person, for a judgment call or a permission, it sets the issue to Needs action and asks there.
AI code review tools FAQ
What is the best AI code review tool?
It depends on what you already use. With a Claude plan, /code-review runs on your plan usage; with ChatGPT Plus or Pro, Codex reviews GitHub PRs. Copilot code review fits teams already on Copilot. CodeRabbit, Cursor Bugbot, Qodo and PR-Agent cover GitHub, GitLab, Bitbucket and Azure DevOps. For free and open source, Open Code Review and PR-Agent.
Are there free or open-source AI code review tools?
Yes. Open Code Review (Apache-2.0), PR-Agent (MIT), Plannotator (Apache-2.0 or MIT) and Anthropic's claude-code-action (MIT) are open source; you pay only for the model. CodeRabbit reviews public repositories free, Sourcery has a free Open Source plan, Greptile's Starter includes 50 credits a month and Graphite's Hobby plan includes limited AI reviews.
Does Claude Code do code review? What is ultrareview and what does it cost?
Yes. /code-review reviews your diff or a PR in your session and counts toward plan usage. /code-review ultra (ultrareview) is a research preview that runs a fleet of reviewer agents in a cloud sandbox and verifies each finding. Pro, Max and Ultra get 3 free runs, then a review typically costs $5 to $25 in usage credits. The managed Code Review service for Team and Enterprise is billed on token usage. See Claude Code code review.
How much does GitHub Copilot code review cost?
It's included in Copilot Pro ($10/user/month), Pro+ ($39), Max ($100), Business ($19 per granted seat per month) and Enterprise ($39 per granted seat per month), not in Copilot Free. Reviews use GitHub AI Credits; GitHub estimates about $0.25 (Lite), $1.50 (Balanced) or $10 (Max) per review.
CodeRabbit or GitHub Copilot for code review?
Copilot is built into GitHub's reviewer menu and rulesets and also works in Azure DevOps; it starts at $10/user/month on Copilot Pro. CodeRabbit adds GitLab and Bitbucket, reviews public repositories free, and starts at $24/user/month billed annually ($30 monthly), charging only for developers who open PRs.
Which AI code review tools work with GitLab, Bitbucket or Azure DevOps?
CodeRabbit, Cursor Bugbot, Qodo and PR-Agent list all three. Codex (beta), Claude Code's /code-review --comment, Sourcery and Plannotator work with GitLab. Copilot code review works with Azure DevOps. Greptile's Enterprise plan lists GitLab Self-Managed and Bitbucket Data Center.
What is open-code-review?
Open Code Review is Alibaba's open-source (Apache-2.0) AI code review CLI and Agent Skill, ocr, with 53.8k GitHub stars as of November 18, 2026. It reviews Git diffs with your own model provider or inside agents such as Claude Code, Codex, Cursor and OpenCode.
Will AI code review replace human reviewers?
Not in the data so far. In Chen and Stratton's November 2026 paper, 79.1% of firms had adopted AI code review tools but still largely kept humans in the loop, and time from review submission to approval rose 56.1% after firms adopted AI agents. The vendors also say their reviewers can miss problems.
Checked against each vendor's site, docs, pricing page and GitHub repository on November 18, 2026.