Comparisons › AI code review and coding agents

AI code review and coding agents

Companies that published how they run AI code reviewers and background coding agents, and the guardrails around them.

21 companies. Claims are quoted or paraphrased from the linked sources; check the source date, because policies change.

Company	Industry / size	What the sources say
Adobe Mandate: unclear	Design / SaaS 5k-50k employees	Claude Code (Time Under Tension, 2026-05-06) Codex (Time Under Tension, 2026-05-06)
Airbnb Mandate: unclear	Travel / marketplace 5k-50k employees	Frontier LLMs (unnamed) (Airbnb Tech, 2025-03) Claude Code (Business Insider, 2026-05-08) Each file went through automated validation steps with retries, and the files the pipeline could not finish were completed by hand. (Airbnb Tech, 2025-03; InfoQ, 2025-03-28) Airbnb migrated nearly 3.5K test files from Enzyme to React Testing Library in 6 weeks. It had estimated the work at 1.5 years by hand. The first bulk run migrated 75% of files in four hours, and the pipeline reached 97% before the rest were finished manually. (Airbnb Tech, 2025-03; InfoQ, 2025-03-28) Chesky said teams are shipping more features and iterating more quickly as a result. (Motley Fool, 2026-05-07)
Anthropic Mandate: unclear	AI lab Under 5k employees	Claude Code (Anthropic, 2025-12-02) Automated Claude code reviewer (Anthropic Institute, 2026-06-04) Most respondents said they can 'fully delegate' only 0-20% of their work. Work generally involves active supervision and validation. (Anthropic, 2025-12-02) Every proposed change is read by an automated Claude reviewer before merge. A retrospective found it would have caught about a third of the bugs behind past claude.ai incidents. (Anthropic Institute, 2026-06-04) Lesse said agents already fix tests autonomously with no human interaction, but leaders are moving carefully on letting agents change the actual codebase on their own. (Business Insider, 2026-10-22) Respondents self-reported using Claude in 59% of their work, for a +50% productivity gain. Anthropic saw a 67% increase in merged PRs per engineer per day after adopting Claude Code. (Anthropic, 2025-12-02) Claude Code went from about 9.8 consecutive tool calls per transcript to 21.2, and human turns fell from 6.2 to 4.1. (Anthropic, 2025-12-02) Employees raised concerns about skill atrophy and reduced mentorship. (Anthropic, 2025-12-02) Head of platform engineering Katelyn Lesse said in October 2026 that Claude writes 'literally all' of her team's code, while people still design systems and verifications. (Business Insider, 2026-10-22) Superseded (2025 survey): staff now say Claude-written code went from worse than human code in late 2025 to roughly at parity by mid-2026. (Anthropic Institute, 2026-06-04)
Atlassian Mandate: unclear	Enterprise SaaS 5k-50k employees	Rovo Dev Code Reviewer (Atlassian) (Atlassian, 2026-04-07) AI review comments are filtered for quality before posting, and humans make the final review decision. (Atlassian, 2026-04-07) Atlassian reports a 30.8% reduction in median PR cycle time and 35.6% fewer human review comments across 1,900+ repos. (Atlassian, 2026-04-07) AI review comments were resolved 38.7% of the time, versus 44.45% for human comments. (Atlassian, 2026-04-07)
Box Mandate: unclear	Enterprise SaaS Under 5k employees	Cursor (Cursor customer story, 2026-02-13) Box AI (Aaron Levie on LinkedIn, 2025-04-30) AI agents for coding (unnamed) (Aaron Levie on LinkedIn, 2025-04-30) Strong data governance and security principles, with human-in-the-loop still required for most areas. (Aaron Levie on LinkedIn, 2025-04-30) Common platforms across the company (e.g., Box AI for corporate knowledge) plus core platforms per function (e.g., AI agents for coding). (Aaron Levie on LinkedIn, 2025-04-30) Teams encode agent guardrails and standards as Cursor rules in code (e.g., a frontend 'AI toolkit') rather than in documentation. (Cursor customer story, 2026-02-13) Box says it ideated over 100 AI agents across every function, then focused on a few 'big bets' to roll out at scale across 2,800 employees. (Box blog, 2025-09-04) Per Cursor's case study, 800+ developers (85% of engineering) use Cursor daily, Levie estimates a 30-50% roadmap throughput increase, and React and design-system migrations finished 80-90% faster than expected. (Cursor customer story, 2026-02-13)
Canva Mandate: unclear	Design / SaaS 5k-50k employees	GitHub Copilot (Canva Engineering, 2025-06-11) Cursor (Canva Engineering, 2025-06-11) Claude (Canva Engineering, 2025-06-11) Claude for Work, ChatGPT Enterprise, Gemini, Glean (Canva Newsroom, 2026-05-19) Forge (internal parallel-agent manager) (Canva Newsroom, 2026-05-19) CTO Brendan Humphreys said engineers apply a 'human touch' to meet the quality bar, and that domain mastery is needed to verify output and keep complexity out of Canva's ~70 million lines of code. (Business Insider, 2026-02-15) Humphreys said agents deliver 'hours and hours' of work done autonomously. (Business Insider, 2026-02-15)
Cisco Mandate: unclear	Networking / security 50k+ employees	Codex (Cisco Newsroom, 2026-03-01) Code reviews are still done by humans, and Cisco requires 'provenance of a human' on AI-written code. (SDxCentral, 2026-05-06; Cisco Newsroom, 2026-03-01)
Cloudflare Mandate: unclear	Cloud / networking 5k-50k employees	OpenCode (Cloudflare blog, 2026-04-20; Cloudflare blog, 2026-04-20) Windsurf (Cloudflare blog, 2026-04-20) AI Gateway (Cloudflare's own) (Cloudflare blog, 2026-04-20) AGENTS.md across about 3,900 repos (Cloudflare blog, 2026-04-20) An AI reviewer (a coordinator agent over specialist agents) runs on every merge request and can block merges for serious problems; a human can comment 'break glass' to force approval, which happened 288 times (0.6% of merge requests). (Cloudflare blog, 2026-04-20) Cloudflare says the AI reviewer 'isn't a replacement for human code review, at least not yet'. (Cloudflare blog, 2026-04-20) The average AI review costs $1.19 (median $0.98, P99 $4.45) and completes in a median of about 3 minutes 39 seconds. (Cloudflare blog, 2026-04-20)
Coinbase Mandate: yes	Crypto / fintech Under 5k employees	GitHub Copilot (TechCrunch, 2025-08-22; Fortune, 2025-08-25) Cursor (TechCrunch, 2025-08-22; Fortune, 2025-08-25) Armstrong: 'You don't want people vibe coding these systems moving money'; AI code must be code reviewed with appropriate checks and humans in the loop. (CIO, 2025-09-03) Coinbase's Rob Witoff said core cryptography is still meticulously researched and verified by human cryptographers, core systems sit in the middle, and internal prototypes are effectively 100% automated. (Cointelegraph, 2026-07-14)
Datadog Mandate: unclear	Observability / SaaS 5k-50k employees	Claude Code (Claude, 2026-07-21) Codex (OpenAI, 2026-01-09) Bits AI / BitsEvolve (Datadog blog, 2026-03-09) Datadog replayed past incidents and found Codex review feedback would have helped in about 22% of them. (OpenAI, 2026-01-09) Datadog follows 'harness-first engineering': instead of reading every line of agent-generated code, it invests in automated checks and production telemetry that verify the code is correct. (Datadog blog, 2026-03-09) An internal tool (Helix) was built in days with Claude Code. (Claude, 2026-07-21) BitsEvolve, an LLM-guided optimizer, shipped 10x speedups on key ingestion functions, verified against live traffic. (Datadog blog, 2026-03-09)
Dropbox Mandate: unclear	SaaS / cloud storage Under 5k employees	GitHub Copilot (Dropbox.tech, 2025-08-19) Internal AI tools (code review, migrations, incident debugging) (Dropbox.tech, 2025-08-19) Nova (internal coding agent platform) (Dropbox.tech, 2026-05-21; Dropbox.tech, 2026-05-28) CTO Ali Dasdan cautions that the volume of AI-generated code does not equal business value. (Dropbox.tech, 2025-08-19) Nova runs agents in isolated environments with validation commands, one branch per session and publication kept outside the agent. A human makes the final call before code reaches production. (Dropbox.tech, 2026-05-21; Dropbox.tech, 2026-05-28) More than 90% of Dropbox engineering uses AI tools. Copilot adoption sat in the low double digits from late 2023 through 2024, until leadership alignment and a change-management plan. (Dropbox.tech, 2025-08-19) Nova accounts for roughly 1 in 12 pull requests at Dropbox (May 2026). It handles flaky-test fixes validated with 100+ CI runs, migrations and dependency upgrades. (Dropbox.tech, 2026-05-21; Dropbox.tech, 2026-05-28)
GitHub Mandate: unclear	Developer tools Under 5k employees	GitHub Copilot coding agent, app and CLI (GitHub's own products) (Latent Space, 2026-06-02; GitHub blog, 2025-11-12) Copilot cloud agents and skills (Latent Space, 2026-06-02) Copilot's coding agent is assigned issues in the core github.com repository; GitHub says its PRs merge at a lower rate than human PRs and that it targets the 'tedious 80%' of work. (GitHub blog, 2025-11-12) GitHub uptime fell below 85% in April 2026, and a merge-queue bug reverted code in 658 repositories, as agent traffic surged. (Times of India, 2026-05-05)
Google (Alphabet) Mandate: unclear	Big tech 50k+ employees	Antigravity (Google, 2026-04-22) Gemini models (internal agentic workflows) (Google, 2026-04-22) AI-generated code is reviewed and accepted by engineers before it lands. (Google, 2024-10-29; Google, 2026-04-22) Pichai said a complex code migration done by agents and engineers together finished six times faster than a comparable one a year earlier, and that engineers are moving to agentic workflows. (Google, 2026-04-22)
LinkedIn Mandate: unclear	Social / professional network 5k-50k employees	Contextual agent playbooks and tools (CAPT, MCP-based) (LinkedIn Engineering, 2026-01-27) Multi-agent AI code reviewer (LinkedIn Engineering, 2026-08-13) The AI reviewer gives a first pass before human review; most AI reviews land before the first human reviewer. (LinkedIn Engineering, 2026-01-27; LinkedIn Engineering, 2026-08-13) LinkedIn rejected off-the-shelf single-model reviewers in favour of a multi-agent approach tuned to each codebase. (LinkedIn Engineering, 2026-08-13) Triage time fell about 70% and data analysis became about 3x faster with CAPT. (LinkedIn Engineering, 2026-01-27)
Netflix Mandate: unclear	Consumer media 5k-50k employees	Claude Code (Claude (Code with Claude SF), 2026-05-06) JetBrains AI coding agents (JetBrains, 2026-10-01) Internal agent platform (Netflix careers, 2026-10-01) A pre-review agent flags drift from the team's architecture decisions before human reviewers open the diff, and a separate agent updates documentation after merge. (QCon San Francisco 2026, 2026-10-01) Netflix engineers became 20-30% more efficient with AI coding agents, and up to 65% on some tasks, according to JetBrains. (JetBrains, 2026-10-01)
Nvidia Mandate: yes	Semiconductors / AI 5k-50k employees	Cursor (Business Insider, 2025-10-15; Cursor customer story, 2026-02-06) OpenAI Codex (GPT-5.5) (Economic Times, 2026-04-25) Teams use custom Cursor rules to automate workflows such as git flow, CI debugging and bug fixes, with Cursor running tests for validation. (Cursor customer story, 2026-02-06) Per Cursor, 30,000+ Nvidia developers use Cursor daily and commit three times more code, with bug rates flat. (Cursor customer story, 2026-02-06) In April 2026 Huang emailed staff that OpenAI's Codex was available to every employee after early access for about 10,000. (Economic Times, 2026-04-25)
OpenAI Mandate: unclear	AI lab 5k-50k employees	Codex (Outlook Business, 2026-05-15; OpenAI, 2026-06-25; The Pragmatic Engineer, 2026-02-17) Bespoke AI code-review model (The Pragmatic Engineer, 2026-02-17) AGENTS.md and skills (The Pragmatic Engineer, 2026-02-17) A bespoke AI code-review model runs on every PR; about 9 in 10 of its comments are judged valid. Non-critical code can merge after AI review alone, while core agent and open-source code still needs a human reviewer. (The Pragmatic Engineer, 2026-02-17) The heaviest (99th-percentile) users run more than 60 hours of agent turns per day across parallel agents. (OpenAI, 2026-06-25)
Reddit Mandate: unclear	Social media Under 5k employees	Huffman: the bottleneck 'is actually code review' now that AI writes more code. (Livemint, 2026-03-22)
Shopify Mandate: yes	E-commerce 5k-50k employees	Cursor (Tobi Lütke memo, 'Reflexive AI usage is , 2025-04-07; Shopify, 2025-10-28; BVP Atlas, 2026-04-02) GitHub Copilot (Tobi Lütke memo, 'Reflexive AI usage is , 2025-04-07) Claude Code (Tobi Lütke memo, 'Reflexive AI usage is , 2025-04-07; BVP Atlas, 2026-04-02) chat.shopify.io (internal) (Tobi Lütke memo, 'Reflexive AI usage is , 2025-04-07) OpenAI Codex (BVP Atlas, 2026-04-02) River (internal Slack agent) (Shopify Engineering, 2026-05-28; Shopify Engineering, 2026-10-22) Aquifer (internal agent platform) (Shopify Engineering, 2026-05-28) All AI requests route through an internal LLM proxy. A senior human reviewer is required on PRs, and AI cannot check in code automatically. (BVP Atlas, 2026-04-02) Engineers are expected to understand two to three layers below the code they ship, to avoid 'comprehension debt'. (BVP Atlas, 2026-04-02) River only works in public Slack channels (no DMs), so every session is a searchable transcript. Agent harnesses run outside the code sandbox. (Shopify Engineering, 2026-05-28) Shopify reports that its internal 'Quick' prototype-hosting sandbox has passed 5,000 sites since launching in mid-2025. (Shopify, 2025-10-28) Every team has unlimited access to top AI models. (Shopify, 2025-10-28) Engineering leadership estimates about a 20% productivity gain from AI tools. (BVP Atlas, 2026-04-02) Shopify gave agents per-session remote dev environments, so agents run in parallel and overnight workflows fix bugs and open PRs. (Shopify Engineering, 2026-10-22)
SpaceX (incl. SpaceXSI, formerly xAI) Mandate: unclear	AI lab / aerospace 5k-50k employees	Grok / in-house coding tools (Times of India, 2026-01-11; Business Insider, 2026-06-29) Cursor (acquired) (Business Insider, 2026-06-29) Claude (via Cursor, cut off January 2026) (Times of India, 2026-01-11)
Uber Mandate: unclear	Mobility / marketplace 5k-50k employees	uReview (internal) (Uber Engineering, 2025-08-12) Claude 4 Sonnet (comment generator) (Uber Engineering, 2025-08-12) OpenAI o4-mini-high (grader) (Uber Engineering, 2025-08-12) Internal coding agent (Business Insider, 2026-03-18) Claude Code (Business Insider, 2026-05-06) Comments go through multi-stage filtering (confidence grading, deduplication and category suppression) before posting. The tool was rolled out gradually with precision and recall dashboards. (Uber Engineering, 2025-08-12) Agent-written changes have 'zero human authoring', but engineers review and approve them before they merge. (Business Insider, 2026-03-18; Business Insider, 2026-05-06) uReview analyzes over 90% of Uber's ~65,000 weekly diffs. Engineers mark 75% of its comments useful, and 65%+ are addressed, versus 51% for human comments. (Uber Engineering, 2025-08-12) Uber estimates uReview saves about 1,500 hours each week, nearly 39 developer-years a year. (Uber Engineering, 2025-08-12)