What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
AI coding-agent pull requests do not have one dependable “agent” failure rate. In Obada Kraishan’s 2026 analysis of 37,623 provenance-labeled pull requests, 90-day reverts ranged from 6.1% for OpenAI Codex to 14.5% for Devin, against an 11.5% matched human baseline. The gap is a vendor-level signal, not proof that one product will perform the same way in every repository.
The direct answer: vendor matters more than the label “AI agent”
A pull request (PR) revert is a merged change that is later undone by a revert-style commit within the following 90 days. On that definition, Codex had the lowest observed rate in the comparison and Devin the highest. The absolute difference was 8.4 percentage points, and Devin’s observed rate was about 2.4 times Codex’s. Both figures came from public GitHub repositories, not from a randomized trial or a controlled coding benchmark.
As an Amazon Associate I earn from qualifying purchases.
The matched human baseline was 11.5%. Codex’s 6.1% was 5.4 percentage points below that reference, while Devin’s 14.5% was 3.0 points above it. Those differences describe what happened in the sampled repositories during the observation window; they do not establish that switching tools alone causes a repository’s revert rate to rise or fall.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →What the vendor comparison actually found
| Author | PRs in revert analysis | Reverted within 90 days | Odds ratio versus humans | 95% confidence interval | Statistical interpretation |
|---|---|---|---|---|---|
| OpenAI Codex | 17,756 | 6.1% | 0.50 | 0.44–0.57 | Significantly below the matched human baseline |
| Devin | 2,185 | 14.5% | 1.31 | 1.11–1.54 | Significantly above the matched human baseline |
| GitHub Copilot | 2,094 | 12.5% | 1.10 | 0.93–1.31 | Not distinguishable from humans in this analysis |
| Cursor | 946 | 11.4% | 1.00 | 0.79–1.25 | Not distinguishable from humans in this analysis |
| Claude Code | 267 | 10.5% | 0.90 | 0.60–1.36 | Not distinguishable from humans in this analysis |
| Matched human baseline | Reference group | 11.5% | 1.00 | Reference | Matched repositories and time window |
The sample-size column is essential. Codex’s estimate is based on 17,756 PRs, whereas Claude Code’s is based on only 267. A percentage calculated from 267 observations is inherently less precise, which is reflected in Claude Code’s wide confidence interval. The confidence intervals for Copilot, Cursor and Claude Code include 1.00, so their differences from the human reference were not statistically distinguishable in this dataset.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
How to read the odds ratios
An odds ratio compares the odds of a revert with the matched human group; it is not the same thing as a percentage-point difference. Codex’s 0.50 means the modeled odds were about half the human odds under the study’s matching approach. Devin’s 1.31 means its modeled odds were about 31% higher. The intervals communicate uncertainty around those estimates and should be considered alongside the raw rates and sample sizes.
Why can two agents have such different revert rates?
The data establish a spread between products, but they do not identify a single causal mechanism. Several factors can produce that pattern, and they can operate at the same time.
Different users and task assignments
Agents are not randomly assigned to tickets. A team may reserve one product for routine dependency updates and another for difficult debugging, migrations or broad refactors. Repositories also attract different contributors and review standards. If a vendor is used on riskier work, its PRs can face more opportunities for reversal even if the generated code is competent. Conversely, a tool used mainly for small, well-tested changes may accumulate fewer reverts.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsDifferent levels of autonomy
“Agent” covers products with very different workflows. Some primarily suggest edits while a developer drives each step; others can inspect a repository, run commands, modify many files and open a complete PR. More autonomy can increase the amount of code changed per task and the number of assumptions made about tests, configuration and deployment. The vendor label alone does not reveal how much human direction each PR received.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
Repository and process effects
Revert decisions depend on branch protection, test coverage, release cadence and reviewer expertise. A repository that merges quickly and fixes problems in follow-up commits may show a different revert pattern from one that keeps changes open for extensive review. Public projects also differ in issue quality, maintainer availability and tolerance for experimental work. These process variables can be associated with vendor choice.
Measurement effects
The analysis identifies reverts through revert-style commit messages within 90 days of merge. A silent rewrite, a follow-up fix, a squashed replacement or a rollback described without revert language can be missed. The resulting metric is therefore “detected reverts under this rule,” not every possible way a merged change was undone.
Is Codex safer than Devin in real repositories?
In this particular public-repository sample and time window, Codex had a lower detected 90-day revert rate than Devin, and both products’ confidence intervals were on opposite sides of the human reference. That is a meaningful empirical difference, but it is not a universal safety ranking.
The observation window ran from December 2024 through July 2025. Product versions, prompting practices, integrations and user populations can change after that period. The authors also caution that the results may not transfer to private repositories, other programming languages or later product versions. A company deciding between tools should treat the published rates as a starting benchmark, then measure its own PR outcomes using the same definition.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
It is also important not to convert 6.1% into “safe” or 14.5% into “unsafe.” A revert can result from a changed requirement, an intentional experiment, a release decision or an integration problem rather than a coding mistake. Conversely, a change can remain unreverted while still creating maintenance or security work. Reverts are a useful operational outcome, not a complete quality score.
What the human comparison says about agent code
The human baseline was matched to repositories and the same general time window, which makes it more informative than comparing agents with an unrelated historical average. Even so, matching does not remove every difference in task mix, contributor experience or review behavior.
The pooled security-smell result points in a different direction from the revert result. Regex detectors found at least one targeted smell in 2.9% of pooled agent PRs and 4.6% of human PRs. That is a pattern signal, not a vulnerability count. The detectors examined added Python, JavaScript and TypeScript lines and looked for regular-expression matches; they did not prove exploitability, severity or whether a finding was reachable in production.
Because the 2.9% and 4.6% figures are pooled across vendors, they cannot show whether Codex, Devin, Copilot, Cursor or Claude Code individually produced fewer smells. Nor can they establish that agents are more secure overall. Security review still needs tests, dependency analysis, permission checks, threat modeling and human judgment.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
Review capacity changes the practical risk
A PR’s outcome depends partly on whether a qualified person has time to inspect it. The analysis found that GitHub Copilot attracted the most human review activity among the compared products. Claude Code had a median time to first human review of 12.6 hours, the longest reported latency in the comparison. The headline does not provide a numeric review-depth value for each vendor, so review volume and review speed should not be treated as interchangeable measures.
For a team, a lower revert rate with delayed review may still create unacceptable exposure if changes reach production before inspection. Conversely, a higher review load can indicate either stronger scrutiny or more problematic PRs. Track both the time to first review and what reviewers actually check: tests, permissions, data handling, error paths and compatibility.
How the analysis was assembled
- Provenance labeling: PRs were attributed to OpenAI Codex, Devin, GitHub Copilot, Cursor, Claude Code or a matched human baseline.
- Repository scope: The sample covered 2,807 public GitHub repositories with more than 100 stars, using Python, JavaScript and TypeScript.
- Volume: The revert analysis included 37,623 provenance-labeled PRs.
- API linkage: The pipeline joined the AIDev dataset to 58,792 cached GitHub API responses.
- Outcome window: Each merged PR was followed for 90 days, and revert-style commit messages were used to detect reversions.
- Static checks: Regex detectors scanned added lines for selected security-smell patterns.
Obada Kraishan summarizes the central design choice this way: “Because every PR carries a vendor label, the analysis can separate ‘agents in general’ from the behavior of individual products, a distinction that turns out to matter.” That separation is why a pooled agent average would hide the difference between Codex and Devin.
What is not established by the published headline table
The comparison framework includes PR size, security-smell presence, normalized post-merge churn, human-review depth and review latency. The vendor table above provides the revert rates, odds ratios, confidence intervals and sample sizes; per-vendor numeric values for PR size, normalized churn, security-smell rate and review-depth measures are not stated in the available summary. They should not be filled in by assuming that a larger PR, a faster review or a particular vendor caused the observed revert outcome.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
How engineering teams should use these numbers
1. Establish a local baseline
Count merged PRs and detected reversions by vendor, repository, language and task type. Use a fixed follow-up period, such as 90 days, and document how you treat rollbacks, squashed merges and silent rewrites. Without a consistent definition, one team’s “revert rate” will not be comparable with another’s.
2. Compare like with like
Separate small maintenance patches from multi-file feature work, incident fixes and migrations. Record PR size, test changes, review time and deployment context. A vendor used for high-risk work should not be judged against one used for low-risk chores without accounting for that assignment.
3. Preserve human review for consequential changes
Require an owner who understands the affected service to inspect authentication, authorization, data validation, error handling, dependency changes and operational settings. A passing test suite does not verify every security or compatibility property.
Recommended Free Tools
4. Watch post-merge outcomes
Monitor reverts, hotfixes, incident links, follow-up churn and customer-facing regressions. A 90-day revert metric is useful for trend detection, but a dashboard that includes only reversions will miss problems resolved through ordinary follow-up commits.
5. Treat security-smell alerts as triage signals
Run a real security review on any flagged pattern. Confirm whether the code is reachable, whether input is trusted, whether escaping or validation is present and whether the finding is exploitable in the deployed configuration. Do not report a regex match as a confirmed vulnerability.
Bottom line for readers choosing an agent
The strongest supported conclusion is narrow: in Kraishan’s 2026 analysis of public repositories from December 2024 through July 2025, vendor-level revert rates differed substantially. Codex registered 6.1%, Devin 14.5% and the matched human baseline 11.5%; Copilot, Cursor and Claude Code were statistically indistinguishable from humans under the stated model. Those figures justify measuring products separately, reporting sample sizes and preserving review controls. They do not justify declaring any agent universally safer, because the comparison was observational, the revert detector was rule-based and the results may not generalize beyond the repositories, languages and product versions observed.
Quick Recap
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.
