The Vercel Agent is currently both expensive and close to useless. Over the last 4 days, it used $52 USD and came back with ~4 code reviews that are mostly inaccurate, implementing them will break my code.

For example, it made the following review:

without knowing the full context. I created the test regex without the ending $ because I want /[locale]/landing-page/* to all get captured by the test.

Comparing to other offerings like BugBot and CodeRabbit, both cost under $60 USD for “unlimited” usage per month and returns ~5 to 10 reviews per PR, Vercel Agent’s suggestions are very expensive and cannot really be trusted.

Thanks for trying it while the feature is in beta. Sharing your feedback with the team so we can investigate why it didn't work well for your team and make adjustments before the general availability release.

@wiiiimm Thanks William, this is very helpful feedback and apologies about the inconvenience. We’re going to be adding debouncing soon, which should help a bunch of reviews being triggered close to each other. We’re also looking further into the context issue you highlighted.

Let me know if you want me to test it again later… but i will need some more test credits. Burnt through all $100 USD now and I only had like under 10 reviews lol

@wiiiimm we just improved some workflows to help reduce the number of reviews which in turn reduces cost.

I’d recommend this flow:

• configure Vercel Agent to not review drafts in Agent settings
• open PRs as draft
• push N commits to the PR
• publish PR as ready
• get a review
• continue with PR workflow

Vercel Agent reviews every commit. If you are pushing many commits to a single PR, it may result in more reviews than you’d like. If a new commit does not have any issues, Vercel Agent will review but not leave any comments. The above workflow enables you to push many commits in draft state without getting reviews.

Email me casey.gowrie@vercel.com with your Vercel team ID and we can help look further.

Vercel Agent doesn’t appear to be picking up BitBucket PRs after being enabled. Is BitBucket supported at this time?

configure Vercel Agent to not review drafts in Agent settings

This is how I had Vercel Agent configured from day one. I was working a lot last week so I don’t have a problem with Vercel Agent working overtime and charges for doing reviews. My concern was the lack of feedback (I got under 5 feedbacks and none of them was on point). In comparison, other code review products returned 10 to 20 suggestions (suggestions + logic flaws etc) on the same PR. I will drop you an email with my Vercel team ID later today.

Did you notice Vercel Agent missing real bugs? Feel free to follow up via email, so we can take a deeper look.

Github is the only supported provider right now.

Hi, in the UI, the only setting for which repositories to review is “All repositories”. What does that mean? Does it mean every repo from which a project in my org is deploying to Vercel?

The thing is, I’d like to use my own account to test-drive Agent at work, but there is no clear path for me to configure which repos and in which GitHub orgs this should run. Every other AI reviewer I’ve tried presents you with a list of repositories in which PRs are reviewed. How would I go about adding Agent to my work org?

Also, I find the credits model weird. I’d find it much more intuitive to simply have Agent usage billed at the end of the month, together with all other platform usage.

"💡 Enable Vercel Agent with $100 free credit for automated AI reviews" how do I stop these ads from showing up on our PRs?

Is there a way to dismiss this for an organization?

I already dismissed this but it came back:

Bug Report: Misleading Error Message for Vercel Agent Code Review

Summary: Users with the "Developer" role receive an incorrect error message when attempting to use "Review with Vercel Agent" on GitHub pull requests.

Expected Behavior: Error message should indicate that the user's role lacks permission to use this feature (e.g., "Your role does not have permission to request code reviews").

Actual Behavior: Error displays: "Unable to request review. [team name] is in a trial period (code reviews are available after the trial ends)."

Impact: This causes confusion as teams believe they have a billing/subscription issue when the actual problem is a role permission issue. The "Member" role can successfully trigger reviews while the "Developer" role cannot.

Title: How does Vercel Agent Code Review handle multiple AGENTS.md files in a monorepo?

Body:

I'm using Vercel Agent Code Review in a Turborepo monorepo with the following structure:

1
/
2
├── AGENTS.md              # Root-level guidelines
3
├── apps/
4
│   ├── acme-app/
5
│   │   └── AGENTS.md      # App-specific guidelines
6
│   └── other-app/
7
│       └── AGENTS.md
8
├── packages/
9
│   ├── ui/
10
│   │   └── AGENTS.md      # Package-specific guidelines
11
│   └── utils/
12
│       └── AGENTS.md
13
└── turbo.json

I have a few questions about how the agent handles this setup:

1. Does the agent read all AGENTS.md files, or only the one at the repository root?

2. If it reads multiple files, how does it merge or prioritize them? (e.g., root + workspace-specific)

3. Are workspace-level AGENTS.md files scoped to PRs that touch files in that workspace?

Understanding this would help me structure my guidelines effectively --- whether I should consolidate everything at the root or take advantage of workspace-level customization.

Thanks!

Hi there! I'm getting an "Error creating pull request" when trying to use Vercel Agent's automated installation feature for both Speed Insights and Web Analytics. My project has a GitHub repository properly connected, and I've verified the GitHub permissions. The error persists even after multiple attempts to regenerate the pull request.