I think any repo your agents push to needs a reviewer on the push, and you should pick those repos by name. Agents push code faster than anyone reads a pull request, and side repos, client repos and internal tools collect commits no human reviews.
Two Nov 13 changes make that easy to switch on. Vercel Security Review now reviews pushes to GitHub repos with no Vercel project, and GitHub lets you turn AI Scan for repositories on or off by REST API. They work differently, so you decide which repos get which.
The Signal
As of Nov 13, Vercel Security Review no longer needs a linked Vercel project. By default it covers every unlinked repo the Vercel GitHub App can access in a connected GitHub account, and Agent settings let you narrow that to selected accounts or repos. Each push to a covered repo triggers a review. (Vercel changelog)
It's a Public Beta on Pro and Enterprise plans. Reviews start from push events rather than pull requests, so direct pushes and force pushes to your production branch get reviewed too.
Findings go to a Vulnerabilities dashboard and a GitHub check on each commit. On private or internal repos with an open PR, they also post as PR comments, code snippets included. Public repos get only the check, so an unpatched vulnerability isn't disclosed publicly.
Scope is the setting that matters. The Repositories to Review default is All repositories, which means every repo the Vercel GitHub App can access, and every push to a repo you select "starts a billed review." You can pick specific repos instead, up to 1,000. (Vercel Security Review docs)
GitHub's AI Scan is the on-demand half. Since Nov 13 you can turn AI Scan for repositories on or off with the REST API (public preview), through a repository_scan setting on PATCH /orgs/{org}/code-scanning/ai-scan and PATCH /repos/{owner}/{repo}/code-scanning/ai-scan. Repos inherit an org's setting and can opt out, but can't turn it on if the org turned it off. Turning it on doesn't start a scan. Someone with write access runs one on demand, and each scan consumes GitHub Code Security AI credits. (GitHub changelog)
Scans cover only the default branch. AI Scan is available on public org repos, plus private or internal repos with GitHub Code Security or GitHub Advanced Security, but not private repos owned by a personal account. It doesn't exhaustively analyze every file, and during the preview a merged fix doesn't trigger a re-scan. GitHub warns that "AI-generated findings can be inaccurate. Review each finding." (AI Scan docs)
The Stack
- List every repo your agents push to: client repos, internal tools, scripts. Mark the ones with no human PR review.
- Those get the per-push reviewer. Turn on "Review Changes Automatically" in Agent settings and choose specific repositories, since every push to a selected repo is a billed review.
- Older or low-traffic repos with lots of existing code get an on-demand GitHub AI Scan instead, where eligible. Re-run it after you merge fixes.
- Name one person to triage findings daily. An agent can run
vercel security vulnerabilitiesto list and summarize them; the person decides dismiss or fix. - Check who can see findings. On private repos, PR comments reach everyone with repo access.
- Treat it as one more reviewer. Vercel says it analyzes your code, not your running application, and "It isn't a security audit."
One decision: by Friday, write down which repos get the automatic reviewer, which get on-demand scans, and who triages.
Prompt of the Day
Paste this with your list of agent repos:
For each repo I've noted who pushes (agent, human, or both), whether a human reviews PRs, how often it gets pushes, whether it's public or private, and what it touches (customer data, payments, or internal only).
Using only those facts, give me: (1) a call for each repo (per-push reviewer, on-demand backlog scan, or none yet) with one reason from what I wrote; (2) repos with no triage owner; (3) a warning for any private repo where PR comments would show findings to a wide group; (4) a one-week rollout order.
Assume only these tool facts: Vercel Security Review bills a review for every push to a selected repo; GitHub AI Scan runs on demand on the default branch and uses AI credits per scan. Don't invent costs, vulnerability counts, or features. Write "unknown" for anything missing.
If no agent pushes to a repo, leave it off.
Bring your repo list to AI Freedom Lab, where builders launch companies and web and mobile apps with AI.
Tool of the Day
Bolt.new, featured in the AI Tools Index.
Bolt.new by StackBlitz turns a plain-language description into a working full-stack web app in your browser. The Free plan has a 300K daily token limit and Bolt branding; Pro is $25 a month with no daily limit. The index notes complex production apps still need manual engineering. (AI Tools Index)
A fast internal tool is exactly the kind of code nobody reviews. A practice to try: once the code lives in a GitHub repo, put that repo on your per-push list before your team relies on it.
Quick Hits
On December 14, 2026, GitHub deprecates Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3 across all Copilot experiences (announced Nov 13). Haiku 4.5 stays for individual subscribers on annual plans. Update workflows before then. (GitHub changelog)
Anthropic's Nov 12 self-hosted Claude Managed Agents lets you run the whole service on your own infrastructure and store agents, sessions and session history there. Model requests go to the provider you choose, Anthropic by default. Access is through your Anthropic account team. (Claude release notes)
Google's Gemini 4 Hydrogen (gemini-4-hydrogen) went generally available on Nov 12 for everyday coding and agentic workflows, on free and paid tiers. It adds an xhigh thinking level and is now the default model for the Antigravity agent in Gemini Managed Agents. (Gemini API changelog)
Closing the loop
Start with the repo your agents push to most and no one reviews, then work down the list. Keep building.
Join AI Freedom Lab, where builders launch companies and web and mobile apps with AI: skool.com/aifreedomlab