Skip to content
Security & Trust Talking point

A Vercel podcast speaker says security fixes are small enough for an agent factory

A speaker on Vercel's podcast said fixes are often tiny, so the chores lie in merging hundreds of them.

W
WebPulse Newsroom
AI-assisted · 2 min read
Share on X LinkedIn
A Vercel podcast speaker says security fixes are small enough for an agent factory
In brief
  • A speaker on Vercel's podcast said most security fixes are maybe ten lines, so automating hundreds of them makes sense.
  • Our reading: in this setup the effort may sit in merging and testing, with review our own addition. Whether attackers can scale the same way is an open question the show did not answer.

A speaker on Vercel's podcast argued that most security fixes are small and repetitive, so a software factory suits them. A software factory is a setup where AI agents take many similar tasks and work through them in bulk. The show is Potentially Their First Win, hosted by Malte Ubl and Andrew Qu. The point came in episode 1, published on 18 November 2026.

What was said

The speaker said everyone seems to be getting hacked. They recalled a summer blog post titled "Everything hackable will get hacked". Vercel is going back through findings from deepsec, a tool it built that runs AI models over a whole codebase. The speaker said a run could return 100 or 1,000 findings, though not all would be critical.

Fixing them, the speaker said, needs little creativity: "most of the security fixes, they're maybe ten line fixes". Some tests get added, but the change stays simple. The hard part is volume. With 200 pull requests, someone must get them merged, work out whether a test failure is real or flaky, and rebase the code.

Their own setup uses a strong model to direct the work and a cheaper, faster model to do the fixing. The speaker said cost can climb at scale. They advised anyone building their own factory to fit it to their company culture, so people are not flooded with pings.

Why it matters

Our reading: if patches are small and alike, many can be written in parallel. In this setup, writing the patch may not be the slow step. The chores the speaker listed are merging, sorting real test failures from flaky ones, and rebasing. Those are what the speaker wanted automated. Review is our own addition, not something the speaker named.

Leaders who buy or run such a pipeline could ask three things. Who approves an automated patch before it ships? How does the team tell a flaky test from a real failure? What does the bill look like at 200 pull requests?

The fix-side point does not touch attackers. But the speaker's warning that everything hackable will get hacked raises the question of whether attackers can scale the same way. That question is our argument, not something the speaker showed.

The other side

The excerpts show no attackers using a factory like this. The ten-line figure carried a "maybe" and described the speaker's experience with deepsec findings, not security work in general. The speaker also said not every finding is critical, so a long list does not mean a long list of dangers.

The speaker described this as work needing little creativity. Harder fixes are not covered. The excerpts also do not say how a team checks that an automated patch is correct.

Written by the WebPulse Newsroom with AI assistance, and checked by our editorial review: every quotation was verified against the recording's transcript. How we use AI.

The conversation this talking point comes from