- A speaker on Vercel's podcast argued that AI agents are excellent at known attacks but weak at imagining ones nobody has thought of.
- In that speaker's view, the valuable human security skill is thinking of new routes in and steering the agent to try them.
A speaker on "Potentially Their First Win," Vercel's podcast hosted by Malte Ubl and Andrew Qu, argued that agents are strong at known attacks but weak at new ones. The human security job, in this view, is imagining attacks nobody has tried. The transcript does not name the speaker.
What was said
The speaker began with a caveat: they are not the security expert. They said the core of the work has not changed. For defenders, one thing has improved. Agents now handle the hard part of working with computers, which once took deep study in many fields. The speaker said agents are essentially experts in all of them.
Then came the contrast. Agents, the speaker said, are extraordinarily good at running through attacks that are already known, and at reusing them against a target. But "they're not good at envisioning what no one had thought about before."
The speaker called the human edge a mindset that looks for other ways to reach a goal, ways nobody had considered. They said it is almost the opposite of what agents do well.
The example was a person called Paulos, though the transcript may misspell the name. He has that mindset and also uses an agent. Because he knows what to prompt, he can trigger behaviour you would not get by default. The excerpt does not say what he asked for.
Why it matters
Our reading: the speaker gave no mechanism, so here is a plain one of our own. An agent-driven security test works like a loop. The agent takes a list of attack methods people have already catalogued. It tries each one against the target and notes what happens.
A loop like that covers the whole list fast, but only the list. A person's prompt can point the agent at a route that is not on it. The agent then does the slow, tedious trying. The person supplies the idea.
If that holds, the job shifts. Running known attacks becomes machine work. Human work becomes imagining new routes and judging what the agent found. A junior who learns only by running known attacks is training for a task an agent already does.
Buyers of security services can ask a plain question: who here imagines new attacks, and who only runs tools?
The other side
The speaker is not a security specialist and said so. This is one person's opinion, backed by one named example and no data on how often agents miss novel attacks.
The excerpt leaves other questions open. It does not say how people build that mindset. It also does not say how a newcomer builds it if entry-level work disappears.
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
- Vercel: Should Agents Wait to Be @-mentioned? | PTFW Ep. 1 (2026-11-18)





