- Antithesis, a software-verification company, has a command-line tool designed for agents to use with its product, said Carl Sverre on Software Engineering Daily.
- The excerpts do not show what the tool does, how agents use it, or why it was built. Any link to Sverre's other remarks is our reading.
Antithesis, a software-verification company, has a command-line tool designed for AI agents to use. Carl Sverre, who leads much of its AI strategy, said so on Software Engineering Daily. A command-line tool, or CLI, is a program run by typed commands instead of clicks. The point: a verification product can be built so machines, not only people, operate it.
What was said
Sverre sits on the company's Red Team. He said each member specialises in one product area, and his is the company's relationship to AI. He wrote the Antithesis AI skills repos. He also said the company has a CLI "designed to be used by agents to interact with the product."
Antithesis sells deterministic simulation testing. In plain words, it runs a customer's software in a fully controlled environment and explores many paths through it. Sverre described the reach of that approach this way: "we have the ability to sort of breadth first search the state space of your software." That means working through the possible states a program can reach, step by step.
He also described a case where a fuzzer, a tool that feeds software odd inputs, found a bug within 15 minutes. It narrowed the bug to the exact order of operations that caused it. The excerpts do not show which system that was.
Later in the episode, Sverre said AI tends to hyper-focus on the goal it is given. He also said Antithesis is designed to find counterexamples to whatever it is given. The excerpts do not show the context of either remark, and he did not tie them to the CLI.
Why it matters
Our reading: a CLI is easy for a program to call, because it needs no screen. Read together, Sverre's remarks suggest a pairing. An agent fixed on its goal could write code, and a tool built to find counterexamples could check it. If agents write more code, a vendor may need its checking product to be drivable by them.
For buyers, that suggests a question to ask. Can the verification tool you are evaluating be run from the command line, by something other than a person? The excerpts do not show how agents actually use Antithesis.
The other side
This is one vendor describing its own product. Sverre said one tool is designed for agents. He did not say agents are the main users of Antithesis. The excerpts do not say what the skills repos contain, how many customers use them, or how well agents do with them.
Sverre also did not present agents as a cure for testing's old problems. He said it is really hard to write tests for the unknown unknowns. Better tools for agents do not remove that limit.
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
- Software Engineering Daily: Software Verification in the Age of Coding Agents (2026-10-15)
