Coding Agents Reshape Exploit Development Practices

On April 4, 2026, Thomas Ptacek argues that frontier large language models and coding agents will, within months, drastically change vulnerability and exploit research. He says agents can be pointed at a source tree to 'find me zero days' because models encode vast bug-class knowledge, excel at pattern matching and constraint solving, and can brute-force-test exploits, reshaping exploit economics and security practices.