Real jobs people get done with AI agents, each tied to its original source, an honest outcome label, a prompt you can edit, and the GitHub repositories involved.
Google's Big Sleep agent started from recent SQLite fixes, looked for related unfixed bugs and found an exploitable stack buffer underflow before it reached an official release.
GhidraMCP exposes Ghidra's decompiler and renaming tools over MCP, so an agent such as Claude Code can list functions, read decompiled code and propose clearer names.
Sean Heelan benchmarked OpenAI o3 on a known ksmbd bug, then ran it across the SMB handlers, where it surfaced a new Linux kernel vulnerability, CVE-2025-37899.