the answer to ai is engineering
The author argues that the surge in security incidents and AI-written code calls for engineering fixes rather than more AI, and walks through using Nix to patch the whole stack from one source of truth. The core idea is that agents make open-source dependencies less necessary, and that Nix overlays let you route around upstream bugs without waiting for maintainers.
- Agent-written code trained on existing vulnerable code is driving a "security take-off," and defenders may need more inference capacity than attackers to keep up.
- Open-source libraries existed because code was costly to write; with agents, the author uses far fewer and prompts for small, purpose-built code instead, reducing Dependabot and supply-chain churn.
- Nix can act as a single source of truth for laptops, CI, agent sandboxes and production images, and lets you monkey-patch tools like git (removing force push) across every environment at once.
Geoffrey Huntley's talk at AI DevCon New York argues that the security incidents of the past couple of weeks, including a breach of OpenCode, a Japanese Prime Minister urging everyone to update PHP, a remote-shutdown bug in an HTTP stack, a 12-year-old GitLab flaw enabling remote code execution on self-managed servers, and a certificate authority whose signing key was just "4" instead of a 116-digit number, are the start of a "security take-off" that will be the new normal. His reasoning is that agents trained on existing code are now writing software full of the same vulnerabilities, and that attackers with racks of inference hardware will autonomously scan any public code and exploit what they find. His practical conclusion is that organisations will need more inference capacity than their attackers to stay secure, and that per-employee token budgets of around $100 a month are unrealistic for that job.
On the dependency side, he says open-source libraries existed because code was expensive to write and hiring needed standardisation, and those assumptions no longer hold. He says he now uses very little open source beyond the Linux kernel, BIND, and similar low-level pieces, and instead prompts an agent to generate small, well-specified tools like a command-line argument parser for his exact use case. He admits this is not a license to vibe-code cryptography, and that a trusted core is still needed. His alternatives are to generate first-party code or vendor open-source projects into the repo so agents can modify them. He also points to the supply-chain cost of Dependabot churn and compromised upstream maintainers as hidden drains on engineering time.
For patching, he recommends Nix, which dates to 2003, as the way to make patching tractable. Previously, patching a dependency meant learning its toolchain, writing a CI job, and hosting artifacts, but an agent can now handle that if you know Nix exists. His demo repo uses devenv to define packages and services in one file that generates developer laptops, CI/CD, agent sandboxes, and production Docker images, eliminating drift between them. He also describes a monkey-patched git build that refuses force pushes, distributed through the same Nix environment so every machine and sandbox gets it. The talk's broader point is that the fastest incident response and the fastest builds come from the same investment: reproducible, centrally controlled environments.