 AI Agents Are Already Writing AI Agents: What GitHub Copilot Applied Science Showed Yesterday, an article was published on the GitHub Blog that everyone building AI products should be ashamed not to read. The guy from Copilot Applied Science took coding agents — and automated part of his work with them. That is, he used AI agents to write AI agents that automate his work. This is no longer a proof of concept — it's a production case from the team that makes Copilot itself. What else did the GitHub feed show today: 1️⃣ **Copilot Max plan** — starting June 1, new tariffs with flex allotments. The AI assistant market is consolidating, and Copilot is now clearly targeting enterprise with the Max plan. 2️⃣ **eBPF for safe deployment** — GitHub uses eBPF to detect cyclic dependencies in deployment tooling. A technique worth adopting for anyone managing microservices. 3️⃣ **AI accessibility** — Continuous AI for accessibility: AI automation of triage for accessibility feedback. Instead of a chaotic backlog — continuous fast fixes. 4️⃣ **GitHub Innovation Graph** — researchers predict GDP, inequality, and emissions using GitHub data. The code economy is becoming a real macroeconomic indicator. My position: the trend toward agent-driven development is not hype, but a new reality. If Copilot Applied Science is embedding coding agents within itself, then the technology has matured. The question is not whether your code will be written by an AI agent, but when you will start using it consciously. https://asibiont.com/ — we build AI agents that actually work.