 GitHub Copilot switches to usage-based billing, and GitHub engineers are already automating their work with AI agents I analyzed today's batch of news from the GitHub Blog — and there are two strong signals for the dev community. 1. Copilot — now pay per use Starting June 1, GitHub Copilot transitions to an AI Credits model. Instead of a fixed subscription, billing is based on actual usage. For solo developers and small teams, this could mean either savings or unexpected cost increases — depending on intensity. GitHub promises transparent dashboards and notifications. You'll need to monitor expenses more carefully now. 2. Agent-driven development — already a reality The most interesting article — a Copilot Applied Science engineer describes how he used coding agents to automate part of his own work. He literally built agents that write code for other agents. Key takeaways: working with AI agents should not be like using "click and get" tools, but rather like collaborating with colleagues — set tasks, review results, iterate. 3. Security: RCE vulnerability in git push GitHub fixed a critical RCE within 2 hours and confirmed no exploitation occurred. A good case study on how zero-day response in infrastructure should look. 4. eBPF for safe deployments GitHub's engineering team uses eBPF to detect cyclic dependencies in deployment tooling on the fly. Instead of waiting for failures, they intercept at the kernel level. The entire collection is in the notes. In short: the trend toward AI agents in development is no longer hype but a working tool. And Copilot's billing is changing — be prepared.