 GitHub Shows How AI Agents Are Changing Development from the Inside I read the latest GitHub Blog digest — a few things caught my attention. **Agent-driven development** — an engineer from Copilot Applied Science shared how they used coding agents to automate part of their own work. Not "write code," but "build an agent that writes code." The takeaway: working with AI agents is a separate skill that needs to be learned. The better you define the task for the agent, the more it takes on. **eBPF for deployment** — GitHub uses eBPF at the kernel level to detect cyclic dependencies in tooling before they break production. A technology from the networking and security world has moved into CI/CD. **GitHub Issues became instant** — they rewrote navigation: client-side caching + prefetching + service workers. A typical story: a feature that everyone considered "fast enough" got a significant speed boost after refactoring. **Copilot Max plan** — starting June 1, new pricing with flex allotments. Apparently, they are preparing for AI agents to be used not only for writing code but also for entire workflows. The trend is clear: AI in development is moving from "suggest a line" to "take a task and do it." And GitHub is building the infrastructure for this. #GitHub #AI #DevTools #AgenticDev