 # Agent-driven development: How I taught agents to do my job When GitHub Copilot first appeared, everyone said: "AI will help write code faster." Two years later — we're not just writing code faster; we're delegating entire chunks of work to AI agents. The other day, I came across an article by an engineer from Copilot Applied Science — he described how he used coding agents to build agents that automated part of his work. Sounds like recursion? Absolutely. ## What has changed Previously, a pull request meant a human wrote the code, and a human reviewed it. Now, agent-driven development means an AI agent writes the code, an AI agent reviews it, and a human looks at the architecture and decides: "Okay, this makes sense." GitHub has already released a guide for reviewing agent-generated PRs — and this is no joke. It contains specific advice: where technical debt hides, what to look at first, how not to miss a logical error in beautifully formatted code. ## What I took away for myself 1. **Verification is the bottleneck.** Agents generate quickly, but their output needs careful checking. Tools like eBPF (GitHub uses it for safe deployment) become must-haves. 2. **Token efficiency is a new discipline.** Agent workflows for every PR can silently burn through your API budget. GitHub instrumented its pipelines, found bottlenecks, and built agents to fix them — a meta level. 3. **Open Source contributions become more accessible.** GitHub released a beginner's guide to OSS contributions — and this is a signal: the ecosystem wants more participants. ## Why this matters for ASI Biont We are building a workforce of AI agents. Agent-driven development is not hype; it's the new reality of software development. The faster a team learns to work effectively with AI agents (setting tasks, reviewing code, measuring efficiency), the greater the advantage. In the near future, I'll do a breakdown: how we at ASI Biont use agents for code review and CI/CD automation. Subscribe, it will be interesting. #AgentDrivenDevelopment #AIAgents #GitHub #ASIBiont #DevOps