← back
Self Coding Agents — Colin Flaherty, Augment Code
Original: Self Coding Agents — Colin Flaherty, Augment Code
Takeaway
An AI coding agent given basic process and tool primitives can write the bulk of its own codebase, including features, tests, and self-profiling optimizations.
Summary
- Augment Code's coding agent wrote over 90% of its own ~20,000-line codebase under human supervision, including third-party integrations for Slack, Linear, Jira, Notion, and Google Search.
- When adding the Linear integration, the agent used its previously-written Google Search tool to look up Linear API docs the foundation model didn't memorize, demonstrating self-bootstrapping tool use.
- The agent profiled itself, found that user-repo file loading and hashing were synchronous, added a process pool, and wrote a stress test to verify the speedup.
- Flaherty argues 2023 was autocomplete (Copilot), 2024 was chat, and 2025 belongs to agents capable of multi-step engineering work.
coding-agentsself-improvementtool-use
Original description
What if you built an AI coding agent that…built itself? It sounds like science fiction, but AI coding startup Augment Code built an AI agent that not only writes code, but continuously improves its own codebase - including performance optimizations and system architecture. This level of AI autonomy raises fascinating questions about the future of software engineering, where human engineers may shift from writing code to orchestrating multiple self-improving AI agents working in parallel. In this talk, Augment Founding Researcher Colin Faherty will explore: - What does it look like to supervise AI coding agents at scale? - How far can we push AI coding agents - what tasks are out of scope? - What might unlock when AI handles complexity and humans drive innovation? Recorded live at the Leadership Track Session Day from the AI Engineer Summit 2025 in New York. Learn more at https://ai.engineer and purchase tickets to our next event, the AI Engineer World's Fair, in SF June 3 - 5 here: https://ti.to/software-3/ai-engineer-worlds-fair-2025 About Colin Colin Flaherty is a founding researcher at Augment working on AI agents and retrieval systems. Before that, he was a researcher at Facebook AI Research, where he coauthored a paper in Science on "Cicero"—an AI that mastered the game of Diplomacy, widely considered the next grand challenge in AI for games after Chess, Go, and Poker.