“Coding Agents: A Crash Course”
Since the launch of GitHub Copilot, coding assistants have come a long way. The current trend is toward “agentic” systems like Aider and Copilot, and they seem to be quite successful. Aider claims that 85% of its own code is self-written, while Cline processed 34 billion tokens in a single day (just from OpenRouter data).
However, using these tools blindly can do more harm than good. That’s why I believe everyone can benefit from this talk. We’ll break down coding agent benchmarks, explore their inner workings, compare the best LLMs and tools, and dive into the latest research trends. Of course, we’ll also look at real-world examples.
Bio
Sebastian, former chairman of the scientific circle Group of Horribly Optimistic Statisticians, is currently a Machine Learning Engineer at deepsense.ai. He specializes in risk estimation, and his interest in LLMs began with the realization that the risk of working with them is… high. He has developed, among other things, an LLM-based trading assistant and the Text2SQL library db-ally. His interests include agentic systems and the latest research on large language models (LLMs).