Open Challenges for Programming Agents
summary
Claims that current coding agents hit an analysis bottleneck — they produce lots of code but miss systematic reasoning — and shows concrete tooling that uses language models to steer abstractions in static analysis, yielding better precision than prior static analysis on functions and small modules. Describes a system that automatically generates API models and tests for JavaScript/Node packages (popular packages under 10k LOC) and reports successful, validated analyses of real codebases.
AI-written summary · about
chapters (9)
- 0:00AI programming automation intro
- 2:58Why programming automation is not solved yet
- 4:04Analysis bottleneck in code
- 11:26Language models with program analysis
- 17:53Automating library modeling
- 20:48New AI tool interactions
- 23:18Reusing code frameworks
- 25:38Future dev models
- 26:21Q&A
related talks
34mProgramming Languages for AI Programing Agents
25mBeyond the IDE: Toward Multi-Agent Orchestration
30mBeyond the Hype: Using AI Coding Agents Today
26mAl and Software: The Good, The Bad, The Scary
38mCoding Agents & Language Evolution: Navigating Uncharted Waters
41mGoing beyond copilot with AI Agents
32mProgramming language evolution in the AI era
48mAI Coding Agents and How to Code Them