Learning To Stop Writing C++ Code (and Why You Won’t Miss It)
summary
Claims that frontier LLMs (Anthropic’s Claude among others) can read, plan and edit real C++ repos: demo shows an agent bootstrapping build configs, running tests and producing a pull request in ~3 hours for a half‑million line repo. Agents use tool calls and retrieval to work around a 200,000 token window, excel at refactoring, build and repo plumbing, tests and boilerplate reduction, but still make contextual errors (ADL/namespace confusions, reward‑hacking, occasional hallucinations) and need careful sandboxing and review.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:21Large language models overview
- 12:28AI skepticism and software future
- 22:47Context windows and agentic tools
- 31:37Agents in AI and software
- 41:09LLMs for code and best practices
- 1:05:17Writing code with LLMs
- 1:28:29Programming languages for AI
- 1:34:54Conclusion and Q&A
related talks
40mHow To Tame Your Agents?
01h 15m10x Development: LLMs For the working Programmer
19mLLM codegen fails and how to stop 'em
27mAgentic Coding with Discipline & Skill: Practical Techniques and Tips
24mAI Hates Legacy Code
30mBeyond the Hype: Using AI Coding Agents Today
16mBeyond the IDE: The Future of Developing Software With Agents
01h 01m10 tips to level up your ai-assisted coding