Automating menial engineering tasks with LLMs
summary
Builds an embedding-based code search called Bloop that feeds retrieved code into GPT-4, evolves from retrieve-then-generate to an agent model with tool calls (path search, diffs, branches) and actually produces diffs and branches in GitHub/Gitpod. Demonstrates the first end-to-end PR created by Bloop, fixes a front-end onboarding bug while the sole front-end dev is on holiday, and ships a patch in under 20 minutes. Notes current limits: prompt length, many files too large to fit a single context.
AI-written summary · about
chapters (7)
- 0:00Blue Project introduction
- 1:26Code search to AI tools
- 3:55Scaling AI for teams
- 7:06AI-assisted code fixes demo
- 10:18AI pull requests demo
- 12:04Fixed pipelines to AI workflows
- 13:47AI in developer tools
related talks
01h 15m10x Development: LLMs For the working Programmer
29mOne-Click Code Fix: Securing Code Using AI
19mHacking Productivity with LLMs
01h 36mLearning To Stop Writing C++ Code (and Why You Won’t Miss It)
48mGoing Beyond Prompt Engineering
45mSupercharge your coding: Integrating ChatGPT and Copilot in long-term projects
24mBeyond Code Completion: The Shift of AI Programming from Pair to Peer
30mLegacy Rails and the AI That Couldn't