How to build your own fun and absurd pair programmer
summary
A cheeky local pair-programmer called Pip the pair programmer is built as an IntelliJ plugin talking to a Spring Boot Kotlin backend, using Quarkus components, a vector database and a local 8B model (Quen 3). 23,000 exported Slack messages provide the personality vectors; retrieval-augmented generation with sparse/dense/multi vectors supplies context. Pip can rename code, generate memes, play tic‑tac‑toe and control Spotify via MCP tools but sometimes mangles files, crashes IntelliJ and refuses risky tool calls, so behavior is powerful but nondeterministic. MCP tool integrations full-line autocritique
AI-written summary · about
chapters (18)
- 0:00Introduction
- 1:42AI bubble and homogenization
- 5:31Introducing Pip
- 7:38Live demo of Pip
- 11:16Agentic systems and workflows
- 15:30Technical stack
- 19:52Renaming code and handling requests
- 22:48Processing requests
- 25:17Fast context retrieval
- 30:17Pip Tic-Tac-Toe demo
- 32:51Model Context Protocol
- 35:20MCP integration and security
- 39:59Pip controlling Spotify
- 43:22Challenges of agentic systems
- 45:50Reflections on Pip and LLMs
- 47:17LLM sustainability and future
- 49:43Experiment with LLMs locally
- 50:45Q&A
related talks
50mHow to build your own fun and absurd pair programmer
38mHow to build your own fun and absurd pair programmer
19mUsing paired programming to have fun & sell your solutions
45mSupercharge your coding: Integrating ChatGPT and Copilot in long-term projects
37mThe battle of the AI coding assistants
42mPrompt. Build. Repeat. – The Vibe Coding Mindset
55mCopilot is my Copilot: Pair programming with an AI
49mCoding Tomorrow