Efficient programming with Scala and LLMs
summary
Shows that strong typing plus tooling reduces hallucinations and that LLMs act like supercharged autocompletion rather than engineers, so humans must verify outputs. Recommends Metals as MCP server (now runnable over stdin/stdout), prefers lightweight CLIs like Scala CLI and Cellard over bloated MCP toolsets, and promotes context compaction and skills to keep token costs down. Describes workflows for small scripts, brownfield edits, and greenfield generation, warns agents can still produce wrong code.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:08Agenda
- 2:34How smart are LLMs really
- 6:32Prompting basics
- 9:44Advanced prompting
- 12:13Context and costs
- 17:34Agentic workflows
- 22:12MCP alternatives
- 26:25Approaches to working with LLMs
- 29:58Summary
related talks
41mUnleashing the Full Potential of Metals
40mSafe Coding with LLMs - Verifiable Scala Silver Bullet
29mMCPs are Boring (or: Why we are losing the Sparkle of LLMs)
44mAccelerating Software Development: The LLM (R)evolution
01h 15m10x Development: LLMs For the working Programmer
02mSCALAR 2023
01h 16mHow Will LLMs Change Haskell, Programming and Open Source?
53mSemantics-based program manipulation in the age of LLMs