Dyalog and AI
summary
Shows concrete tooling work: a CLI agent loop that executes commands, runs APL from the command line, queries AppleCart and the defense workspace (PMAT) and runs TAP unit tests; a vector index over Dyalog docs used for semantic search and a hosted retrieval-augmented notebook. Models like Claude Opus and recent GPT-5 codec explain APL well but are ~18–24 months behind Python for code generation; after about a day of setup the agent solves an Advent of Code instance but overall productivity is currently a net negative. Mentions Model Context Protocol (MCP) and OpenAI-compatible client libraries.
AI-written summary · about
chapters (14)
- 0:00About Stefan and this talk
- 1:18AI for developer productivity – a reality check
- 3:50Pace of AI improvement
- 6:52Claude Opus explains APL
- 8:48The state of AI tooling
- 12:42What should Dyalog's AI tooling look like?
- 15:50Console agents
- 19:07Claude Opus tries writing APL
- 27:18Models are not yet good at APL
- 28:32Semantic search and RAG
- 31:06Searching Dyalog's documentation with AI
- 33:35Building AI features for your users
- 37:27The state of AI for APL
- 40:45Questions
related talks
01h 01mAI Agents In-Depth – Function Calling, MCP and Tool Use Under the Hood
01h 04mBuilding AI Applications the Pydantic Way
54mKeynote: AI-Powered App Development
32mProgramming language evolution in the AI era
24mThe truth about AI development
41mGoing beyond copilot with AI Agents
01h 28mAI Coding Agents and how to code them
31mDesigning AI Agents with the Model Context Protocol: From Answers to Actions