Keynote: What Comes Next
summary
Claims that most coding is now LLM-driven and that work shifts to supervising long-running agents which run 20-hour agent workloads and continuously build, test, deploy and fix staging systems. Describes tools actually built: a reverse proxy on Bandit, socket_disc over websockets for distributed messaging, EKV key‑value with CASPaxos (benchmarked ~86k mixed ops, 7k CP writes) and a network block device filesystem in Elixir matching Go performance. Shows nightly fuzzing, end‑to‑end agent loops that trace, benchmark and hot‑upgrade running runtimes.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:06Elixir in compute and AI
- 3:12Coding to code generation with LLMs
- 6:26Agents as stateful systems
- 9:50Agent-driven software deployment
- 11:37Language and runtime in distributed systems
- 13:43Building with AI: craft and confidence
- 18:23Top-down problem solving
- 22:35Real projects with LLMs in Elixir
- 28:14Performance and fuzz testing with agents
- 31:03Erlang bottlenecks and socket_disc
- 36:39Sprites control plane demo
- 42:16Confidence in AI-generated code
- 45:45Elixir’s evolution
- 47:30Erlang/Elixir for AI era
- 50:38Future of self-healing systems
- 51:39Q&A
related talks
24mCoding for the Future Panel
23mUsing "Program shaping" and algorithmic skeletons (...)
21mModernizing Our Oldest, Scariest Code with AI
20mFully Integrated AI that Actually Ships
01h 19mKeynote: Q&A on LiveView
29mKeynote: Kent C. Dodds - The Web's Next Transition
26mKeynote
33mKEYNOTE: Making Movies With Web Applications