Thinking Elixir Podcast 292: Sage Advice for AI Agents
19 snips
Feb 17, 2026 They unveil Sagents, an Elixir library for building LLM agents with human-in-the-loop oversight and real-time debugging. They demo a Live Debugger and discuss tracing, virtual filesystems, and safe tool authorization. Conversation covers why Elixir may suit AI, recent tooling like LiveDebugger v0.6, a minimal MCP server, a Lua revamp, job market salary analysis, and impressions of Opus 4.6.
AI Snips
Chapters
Transcript
Episode notes
Observe Agents In Real Time With Live Debugging
- Deploy a LiveView debugger to inspect agents in real time including their assembled system prompts, middleware stack, and spawned subagents.
- Mark's Sagents Live Debugger shows agent presence, message history, subagent events, and the exact prompt the LLM sees for faster diagnosis.
Search For Binary To Term To Fix A Remote RCE Risk
- Immediately scan Phoenix projects exposed to the internet for binary_to_term usage to avoid insecure term deserialization.
- Mark warns that base64-decoded terms from users can enable remote code execution and recommends the Praxial.io serialization security article.
Elixir's Language Traits Favor LLM Coding Success
- José Valim and recent studies suggest Elixir scores highly as a target language for LLM coding due to immutability, readability, and stable tooling.
- Hosts link the Tencent/TideWave study and Dashbit blog arguing Elixir's REPL, docs, and low operational complexity improve agent accuracy.