Elixir Mentor

Mike Hostetler on Jido v3

19 snips
Oct 2, 2026
Mike Hostetler, creator of Jido and ReqLLM, returns to explore Jido v3’s declarative DSL, agentic back pressure, and durable workflows with Runic and Jido Flow. They compare deterministic actors with LLM-powered agents, discuss multiplayer OTP harnesses, personal agents, IoT reverse engineering, model tooling, vibe-coded contributions, and what AI means for engineering.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
ADVICE

Give Agents Immediate Feedback With Back Pressure

  • Build agentic back pressure into every workflow through tests, fixtures, schemas, or other project-specific validation.
  • ReqLLM uses live API fixtures across 15 providers, while Jido uses acceptance, property, and fuzz tests to catch changing behavior.
INSIGHT

Why Ten Thousand Agents Belong Outside Coding

  • Software teams will likely coordinate roughly 10 to 20 agents because humans still need to validate design and nuanced output.
  • Non-software tasks can support thousands of agents because information movement and routine chores have a much higher automation ceiling.
INSIGHT

Most Actors Do Not Need An LLM

  • Jido treats agents as either LLM-driven programs or classical actors powered by rules, state machines, and behavior trees.
  • An LLM can handle exceptional cases while deterministic logic handles routine behavior, reducing inference cost without abandoning flexibility.
Get the Snipd Podcast app to discover more snips from this episode
Get the app