In this episode of the Elixir Mentor Podcast, I chat with Coby Benveniste and Daniel Garcia-Shulman from MarkeTeam.ai about building intelligent AI marketing agents with Elixir. They share their experience migrating from Python and React to a full Elixir and LiveView stack, and explain why the BEAM VM is ideal for powering autonomous agent workflows.

Coby and Daniel explain their approach to agent architecture, including why they chose gen state machine over gen server for managing agent state machines. They walk through the ReAct pattern (reasoning, actions, observations) and how it maps naturally to Erlang's state machine behaviors. The conversation covers their custom marketing strategy LLM, how they use RAG patterns for brand context, and why specialized agents outperform single all-purpose agents.

We explore the technical details of their stack, including how they handle DevOps without a dedicated team using mix release, their use of Fun with Flags for feature flagging, and how Broadway and Oban power their data pipelines. The discussion also covers practical workflows with Claude Code, context management using Beads, and the usage rules library for better LLM documentation.

The episode wraps up with insights on hiring Elixir developers, the emerging field of AEO (Answer Engine Optimization), and advice for developers learning Elixir with LLM assistance. Whether you're building AI agents, exploring marketing automation, or curious about advanced Elixir patterns, this conversation offers practical insights from engineers shipping production AI systems.

Resources Mentioned:
- MarkeTeam AI: https://www.marketeam.ai
- Beads (Claude Code context tool): https://github.com/steveyegge/beads

Connect with Coby & Daniel:
- Coby: https://www.linkedin.com/in/coby-benveniste/
- Daniel: https://www.linkedin.com/in/danielegsh/

SUPPORT ELIXIR MENTOR
- Elixir Mentor: https://elixirmentor.com

Romanian Space Time Portal to Tibet & Egypt's Hall of Records - Interview with Peter Moon