Top AI Product

If you’ve been watching the AI agent space, you’ve probably noticed that most frameworks are built on Python or TypeScript. So when [Jido 2.0](https://jido.run/blog/jido-2-0-is-here) dropped this week — an agent framework built entirely on Elixir and the BEAM — it caught a lot of people off guard. In a good way. The project racked up 317 upvotes on its [Show HN post](https://news.ycombinator.com/item?id=47263036) and landed at #2 on [bestofshowhn.com](https://bestofshowhn.com), which tells you something about how hungry developers are for alternatives to the Python-heavy status quo.

The core idea behind Jido is refreshingly simple: agents are just structs. No magic classes, no hidden state machines — an agent is a plain Elixir data structure with state, actions, and tools. Everything runs through a single `cmd/2` function. Actions go in, an updated agent and a list of directives come out. It’s pure functional, which means your agent logic is deterministic and actually testable. That alone puts it ahead of a lot of frameworks where debugging agent behavior feels like reading tea leaves.

What makes this really interesting is the BEAM runtime underneath. If you’ve ever used Erlang or Elixir, you know about OTP supervision trees — when a process crashes, it just restarts. Jido leans hard into this. Each agent can be wrapped in a `Jido.AgentServer` (a supervised GenServer), and you get signal routing, directive execution, and parent-child agent hierarchies out of the box. The framework ships with Direct and FSM strategies for action execution, and the [jido_ai](https://github.com/agentjido/jido_ai) package adds ReAct, Chain-of-Thought, and other LLM reasoning strategies on top.

One feature that stood out to me is the new Ash_jido integration. If you’re already using the Ash framework for your Elixir app, your existing CRUD operations on Ash resources can automatically become tools that an LLM can call. That’s a pretty clever way to bridge the gap between your existing business logic and an agent layer without rewriting everything.

The name “Jido” (自動) means “automatic” in Japanese, and the project has been in the works for about 18 months — it actually started as a bot platform called BotHive back in 2024 before pivoting to agents. The [GitHub repo](https://github.com/agentjido/jido) is active, there’s a growing ecosystem around it, and developers are already building coding assistants, workflow orchestrators, and production support systems on it. You can grab it from [Hex](https://hex.pm/packages/jido) right now.

Whether the BEAM turns out to be the best runtime for agent systems is still an open question, but Jido 2.0 makes a compelling argument. If you’re an Elixir shop looking to add agent capabilities, or just tired of the Python monoculture in AI tooling, it’s worth a serious look.