What is an LLM agent?

An LLM agent combines a language model with the ability to call functions. The core loop, known as ReAct (Reasoning and Acting), works like this: the LLM examines the conversation and available tools, decides whether to call a tool or respond, and if it calls a tool, the result gets fed back into the conversation. The loop continues until the agent produces a final answer or hits a step limit.

Why does it matter that most agent frameworks are Python-first?

Organizations running JVM infrastructure or Erlang/OTP systems face the choice of either moving agents to Python or building them in the runtime they already operate. Python's largest AI ecosystem (every major LLM provider ships a Python SDK) makes it convenient for some teams, but other teams can avoid the operational cost of adopting Python by building agents in their existing runtime, where Clojure and Elixir offer different production strengths.

What are the key differences in how Clojure and Elixir handle agent state?

Clojure represents agents as data transformations on immutable maps: each agent iteration produces a new state without modifying the previous one, allowing states to be diffed, serialized, stored, and replayed. Elixir models each agent as a lightweight process using the Actor Model, where each process maintains isolated state that other processes cannot directly access, preventing accidental state corruption across agents and allowing the BEAM VM to run millions of processes with preemptive scheduling.

Large Language ModelsAI Coding AssistantsOpen-Source AIHacker NewsAug 13, 2026Aug 14, 2026, 04:01 JST

Agent tooling expands beyond Python to Clojure, Elixir

Send on LINE

3 Key Points

What happened
A functional programming team built LLM agent systems in Clojure and Elixir, comparing them directly with Python implementations to show how each language handles the core agent loop (where an LLM examines available tools, decides whether to call them, and feeds results back into the conversation).
Why it matters
Most agent frameworks (LangChain, AutoGen, CrewAI, LangGraph) are Python-first, forcing organizations running JVM or Erlang/OTP infrastructure to choose between moving agents to Python or building them in their existing runtime. This work demonstrates concrete trade-offs: Python offers the largest AI ecosystem and ready frameworks; Clojure provides immutable state that can be diffed and replayed for testing; Elixir's Actor Model (lightweight processes with supervision) handles fault tolerance and distribution natively, features Python requires external tools like Kubernetes or Celery to achieve.
What to watch
The comparison shows that Elixir's process-based model maps directly to agent workflows—prompt chaining becomes message-passing between processes, routing becomes a classifier process dispatching to specialized agents, and orchestration becomes spawning and managing worker processes. For teams already operating Erlang/OTP systems, the functional approach eliminates boilerplate versus Python frameworks and simplifies concurrent agent deployment.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

The article addresses a practical gap in the agentic AI ecosystem: most tooling assumes Python, but many production organizations run on the JVM (Clojure) or Erlang/OTP (Elixir). The comparison is grounded in a real constraint—whether to adopt Python or build agents in the infrastructure already in place—rather than a theoretical exploration.

Python's advantage is well-established: the largest AI ecosystem, ready-to-use frameworks (LangChain, AutoGen, CrewAI, LangGraph), and first-party SDKs from every major LLM provider. However, Python's mutable data structures and reliance on external tools (Kubernetes, Celery, Ray) for distribution and fault tolerance come with operational costs that matter at scale.

Clojure's functional model eliminates a class of bugs entirely: because state is immutable, tool functions cannot corrupt state through hidden references, and the REPL allows developers to inspect and replay any intermediate state during development. Testing requires no mocking libraries because there are no framework internals to mock. The trade-off is a smaller AI-specific ecosystem; integrations must often be written by hand.

Elixir's inheritance of Erlang/OTP patterns (supervision trees, clustering, preemptive scheduling of lightweight processes) means that production requirements that Python must solve with external infrastructure are solved in the runtime itself. A single machine can run millions of concurrent agents by default, and distribution across machines requires no code changes. This is particularly valuable for teams already operating Erlang/OTP systems, where the agentic model maps directly to existing patterns (orchestrator processes, supervisor trees, message-passing).

FAQ

What is an LLM agent?: An LLM agent combines a language model with the ability to call functions. The core loop, known as ReAct (Reasoning and Acting), works like this: the LLM examines the conversation and available tools, decides whether to call a tool or respond, and if it calls a tool, the result gets fed back into the conversation. The loop continues until the agent produces a final answer or hits a step limit.
Why does it matter that most agent frameworks are Python-first?: Organizations running JVM infrastructure or Erlang/OTP systems face the choice of either moving agents to Python or building them in the runtime they already operate. Python's largest AI ecosystem (every major LLM provider ships a Python SDK) makes it convenient for some teams, but other teams can avoid the operational cost of adopting Python by building agents in their existing runtime, where Clojure and Elixir offer different production strengths.
What are the key differences in how Clojure and Elixir handle agent state?: Clojure represents agents as data transformations on immutable maps: each agent iteration produces a new state without modifying the previous one, allowing states to be diffed, serialized, stored, and replayed. Elixir models each agent as a lightweight process using the Actor Model, where each process maintains isolated state that other processes cannot directly access, preventing accidental state corruption across agents and allowing the BEAM VM to run millions of processes with preemptive scheduling.

Hacker NewsRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.