"Should this be an agent or a workflow?" gets asked in every architecture review of an AI system. The answer is rarely binary. Most production systems are hybrids. But you should know which mode you are in for any given subsystem, because the operational cost is wildly different.
Here is the practical decision framework.
If you have a LangGraph agent in production and want to move to AgentFlow, this is the playbook. The graph mental model is the same, so the port is mostly mechanical. Most teams complete it in an afternoon.
We will walk a real, two-agent example end-to-end: imports, state, nodes, edges, checkpointing, API serving, and the TypeScript client.
The ReAct pattern (Reason → Act → Observe → loop) is the workhorse of modern agents. It is also where most "hello world" tutorials end and real engineering begins. What does the loop look like when the tools actually call external APIs? When the API rate-limits? When it returns malformed JSON?
Here is the end-to-end pattern, with the failure modes baked in.
The "hello world" agent works on day one. The "we have paying customers and a pager" agent works on day 200. The gap is operational: observability, retries, idempotency, graceful shutdown.
These are the patterns we see in production Python agent codebases, and the failure modes they prevent.
A blocking 30-second response is not a product. The first token in 200ms is. Server-Sent Events (SSE) is still the simplest way to stream agent output from a Python backend to a browser, and it composes well with auth, reconnection, and tool calls.
Here is the production-shaped pattern.
"Memory" is doing a lot of work in agent literature. It can mean: chat history within a session, long-term semantic recall across sessions, persistent state for resumable runs, or all three. They are different problems with different solutions.
Here is the practical taxonomy and how to wire each one in Python.
LangGraph put graph-based agent orchestration on the map. It is also a fairly opinionated piece of the LangChain ecosystem, and many teams hit limits. Operational, architectural, or licensing. That send them looking for alternatives.
This is a survey of the five strongest LangGraph alternatives in 2026, what each one is good at, and how to decide.
Most "multi-agent" demos collapse into a for loop calling two LLMs. That works for a blog post and not much else. Real multi-agent systems need explicit control flow, shared state, and a way to debug when one specialist agent goes off the rails.
Here are seven orchestration patterns we see ship in production Python codebases, and when each one is the right tool.
If you have a Python codebase and an LLM provider, you can ship a working AI agent today. The hard part is no longer "can the model use a tool?". It is "how do I keep this thing reliable in production, with memory, streaming, and a UI."
This guide walks the full path: from a single-tool prototype to a deployed Python AI agent with persistent threads and a typed frontend client. Every snippet runs.