CrewAI is described as 'role driven' and LangGraph as 'state driven.' Unpack what that means in terms of the core abstraction each one asks you to think in, and explain how that difference shows up in the kind of bugs each framework makes hard or easy.
CrewAI gives you Agent + Task + Process; LangGraph gives you State + Node + Edge. The first invites role-bleed bugs, the second invites state-merge bugs.
Picture two ways to organise a kitchen. CrewAI is the brigade kitchen: each cook has a title (saucier, pastry chef, expediter), a backstory of training, and a list of dishes they own. You write the menu by saying who cooks what in what order. LangGraph is a dish-assembly line on a conveyor belt: a tray (the state) carries the dish from station to station, each station reads the tray, adds something, and passes it on, and a switch decides which station gets it next. Brigade kitchens go wrong when two cooks step on each other's roles. Conveyor belts go wrong when two stations write to the same spot on the tray and clobber each other.
Concept explanation~2 min read
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Concept explanation~2 min read
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
The CrewAI versus LangGraph question is one of the most frequently asked in 2026 LLM-engineering interviews, and it is also one of the most frequently botched. Candidates treat the two as competitors offering different APIs for the same problem. They are not. They are different shapes for different problems, and the right answer starts with naming the shape of the work before naming the framework.
CrewAI is a persona-orchestration framework: the primitives are people-shaped (role, goal, backstory) and the work is described as a team meeting (who does what, in what order, who delegates to whom). LangGraph is a typed-state finite-state machine on top of an LLM as node: the primitives are graph-shaped (State, Node, Edge) and the work is described as state evolving through a directed graph with conditional branches.
This deep dive walks both abstraction models in detail, the failure modes each model invites, the production-grade infrastructure each ships with (or does not), the decision rule for picking between them on a fresh project, and the migration pattern when one outgrows the other.
CrewAI's persona model: agents, tasks, and the manager LLM
A CrewAI app starts with Agent instances. Each agent has four required fields: role (a short title), goal (a one-line objective), backstory (paragraphs of context that anchor the persona), and tools (the list of callable tools the agent can invoke). At runtime, these four fields concatenate into the system prompt of every LLM call the agent makes.
Work is expressed as Task instances. Each task has a description (the prompt that becomes the user message), an expected_output (a brief specification of what 'done' looks like), and an agent field that pins the task to a specific agent. Tasks can also declare context (the list of upstream tasks whose outputs are injected) and tools (overriding the agent's default toolset for this task).
The Crew object ties it all together with a Process. Sequential runs the tasks in declared order, threading outputs through context. Hierarchical inserts a manager LLM (configured via manager_llm) that receives the goals and decides which agent to delegate each step to. The manager itself is an LLM call per delegation decision, which is the source of much of CrewAI's latency budget.
The elegance is real for genuinely persona-shaped work. A research crew with a researcher (gathers facts), analyst (synthesises themes), and writer (drafts the report) maps perfectly onto three Agents and three Tasks. The mental model is transparent to non-engineers, which is part of why CrewAI took off in 2024.
Situations where this technique stops working.
2–4 min · Everything important, quickly.
Real products, models, and research that use this idea.
- CrewAI's reference research-crew template (researcher + analyst + writer) is the canonical persona-orchestration example; ships in the CrewAI docs.
- LangGraph's create_react_agent and supervisor patterns power LangChain's own agent reference implementations in 2026.
What an interviewer would ask next. Try answering before peeking at the approach.
QWhen would you actually pick CrewAI over LangGraph for production?
When the work is conceptually a team of specialists and the workflow is mostly linear (Sequential process) with no need to checkpoint or interrupt. Content pipelines (research, draft, edit, publish) are a sweet spot. The persona as prompt mental model also helps non-engineers contribute to agent design.
Red flags & common mistakes
The phrases that signal junior thinking. Click to expand.
Red flags & common mistakes
The phrases that signal junior thinking. Click to expand.
Picking CrewAI for a pipeline that is really a state machine, then fighting role descriptions to encode branching logic. The shape of the work should pick the framework, not vice versa.
60 second bullets to scan on the way to the call.
CrewAI's Agent, Task, Process triple
LangGraph's State, Node, conditional-edge triple
Primary sources. Browse if you want the original framing.
Same topic, related formats. Practice these next.