Agentic Workflow
A structured process in which one or more AI agents independently decompose a complex objective, execute steps using tools, maintain state across turns, and adapt their strategy based on intermediate results—without requiring a human to drive each step.
An **agentic workflow** is what happens when you replace a human operator managing a sequence of steps with an AI agent that runs the sequence itself. The agent receives a high-level goal, generates a plan, executes each step using available [[tool-use]], evaluates the result, updates its plan if needed, and continues until it reaches a termination condition or asks a human for guidance. The defining characteristic is that the agent maintains persistent state across steps—it remembers what it did in step 3 when deciding what to do in step 7.
Well-designed agentic workflows compose atomic, reliable actions rather than attempting one giant model call. Each tool invocation should be [[idempotency|idempotent]] where possible, so the agent can retry on failure without corrupting state. The [[agent-loop]] governs the individual agent's execution cycle; the agentic workflow may span multiple agents, external APIs, and approval gates. Large workflows often include structured checkpoints where a [[human-in-the-loop]] reviews progress before the agent continues with irreversible actions.
The most common misconception is that an agentic workflow requires a single monolithic agent. In practice, most production workflows decompose into a [[planning-agent]] that creates the task graph and specialist agents that execute individual steps—a pattern enabled by [[multi-agent-system]] design. Another misconception is that reliability comes automatically from better models; it actually comes from the workflow design: scoped tool permissions, [[audit-trail]] logging, retry logic, rollback paths, and [[agent-budget]] guards. A workflow that can fail safely is more valuable than one that occasionally achieves brilliance. See [[claude-tool-use-from-zero]] for a hands-on walkthrough of building a reliable agentic workflow from first principles.
## Related Terms
- [[glossary/multi-agent-system|Multi-Agent System]] — the broader architecture in which multiple specialized agents collaborate - [[glossary/planning-agent|Planning Agent]] — the agent responsible for decomposing goals into executable sub-tasks - [[glossary/autonomous-agent|Autonomous Agent]] — an agent that completes tasks independently without step-by-step human approval - [[glossary/agent-loop|Agent Loop]] — the iterative perceive-act-observe cycle the harness executes - [[courses/claude-tool-use-from-zero|Course: Claude Tool Use from Zero]] — hands-on practice with the concepts covered in this entry