Anthropic: Building effective agents
Primary guidance on simple workflows, tool use, orchestration, and adding autonomy only when it creates measured value.
← Back to Compute College
Compute College
What is an AI agent?
Understand an AI agent as a model-driven system that observes context, chooses actions, and works toward a goal.
Plain-English definition
An AI agent is a system that uses a model to choose steps toward a goal while interacting with tools, data, or an environment. The model is one component; instructions, state, permissions, tools, stopping rules, and evaluation determine what the system can actually do.
Memory trick: An agent is a model inside a controlled loop.
Why it matters
Calling every model workflow an agent hides important design choices. A controlled agent can be useful when the path is uncertain, but it also creates more calls, more state, and more ways to fail than a fixed workflow.
- The design makes a hidden tradeoff measurable before it reaches users.
- A clear boundary reduces retries, unsafe actions, or unnecessary capacity.
- The operating choice should be tested against the workload rather than assumed from a demo.
Simple example
A research assistant receives a question, searches approved sources, extracts evidence, asks for clarification when necessary, and drafts a cited answer. It cannot send messages, change records, or access sources outside its allowlist.
- The example names the input, decision, and expected boundary.
- A deterministic or human checkpoint owns the part a model should not guess.
- The result can be measured with quality, latency, safety, and cost metrics.
Example figures are illustrative calculations, not current quoted market prices.
Current example
Primary reference
This source explains the method or risk boundary. It does not replace testing the actual model, data, provider, and workload conditions in production.
Source discipline: examples on this page are illustrative; measure the real workload before making a release or capacity decision.
Common mistake
An agent is more than a smarter prompt. It is a system with tools, state, permissions, and failure handling; adding a loop changes cost and risk.
Practical takeaway
What you can do with this
Draw the proposed agent as model, instructions, tools, state, environment, and evaluator. Mark which actions are read-only, reversible, or irreversible before building.
- Start with a narrow workflow and explicit failure boundary.
- Instrument the path before optimizing it.
- Review the design whenever model, tool, traffic, or policy changes.
Decision check: can you explain what the agent may do, what stops it, and how a human can inspect the result?
Compute College learning path
AI Engineering
Step 27 of 48: What is an AI agent?