← All field notes
01 / Architecture

Workflow or agent? Choose the simplest system that can own the job.

Autonomy is an architectural choice, not a product adjective. Start with the uncertainty in the work, then add only the freedom the system needs.

Direct answer

The fastest way to make an AI product fragile is to give it more autonomy than the task requires. The second fastest is to force an open-ended task through a rigid chain that cannot adapt.

Architecture decision treeWho should control the path?
Can the steps be defined before execution?Start with the shape of uncertainty
Mostly yes ↓
Use a workflowsoftware owns the route
  • known sequence
  • predictable checks
  • bounded exceptions
Mostly no ↓
Use an agentmodel adapts the route
  • runtime discovery
  • tool feedback
  • safe stopping conditions
Start with deterministic structure. Add autonomy only where the next correct action depends on evidence discovered during the run.

Anthropic draws a useful line: workflows use predefined code paths to orchestrate models and tools, while agents dynamically decide how to use tools and complete the work. Both are agentic systems. They solve different shapes of uncertainty.

The difference is who controls the path

In a workflow, the product team owns the route. A request may be classified, sent through a sequence of prompts, checked, and either accepted or returned for revision. The model supplies judgment inside a structure the software defines.

In an agent, the model owns more of the route. It can inspect the environment, choose a tool, evaluate the result, revise its plan, and continue until it reaches a stopping condition or needs human input.

Use a workflow whenThe steps are known, consistency matters, and exceptions can be enumerated.
Use an agent whenThe next step depends on what the system discovers while doing the work.

Where workflows win

Workflows are underrated because their control is visible. That visibility is a strength. They are easier to inspect, test, price, and explain. Prompt chaining can make each model call narrower. Routing can send distinct request types to specialized logic. Parallel steps can reduce latency or collect independent judgments.

A workflow is usually the better starting point when:

  • the task has a stable sequence and a clear definition of done;
  • you need predictable latency or cost;
  • an intermediate artifact must pass a deterministic check;
  • the system operates in a high-consequence environment; or
  • the team does not yet have evaluation data for autonomous behavior.

For example, turning a meeting transcript into a product brief may be a chain: extract decisions, identify unresolved questions, draft requirements, then validate the document against a schema. There is no prize for letting an agent invent a new route every time.

Where agents earn autonomy

Agents become valuable when the route cannot be specified in advance. A coding task may touch one file or twenty. A research task may reveal that the original question is based on a false premise. A production incident may require inspecting logs, code, deployment history, and infrastructure in an order determined by each finding.

The agent loop works because the system receives ground truth from its tools and environment. It acts, reads the result, and adjusts. That makes tool quality, permissions, stopping conditions, and observability part of the product—not implementation detail.

Operating principleIncrease autonomy only when the value of adapting at runtime is greater than the added cost of evaluation, latency, and failure recovery.

Use an autonomy ladder

Do not jump from one prompt to an unconstrained agent. Move through a ladder and stop when the system can reliably own the outcome:

  1. Single call: one bounded transformation with a clear input and output.
  2. Workflow: several calls connected by code, checks, and routing.
  3. Tool-using loop: the model chooses among a small, well-defined toolset.
  4. Orchestrated agent: the system plans, delegates, and revises across a longer horizon.
Autonomy ladderStop when the system can reliably own the outcome
Single callone bounded transformation
Workflowcode owns steps + checks
Tool loopmodel chooses bounded tools
Orchestrated agentplans + delegates + revises
Capability increases upward—and so do evaluation cost, latency, and the need for recovery controls.

At each step, compare the new architecture against the simpler baseline. Measure task completion, quality, cost, time, recoverability, and the amount of human correction. If complexity does not create a measurable gain, remove it.

Five questions before you choose

Architecture fit matrixUncertainty × consequence
higher consequence ↑
Guarded workflowknown path · high consequence
Agent + human gatesunknown path · high consequence
Simple automationknown path · low consequence
Bounded agentunknown path · low consequence
runtime uncertainty →
High uncertainty calls for adaptation. High consequence calls for stronger evidence and control.
  1. Can we write the steps before the work begins?
  2. Does the environment provide reliable feedback after each action?
  3. Can we define a safe stopping condition?
  4. What happens when the model makes three plausible but wrong decisions in sequence?
  5. Do we have evaluations that measure the whole outcome, not just the final prose?

The right architecture is rarely the most autonomous one. It is the smallest system that can encounter the real world, make the necessary decisions, and still produce evidence that it did the job correctly.

Sources and further reading

  1. Anthropic — Building effective agents
  2. Anthropic — Effective context engineering for AI agents
Next field noteWhen a multi-agent system earns its complexity.
Continue to note 02 →