Skip to main content
Design agents that are effective, reliable, and easy to maintain.

Core Principles

Start Simple

Begin with a narrow scope and expand as the agent proves itself.

Clear Mission

Define a focused purpose that the agent can understand and execute.

Right Tools

Give agents only the capabilities they need — no more, no less.

Iterate Often

Improve through conversation and testing, not upfront perfection.

The Intern Analogy

Think of setting up an agent like onboarding a new intern:
1

Explain the job

What should they do? What’s the goal?
2

Provide context

What do they need to know? What resources should they access?
3

Set boundaries

What should they never do? When should they ask for help?
4

Start supervised

Monitor their work until they’ve proven themselves.
5

Gradually expand

Give more responsibility as trust is earned.

Agent Scope

ScopeProsCons
NarrowPredictable, safe, easy to testLimited flexibility
BroadVersatile, handles edge casesNeeds more oversight, harder to test
Start narrow and expand. A well-functioning narrow agent is better than a broad agent that makes mistakes.

Instructions Best Practices

Do:
  • Be specific about the expected outcome
  • Explain the “why” behind important rules
  • Provide examples of good and bad behavior
  • Define escalation paths for edge cases
Don’t:
  • Write novel-length instructions upfront
  • Assume the agent will infer unstated requirements
  • Include irrelevant context
  • Forget to update instructions as needs change

When to Split Agents

Consider multiple specialized agents when:
  • Tasks require different capabilities
  • Different security profiles are needed
  • Workflows can be parallelized
  • Maintenance would be simpler with separation
Multiple focused agents working together often outperform one complex agent trying to do everything.