1.2What Makes an Agent Different

An agent does not follow a fixed script. Instead, it operates in a loop:

1. Observe the current state of the system
2. Reason about what to do next, considering the objective and what it has already tried
3. Act on its decision
4. Evaluate whether the objective has been met

This loop -- called the ReAct pattern (Reason + Act) -- is the foundation of all agentic testing. Originally described for general-purpose LLM agents, it maps directly to testing workflows. And it is no longer a research curiosity: as of July 2026, agentic testing is a mainstream discipline, with the ReAct loop packaged inside first-party tools (Playwright Test Agents) and widely adopted SDKs (Stagehand, browser-use).

Here is the same login test, expressed as an agent objective:

# Agentic test -- adaptive, goal-oriented
agent = TestAgent(
    objective="Log in with test@test.com / password123 and verify dashboard loads"
)
result = agent.run(max_steps=20)
assert result.status == "pass"

When the agent encounters the scenarios that break scripts, it adapts:

Scenario	Script Response	Agent Response
Button ID changed	Fails: element not found	Searches for alternative selectors, finds the button by text
Cookie popup appears	Fails: popup blocks interaction	Observes popup, dismisses it, continues
Page loads slowly	Fails: timeout	Observes loading state, waits, retries
2FA field added	Fails: unexpected page	Observes new field, reports environment configuration issue
Account locked	Fails: wrong URL	Reads error message, reports "Account locked -- environment issue"

The agent provides diagnostic information, not just pass/fail.