Testing Agents: QA Without Humans

Quality assurance has always been the team that runs out of budget first. Companies spend on features, designers, and marketers. QA gets what's left. The result is well-known: bugs ship, customers complain, engineers get paged at 3 AM.

The testing agents in msitarzewski/agency-agents don't fully replace a QA team, but they can dramatically extend whatever QA capacity you have. This article walks through the roster and explains how to deploy it.

Key Takeaways

Testing agents cover unit, integration, E2E, accessibility, performance, and security testing
Agents write test plans, author test code, and review existing test suites for gaps
Best used as a force multiplier for existing QA, not as a full replacement
The Test Planner agent alone can save a day per sprint on planning overhead
All agents work with any modern testing framework: Jest, Vitest, Playwright, Cypress, etc.

The testing roster

Test Planner

The Test Planner agent is the project manager of your test suite. Given a feature spec or PR description, it produces a structured test plan covering happy path, edge cases, error conditions, and non-functional requirements.

The plans are organized by test type (unit, integration, E2E), prioritized by risk, and annotated with effort estimates. A typical plan for a medium-sized feature takes 5 minutes to generate and would take a human QA lead 30-60 minutes to produce at similar quality.

Unit Test Author

Writes unit tests in your framework of choice. Supports Jest, Vitest, Pytest, Go's testing package, RSpec, JUnit, and a dozen others. Tell it which framework and it will match conventions.

What makes this agent strong is its focus on meaningful assertions, not just coverage metrics. A generic AI will write tests that achieve 100% coverage but assert nothing useful. This agent won't.

Integration Test Author

Handles the trickier world of integration tests: database interactions, external API mocks, message queues, and service boundaries. Uses patterns like test containers and fixture factories.

E2E Test Author

Writes Playwright, Cypress, or Puppeteer tests for browser-based E2E scenarios. Knows to wait for network activity, handle timing flakiness, and use stable selectors.

Accessibility Tester

Complements the Accessibility Specialist design agent with automated test scenarios. Writes axe-core assertions, keyboard navigation tests, and screen reader compatibility checks.

Performance Tester

Writes load tests in k6, Artillery, or JMeter. Includes realistic user behavior models, not just raw request flooding.

Security Tester

Partner to the Security Engineer. Writes tests that verify security controls work as expected: auth checks, input validation, rate limits, CSRF protection.

Test Reviewer

A specialized reviewer that audits existing test suites for quality issues. Flags:

Tests that assert nothing meaningful
Tests that will pass even when the code is broken
Flaky test patterns
Missing edge cases
Coverage gaps in critical paths

Run this on any legacy codebase and prepare to be humbled by what it finds.

How to deploy them

For a small team without dedicated QA, here's a recommended starting lineup:

Test Planner runs on every major feature spec
Unit Test Author is invoked by developers during implementation
Test Reviewer runs on every pull request as a final gate
E2E Test Author handles critical user flows at release time

Four agents cover 80% of what a small QA team would cover, at a fraction of the coordination cost.

Example: catching a bug that humans missed

A team we work with was shipping an onboarding flow. Their human QA ran through the happy path, tested a few edge cases, and signed off. After deploy, customers started reporting that the flow broke when their company name had an apostrophe.

Post-mortem, the team ran the Test Reviewer agent on the existing test suite. Within seconds, it flagged: "No tests cover inputs with special characters including apostrophes, quotes, and backslashes. This is a common source of SQL injection and rendering bugs. Recommend adding a dedicated edge case test category."

The bug wouldn't have shipped if they'd run the agent first. Total cost of the agent: about $0.15. Cost of the incident: several hours of engineering time and customer frustration.

Where agents struggle with testing

A few honest limitations:

They don't run the tests. The agents write test code, not execute it. You still need a test runner and CI infrastructure.

They're weak on domain-specific business logic. A tax calculation test needs domain knowledge the agent doesn't have. You have to provide the expected values.

They can't explore an app manually. Exploratory testing is still a human skill. Agents write scripted tests; they don't play with the product.

They occasionally write flaky tests. Modern frameworks help, but agents can produce tests with timing dependencies that break on CI. The Test Reviewer catches most of these.

Pairing with development agents

Testing agents pair beautifully with development agents. The pattern looks like this:

Frontend Developer implements the feature
Test Planner produces a test plan
Unit Test Author writes unit tests
E2E Test Author writes end-to-end tests
Test Reviewer audits the result before merge

This entire pipeline can run in a single session with the Agents Orchestrator in charge.

Frequently Asked Questions

What testing frameworks do they support?

All major frameworks: Jest, Vitest, Mocha, Jasmine, Pytest, unittest, Go testing, RSpec, JUnit, NUnit, Playwright, Cypress, Puppeteer, k6, Artillery, JMeter.

Can they write tests for untested legacy code?

Yes. The Legacy Code Rescuer agent (in the engineering category) pairs with Test Reviewer and Unit Test Author to add characterization tests to legacy systems.

How do I run agent-generated tests in CI?

Same as any other tests. The agents write code; your CI pipeline runs it. Claude Code can even commit the tests directly if given permission.

Are these better than AI test generation tools like Codium?

Different approach. Dedicated tools are good at bulk generation; agents are better at strategic, focused tests with meaningful assertions.

What about mutation testing?

Not directly supported by the agents, but the Test Reviewer agent evaluates tests using similar heuristics (would this test catch the bug if I introduced one?).

Scale your QA

Testing is high-leverage work, and agents are the ultimate leverage for high-leverage work. A small QA team plus a handful of testing agents punches well above its weight. Install them today and ship with more confidence tomorrow.

Browse all 150 agents at aiskill.market/agents or submit your own skill.