AI
The core principle of Step is to bring teams and technologies onto a single, scalable platform and unify software automation disciplines:
- Functional testing
- Load and performance testing
- Synthetic monitoring
- Robotic process automation (RPA)
The Step AI ecosystem brings greater autonomy across the testing and automation lifecycle through a set of AI-powered technologies, workflows and capabilities designed to reduce effort while improving the quality of automation outcomes. It can be adopted at different levels of autonomy, from exposing individual Step capabilities to your AI assistant with the Step MCP to delegating complete automation workflows to the Step AI Agent.
Choose your level of autonomy
The Step AI ecosystem consists of three complementary layers, each making a different trade-off between flexibility, guidance and autonomy:
| Product layer | Agentic primitive | What it provides | User considerations |
|---|---|---|---|
| Step MCP | Tools | Atomic Step capabilities | Maximum flexibility, more integration and prompting effort |
| Step Skills | Tools + instructions | Reusable expert workflows | Faster adoption, guided and reusable patterns |
| Step AI Agent | Tools + instructions + workflow | Operational autonomy | Minimum effort, standardized operation, less flexibility |
The three layers can potentially address every step of the testing lifecycle, from creating and consolidating the test plan, through the technical implementation, to analysing results and feeding the findings back. Those steps can be addressed in isolation with Step MCP, or as partial or complete workflows with Step Skills and the Step AI Agent.
Step MCP exposes Step capabilities to the AI. Step Skills teach the AI how to use those capabilities. The Step AI Agent turns them into an engineered operational workflow.
There is no requirement to move from one layer to the next. The right approach depends on the level of flexibility and autonomy required.
The layers are also not mutually exclusive: they compose. A Skill can instruct an AI assistant to use Step MCP, and individual steps of the Step AI Agent’s workflow can themselves be invoked through Step MCP.
Step MCP
The Step MCP Server exposes Step capabilities to AI assistants and coding agents such as Claude Desktop and Claude Code through the Model Context Protocol (MCP).
Built on top of the Step CLI and Step APIs, it gives AI agents direct access to capabilities across the testing lifecycle. They can create and update automation, execute it on Step infrastructure, analyse results and inspect artefacts.
This is the most flexible way to use Step with AI: you choose the AI, tools, prompts and workflow.
👉 Go to the Step MCP Server documentation
Step Skills
Step Skills add expert guidance on top of these capabilities.
They package testing know-how, Step-specific automation patterns and repeatable procedures into instructions that AI agents can invoke on demand, including how to use the Step MCP Server.
Instead of prompting an AI assistant from scratch, teams can reuse established procedures for tasks such as creating automation, validating it, analysing executions or running regression workflows.
Step Skills remain flexible: the user or AI assistant orchestrates the overall workflow, while the Skill provides a proven way to perform a particular task.
Step Skills are not yet available as a separate download.
👉 Learn more about Agent Skills, or reach out to discuss your use cases.
Step AI Agent
The Step AI Agent provides a more autonomous approach by taking responsibility for the overall automation workflow.
The agent takes high-level requirements, the systems under test (SUTs) in scope, and existing test cases or an existing codebase where available. From there, it scans the existing assets, evaluates the gaps, plans the implementation, scaffolds the technical setup for each technology involved, then generates a Step Automation Package, executes it, validates the result, and uses execution feedback to repair and stabilise the automation before handing it over.
Within that workflow, it investigates only where needed and acts deterministically wherever possible. The workflow is engineered for reproducibility, reliability and efficient evolution of growing test suites, and it contains the guardrails to maintain those objectives.
Similar workflows can be assembled using general-purpose AI assistants, MCP servers and Skills. You can design and maintain the orchestration yourself, including the controls required for reproducibility, reliability and efficient suite evolution. With the Step AI Agent, this workflow is engineered, standardized and validated as part of the product.
This becomes particularly valuable as AI-generated automation moves beyond individual experiments and needs to operate reliably across teams and growing test suites.
👉 Go to the Step AI Agent documentation
Engineered for reproducibility
Agentic systems contain inherently non-deterministic components. Reliable automation therefore requires more than producing a working test once.
The Step AI Agent follows a controlled workflow for planning, generation, execution, validation and repair. We benchmark this workflow repeatedly to evaluate both:
- Operational usability: success rate, stability (absence of flakiness), test-to-specification fidelity and package compliance
- Operational efficiency: workflow duration and cost
Our first benchmarking study evaluates these properties through repeated independent runs of the same end-to-end browser scenario against a self-hosted e-commerce application.
📊 Read “Agentic Testing - Part 1: Reproducibility”
Built to evolve test suites
Generating the first test case is only the beginning. In real projects, automation suites grow continuously as new requirements and scenarios are added.
One approach is to feed the entire existing codebase back to the model when generating each new test, which becomes increasingly expensive as the suite grows. The Step AI Agent instead works with the existing automation artefacts and extends them incrementally, reusing what has already been created.
Our second benchmarking study measures the cumulative cost of growing a suite under both strategies, for growth in breadth and in depth. Cumulative cost is the figure that matters in practice, since automation is extended continuously over the life of an application rather than recreated from scratch.
📊 Read “Agentic Testing - Part 2: Scaling Test Suites”
Which approach should I use?
Use Step MCP when you want maximum control over your AI stack and workflow. This is a natural fit for developers already working with coding agents, IDEs and CI/CD pipelines, and for individual initiative rather than standardized practice.
Add Step Skills when you want to give those agents repeatable testing procedures and Step-specific expertise without giving up control of the overall workflow.
Use the Step AI Agent when you want the complete automation lifecycle to be handled as a standardized operation, with planning, generation, execution, validation, repair and incremental suite evolution engineered as part of the product.
These approaches are complementary and can coexist within the same project. Teams can use Step MCP and Step Skills for highly flexible workflows, particularly during exploration and early development, and adopt the Step AI Agent for more standardized, autonomous operation as projects and systems mature.
Step MCP and Step Skills give you the workshop; the Step AI Agent gives you the assembly line.
Frequently asked questions
Do I need the Step AI Agent if I already use Claude, Playwright, MCP and Skills?
Not necessarily. General-purpose AI assistants, MCP servers and Skills give developers the building blocks to create sophisticated AI testing workflows themselves. In fact, this is where we started too.
As we developed and tested these workflows, however, we found that relying on the AI and its instructions to orchestrate the overall process made it harder to control not only the outcomes, but also how those outcomes were produced. This led us to progressively formalize the workflow around the AI: defining how it plans, generates, validates, executes and repairs automation, and where deterministic logic should take over.
The Step AI Agent productizes this approach into an engineered and validated workflow, while Step MCP and Step Skills remain available for teams that prefer greater flexibility and control over their own orchestration.
Do I need an LLM in my CI/CD pipeline to run AI-generated automation?
No. The Step AI Agent produces a standard Step Automation Package, which runs on the same Step execution infrastructure as hand-written automation, with no AI involved at execution time. Generated packages integrate like any other: see Integration with CI/CD pipelines.
Getting started
See Getting started with the Step AI Agent for prerequisites, setup and a first run, or reach out at contact@exense.ch to discuss your use case and see a demo.