Step 31: Meet Step Studio and the Step AI Agent

What if everyone on your team could build automation the way they prefer, visually or in code, on their own machine, with AI doing the heavy lifting?

With the release of Step 31 “Margna”, we’re introducing two major innovations: Step Studio, which brings the low-code experience of Step to your local machine, and the Step AI Agent, which turns requirements into validated, self-healing automation packages.

Below you’ll find the highlights of the release. For the complete list of bug fixes and enhancements, visit the release notes.

Step Studio (beta): Low-Code, Now Local

For years, the visual editors of the Step platform have let testers and business users build automation without writing code, while developers worked on the same automation as code in their IDE and pipelines. Step Studio brings these low-code advantages from the central platform to your local machine.

Less technical users get a local-first workflow without leaving the visual editor. Developers get low-code editing right next to their code.

Why Step Studio

Local-first development for low-code users: Testers and business users can now apply modern development practices such as Git, versioning, branching and CI/CD to their automation, while still building and maintaining it visually, without writing code.
Low-code editing for developers: Developers combine visual automation with code-first development, switching between the visual editor and their IDE at any time, depending on the task.
Unified automation across teams: Developers, testers and business users collaborate on the same codebase and automation assets, each with the tools and level of abstraction that suit them.
Faster development with local execution: Teams build, run, debug and iterate on automation locally, with fast feedback and less dependence on central infrastructure.
One-click deployment to the Step platform: Locally developed automation is deployed to the Step platform in a single click, ready for distributed execution, orchestration, scheduling and reporting.

Step Studio is also the gateway to the second highlight of this release: the Step AI Agent. With Create with AI, users generate and extend automation from plain-language test cases, then inspect, refine and maintain the result visually in Step Studio or directly in code.

Under the hood

Step Studio works directly on your Automation Package files, the same YAML descriptors and source code your IDE uses. There is no duplication and no synchronization step. With Step Studio you can:

Define keywords, build plans with the visual plan editor, and manage parameters and schedules
Execute plans locally on all Step agent types (Java, Node.js and .NET)
Execute plans or whole packages on a remote Step instance, or deploy them there directly
Reload changes made outside of the Studio, for instance in your IDE or after a git pull

Get started

👉 Step Studio ships with the Step CLI. Launch it with step studio
👉 Check out the Step Studio documentation to get going

Step AI Agent: From Requirements to Validated Automation

The Step AI Agent is an autonomous agent that turns high-level requirements, test cases or existing code into validated and executable Step Automation Packages. In this first release, it covers functional testing and RPA, with load testing and synthetic monitoring coming soon.

The AI Agent combines AI reasoning with a controlled, multi-phase workflow. AI handles the steps that need interpretation and investigation, and deterministic logic handles everything that can be specified reliably. The workflow is engineered for reproducibility, reliability and the efficient evolution of growing test suites.

The result is ordinary Step automation that can be versioned, reviewed, shared and executed on the same infrastructure as the rest of your automation, without depending on AI.

Key features

End-to-end agentic workflow: understand the request, plan, review the plan, explore the live application, implement, execute, repair and report
Self-healing and stability checks: the AI Agent diagnoses failures from real execution evidence and repairs the automation. It tells broken automation apart from invalid test cases and genuine application failures
For testers and developers: testers run the AI Agent from Step Studio with Create with AI, developers from their IDE or CI/CD pipeline through the Step CLI
Local or central execution: validate the generated automation locally first, then on a Step instance to catch flakiness on real infrastructure
Modular, reusable automation: the AI Agent takes full advantage of Step’s unique modularization and abstraction features. Generated keywords and plans become reusable building blocks, so later runs extend existing automation instead of starting from scratch
Full transparency: every run produces a report with the verdict, coverage per test case, duration and cost, plus the intermediate plans, specifications, logs and traces

Step AI Agent at a glance. On the left, its users: testers working from a UI and developers working from an IDE. In the middle, the agent plans, generates an automation package and executes it in CI/CD, with an AI feedback loop running back from each stage. On the right, the use cases served: functional testing, load and performance testing, synthetic monitoring and robotic process automation.

Get started

Beyond Studio and AI

Step Studio and the Step AI Agent are the headline features of Step 31, and the release also brings improvements to Automation Packages, the CLI, the Step MCP Server and reporting.

Key highlights

Automation Packages: string interpolation lets plain values embed variables with the ${...} syntax, for instance http://${host}:8080/api, both in the UI and in YAML (see Embedding variables in text values), and the callPlan syntax has been simplified
CLI: the local execution of Automation Packages now supports all Step agent types (Java, Node.js and .NET), and new Step CLI Docker images make CI/CD integration easier
Step MCP Server: optimized report browsing, improved YAML validation with better error reporting, and local execution of all Automation Packages, like the CLI. The server now also starts without a central Step instance
Reports & analytics: performance analytics in node details, customizable cross-execution views, and CSV export of table dashlets

⚠️ Step 31 contains breaking changes: Keyword Packages are removed and replaced by Automation Packages, and plain values containing ${ are now evaluated. Existing content is migrated automatically in most cases. See the migration guide before upgrading.

📖 Explore everything that’s new in Step 31.

Read the full release notes