AI Test Authoring brings agentic, natural language test creation into Test Builder, grounded in the metadata-driven engine enterprises already trust.
Every team wants AI-generated tests. Very few teams trust them.
That gap is the real story of AI in test automation today. Generic AI agents can produce a test script in minutes, however what they produce is a guess: a script written in the backend, based only on what the AI thinks it sees on screen. It looks finished, then it fails on the first real run, and someone spends the afternoon fixing it.
Provar’s AI Test Authoring closes that gap. Open a copilot chat inside Test Builder, describe your scenario in plain English, and the AI agent reasons through the prompt, drives the browser, and builds the test in front of you.
Not in a black box. Not in a background job you review later. On your screen, step by step, as it happens.
1. Describe the business scenario
Type the scenario the way you would explain it to a colleague:
“Create a new Lead, Qualify it to convert it into an Opportunity, and change the stage to Closed Won.”
Test Builder drives the browser and creates the steps automatically. Authoring drops from hours to minutes.
Just as important, it changes who can author. Business analysts, manual testers, and subject-matter experts can produce a working first draft themselves, without writing a line of code. The people who know the business process best are no longer waiting in a queue behind automation engineers.
2. Watch the test build and execute live
For end-to-end flows, accuracy is everything. So the test builds live in the browser, and every step is executed right there as it is created.
You watch it happen. If a step goes wrong, you catch it in the moment, not three days later in a CI failure. Human validation stays in the loop by design, which is exactly where enterprises need it.
Compare that with tools that generate scripts in the backend: the script arrives, it has to be reviewed, corrected, and re-run, and the time you saved on authoring is spent again on rework.
3. Grounded in Provar’s application intelligence
Most AI testing agents are a layer of AI on top of a browser driver like Playwright. They can only reason from what is rendered on screen: the DOM, the labels, the pixels. That inference is fragile, and the tests it produces are brittle.
AI Test Authoring reuses Provar’s existing Salesforce, NitroX component capabilities and locator intelligence across multi-page, mixed-technology flows. Tests are grounded in real application metadata: object schemas, field API names, page structures.
The AI is not guessing what that lookup field is. It knows, because the engine knows.
The result is a test that behaves like one an experienced Provar engineer would have built by hand, only faster.
4. AI authors. Provar’s deterministic engine executes.
AI is non-deterministic by nature. It can behave differently every time, and that is a problem when an enterprise needs the same test to produce the same result on every run.
Provar uses AI deliberately and only where it adds value. AI handles the reasoning and the building of the scenario. Once the test is built, execution is fully deterministic, running on Provar’s existing locator engine and metadata technologies.
This split also keeps AI costs predictable. Execution never calls an LLM, so you are not paying tokens every time a test runs. AI is spent once, at authoring, and the reasoning itself is optimised to build the scenario in fewer steps and fewer tokens. For teams running thousands of executions a day, that difference shows up directly in the AI bill.
5. Enterprise control: your LLM
Customers connect their own LLM provider to power AI Test Authoring. Prompts and application data flow through the customer’s own model, within their own infrastructure, under their own security policies. Enterprises keep control of data handling, security, cost, performance, and vendor choice.
An honest note here: bring-your-own-LLM on its own may not be unique to Provar, and we will not pretend it is. What is different is what your model is asked to do here: reasoning at authoring time, grounded by Provar’s metadata, and never sitting in the execution path. Because everything stays within your infrastructure, model choice becomes genuine control over data, security, and cost, rather than a checkbox.
6. Business aware specialised app skills
Pluggable application skills give the agent business awareness of the app under test: it understands Salesforce objects, record pages, and business flows, and Microsoft Dynamics entities and forms, rather than treating every screen as a generic web page.
And skills are not limited to what ships in the box. Rather than relying on UI properties alone, your team can define application-specific business flows inside a skill, making the AI aware of how your business actually works. Salesforce, Microsoft Dynamics, and generic web applications are supported today, with more skills planned.
Generic AI agents vs. AI Test Authoring
Same promise on the surface, plain English in and tests out. The difference is what the AI works from, and where it stops.
Generic AI agents generate scripts. Provar generates trust.
See it in action
AI Test Authoring is available in Test Builder. Describe your first scenario in plain English and watch it build, execute, and prove itself live. Book a demo or reach out to your Provar account team.
