Agentic Execution of Manual and Gherkin Tests: Questions to Ask Before You Start

Test execution remains a pain point for many QA teams. Manual test cycles take up a lot of time. Automated scripts work well for stable, frequently run scenarios, but they require technical skills, development time, and ongoing maintenance, even with AI assistance.

Lynqa performs agentic execution of your manual tests and Gherkin scenarios. It interprets the test intent, observes the application, and adapts its actions like a manual tester. A manual test cycle planned over two days with two QA testers can be completed in two hours through parallelized agentic execution, including verification.

Which tests can you hand over to Lynqa, under what conditions, and for what benefit? This article covers those topics.

Agentic test execution relies on AI models’ computer-use capabilities. The agent uses the application with a virtual mouse and keyboard. It reads the test, observes the screen, chooses its actions, and checks the result. Like a manual tester, Lynqa executes the test scenario based on the application’s actual state.

Figure 1 – The agentic loop when executing a manual or Gherkin test

Before starting agentic test execution, examine three topics: applicability to your context, reliability and data security, and expected benefits.

Agentic execution applies to functional UI tests. It can handle manual tests written in natural language, Gherkin/Cucumber scenarios, and tests based on the acceptance criteria of a user story.

It is particularly relevant for:

The level of detail in how a test is written can vary; Lynqa will interpret it and follow your test instructions. A manual QA tester understands the objective, finds the right path, and checks the expected result. Lynqa follows the same logic. It can interpret a high-level instruction and build the path needed to execute it.

Since its release in late 2025, Lynqa has executed thousands of test cases across a wide range of web applications, demonstrating its ability to adapt to varied writing styles and to applications that differ greatly from one another.

A functionally stable environment is not a prerequisite. Agentic execution adapts when the application and its user interface change. 

During agentic execution, Lynqa observes the user interface. It can find a “Submit” button or an “Address” field based on its role in the flow, even if its position or implementation has changed.

With Lynqa, you can run agentic execution directly from your test management tool or trigger it from an AI-powered QA toolchain via the Lynqa API.

Lynqa is already natively integrated with Jira Cloud / Xray, Xqual, and SquashTM.

The Lynqa API enables you to integrate agentic execution into AI-powered QA toolchains, which often include user story refinement and test generation. Lynqa is very well suited to the agentic execution of AI-generated tests.

Agentic execution suits evolving flows, functional tests written in natural language, and the need for immediate validation during the sprint. Scripts remain the right choice for stabilized regression scenarios on stable applications that are run very frequently in a CI/CD pipeline.

The two approaches can follow one another. A team can execute a new test with Lynqa as early as the sprint in which it is created. Once the flow stabilizes, the team can select the scenarios that justify long-term scripted automation.

A test agent must respect the scenario’s intent, check the expected results, and produce a correct verdict.

Smartesting evaluates Lynqa using the Lynqa Assessment Framework, established in 2024, to measure the agent’s reliability. Current benchmarks show a success rate above 92%, measured as “true accuracy”: the verdict is correct for the test and each step, exactly as the QA tester expects. The full Lynqa assessment dataset includes 1,247 scenarios and 215 websites.

The assessment does not focus solely on the final status. It analyzes the step-level verdict, repeatability on an unchanged application, defect detection, and twelve action categories. These include navigation, forms, attachments, data manipulation, logic, and reasoning.

This level of performance makes agentic execution with Lynqa operationally viable. The pilot phase confirms integration into the QA team’s workflow and measures the actual benefits.

Trust also depends on how easily QA testers can check agentic execution. A simple PASS or FAIL status is not enough. With Lynqa, for each step, the tester will find:

Figure 2 – The evidence provided by Lynqa makes agentic execution easy for the QA tester to verify

Verification is part of the process. The agent handles the actions and the evidence. The tester keeps the final judgment. This traceability has been a key factor in adoption: the QA tester makes the final decision on the test verdict.

Figure 3 – Division of tasks in agentic execution: the human decides and verifies, and the agent executes.

Edge cases remain instructive. A misinterpretation may reveal an agent limitation. It may also highlight an ambiguous expected result, inconsistent data, or an implicit business rule. Agentic execution thus helps improve the test repository.

The agent accesses the application in a test environment (never in production) and faithfully performs the specified test steps based on the information visible on the screen. Agentic execution is therefore a controlled and safe use case for AI agent technologies.

Smartesting has achieved SOC 2 Type II compliance for Lynqa Cloud. The independent audit reached a favorable conclusion on the design and operating effectiveness of the security controls and the report noted no exceptions in control testing. 

The Lynqa Self-Managed edition allows on-premises deployment in the customer’s environment, using the customer’s own AI model.

For more details, see our article “Agentic Test Execution: What Security and Compliance Guarantees Should You Require?”.

From two days to two hours for the manual test cycle

The first major benefit is a drastic reduction in execution cycles. Thanks to parallelization and the AI agent’s immediate availability, tests launch on demand. Teams can thus absorb workload peaks effortlessly and speed up critical validations before each release.

Figure 4 – Strong, immediate acceleration of manual test cycles

The benefits also come from sharply reduced costs. Agentic execution is automatic and requires no scripting, no coding, and none of the skills of a test automation engineer. Lynqa pricing starts at €0.89 per test, regardless of test complexity (detailed pricing at www.smartesting.com/en/lynqa/pricing).

The Lynqa ROI calculator, available online, uses the following parameters to determine the result in your context:

Try it: https://roi.lynqa.smartesting.com/.

Three indicators are useful for measuring the benefits of an agentic execution pilot with Lynqa. The first is the number of hours saved per test cycle, including review, the second is the ability to detect the same defects as a comparable human execution and the third is how you use the freed-up time.

Agentic execution is worth evaluating if your context meets several of these criteria:

Start with a representative set. Include common flows, a few edge cases, and end-to-end scenarios. Measure reliability, review time, and benefits over several cycles. You will then have what you need to determine the scope to hand over to the Lynqa agent.