Beyond Manual vs Automation: A New Test Execution Paradigm

For years, functional testing has existed in a sort of ongoing tension between two approaches: manual testing and automated testing. In a recent post, I argued that framing them in opposition misses the point. The real challenge isn’t choosing one over the other, but deciding thoughtfully when automation makes sense, and when keeping tests manual delivers more value.

On one side, there’s manual testing driven by people, exploration, and contextual understanding. Testers interact directly with the application, clicking through features, validating business rules, spotting inconsistencies, and catching behaviors that are challenging to formalize. Manual testing is especially valuable for exploring new features, evaluating usability, and interpreting unexpected behaviors. However, it can become repetitive, time-consuming, and difficult to scale as delivery cycles accelerate.

On the other side, there are automated testing scripts, frameworks, and CI/CD pipelines executing predefined scenarios. Automation shines when tests need to be repeated quickly and reliably, especially for regression protection and fast feedback. But it also introduces costs: maintenance effort, UI fragility when interfaces evolve, and the continuous work required to keep test suites aligned with the application’s reality.

For a long time, the conversation has sounded like a duel: manual vs. automated. Even more striking, I’ve often heard people say:

“Manual testing is dead; everything will be automated soon.”

But a new test execution paradigm is emerging, and it’s not about replacing one with the other. It’s about introducing a third player: manual test execution by AI agents.

Manual test execution by AI agents is exactly what it sounds like: AI agents executing tests written in plain English (or another language) that were traditionally performed manually, without relying on rigid, line-by-line automation scripts.

Instead of hard-coded selectors and brittle test steps, AI agents interpret intent.

You don’t necessarily describe how to click every button. You describe what needs to be validated.

For example:

“Make sure a new user can register, receive a confirmation email, and log in.”

An AI agent can interpret that objective, navigate the interface, adapt to layout changes, identify elements visually or semantically, and execute the flow, much like a human tester would. The difference? It doesn’t get tired. It scales instantly. And it doesn’t require you to maintain hundreds of fragile scripts.

This shifts the effort:

It feels less like writing code, and more like delegating a mission.

When electric scooters appeared in cities, they didn’t eliminate walking. And they certainly didn’t kill bicycles. Walking is still the most natural way to move short distances. Bikes are still perfect for longer commutes or sport. But scooters introduced something new, an in-between option. They reshuffled urban mobility.

Some people who used to walk started using scooters for slightly longer distances. Some cyclists switched for convenience. And new habits emerged that didn’t exist before.

The ecosystem didn’t shrink; it reorganized. That’s exactly what’s happening with manual test execution by AI agents.

Let’s be clear: manual test execution by AI agents is not here to eliminate manual testers. Humans remain essential for:

And it’s not here to eliminate traditional automation either.

Scripted automation is still unbeatable for:

What changes is the distribution of effort.

There’s a whole category of tests that used to sit in an uncomfortable middle:

That’s where manual test execution by AI agents shines: it absorbs that middle layer.

It’s not a replacement story. It’s a role redistribution story.

A quick disclosure: at Smartesting, we are exploring this paradigm directly with Lynqa, an AI agent for manual test execution integrated with Xray. It executes manual tests and produces detailed step-by-step execution reports.
That experience informs the perspective shared in this article.

You should use AI agents for test execution when the goal is to scale the execution of existing manual tests, especially in UI-driven functional testing while preserving traceability and reducing the maintenance burden of scripted automation.

If a test is:

Then an AI execution agent is often an excellent option.

Manual test execution by AI agents is a paradigm shift, not because it wipes out what existed before, but because it changes how everything fits together.

Just like electric scooters didn’t replace walking or cycling, AI agents don’t replace manual or automated testing, they rebalance the ecosystem. The real shift isn’t just technical. It’s cultural. It changes how teams think about execution, ownership, and scale.

The question is no longer: “Should we do this manually or automate it?” It becomes: “Who, or what, is best positioned to execute this test?”

A human? A script? An AI agent?

That’s the reshuffle. And once you see it that way, it’s difficult to go back to the old binary debate.