Agentic Software Quality Assurance Platforms

Agentic Software Quality Assurance Platforms is the name Gartner's market is transitioning toward. The Magic Quadrant under the old name, AI-Augmented Software Testing Tools, published its first and so far only edition on October 6, 2025, and that scorecard is the anchor for this article. The rename is the story.

What the two names score differently

The old name describes a tool: testing, augmented with AI. Faster tests, self-healing scripts, generated cases. The new name describes a function: quality assurance, owned by agents.

That is a bigger promise. An agentic SQA platform does not run tests faster. It plans what to test, generates the coverage, executes it, decides what the results mean, and reports to humans who now supervise instead of author. The old category scored the speed of the test suite. The new category will score the judgment of the agent.

Testing is the first discipline the AI era put under pressure. The software now writes itself, which means the test suite has to write itself too, and the vendor that owns quality owns the whole delivery loop.

What the inaugural field actually contains

The first-ever Magic Quadrant for AI-Augmented Software Testing Tools published October 6, 2025, authored by Joachim Herschmann, Sushant Singhal, Ross Power, and C.A. Swan. Four Leaders emerged, and they won four different arguments.

Tricentis took the highest position on Ability to Execute, the incumbent test-automation company that grew into AI rather than being born in it. Keysight leads with Eggplant, intelligent computer vision, autonomous test design, and self-healing capabilities, including support for air-gapped environments, which matters to the defense and industrial buyers nobody mentions in AI demos. UiPath entered from the automation platform side, a vendor that treats testing as one workflow among many. OpenText leads with its Core Software Delivery Platform, where AI testing sits inside a full delivery stack.

Four different companies won four different arguments, and the quadrant rewarded all of them. Applitools and Katalon appear in the field but outside the Leader rung, the specialist layer the market is already consolidating around.

The October 6, 2025 Magic Quadrant for AI-Augmented Software Testing Tools, in full

The report is young enough that its projection is still the headline: Gartner expects 70 percent of enterprises to have integrated AI-augmented testing tools into their software engineering toolchains by 2028, up from 20 percent in early 2025.

Read the two numbers together. The quadrant exists before the adoption. Most enterprises are still buying traditional test automation while the vendor field has already moved to a new name. The scorecard is ahead of its market, which is exactly where a first edition belongs.

The adoption gap the prediction exposes

A 20 percent baseline in 2025 rising to 70 percent by 2028 is not a smooth curve. It is a forced migration, and forced migrations are where buyers make their worst decisions.

The gap is between what the category promises and what most organizations can absorb. AI-augmented testing assumes the test estate is already structured enough to augment. Most test estates are not. The vendors sell the destination, and the quadrant scores the vendors, but the journey runs through the buyer's own cleanup first.

The agentic half that does not exist yet

The transition name promises agents. The published scorecard mostly scores augmentation: generated cases, self-healing scripts, computer vision that finds what changed on screen.

True agentic quality assurance, where the agent owns planning, coverage decisions, and pass-or-ship judgment, remains mostly roadmap. Gartner's own ITSM-adjacent research found the same pattern in its market: nearly every vendor promotes agentic AI, and very few deliver genuine autonomous planning and multistep execution. The quality assurance market is one cycle behind the same curve.

The quadrant scored augmentation. The market is buying agents. The next edition under the new name will have to score the difference.

What the first scorecard cannot settle

A first edition has no track record of its own. The four Leaders have not yet defended their placements against a second scoring, and the rename means the next edition will run under different criteria anyway.

The honest limit is the name itself. Between the October 2025 scorecard and the next one, the category changes identity, and placements earned under the old definition transfer only partially. A Leader in augmentation is not automatically a Leader in agency.

Four questions for the quality owner

What does the agent own, exactly? Generation, execution, or judgment? Most vendors sell the first two. The third is the category's future and the buyer's risk.

Can the platform see what the user sees? Computer-vision testing is the difference between testing the code and testing the product. Ask which one the vendor actually does.

Does it work air-gapped? The interesting deployments are the ones the marketing does not mention, and the quadrant's citations show which vendors planned for them.

Which argument do you need? The incumbent, the vision platform, the automation vendor, and the delivery stack each won for a different reason. The quadrant will not choose for you.

Gartner does not endorse any vendor, product or service depicted in its research publications, and does not advise technology users to select only those vendors with the highest ratings or other designation. Gartner research publications consist of the opinions of Gartner's research organization and should not be construed as statements of fact. Gartner disclaims all warranties, expressed or implied, with respect to this research, including any warranties of merchantability or fitness for a particular purpose.

The same shift shows up next door, in AI-Augmented Code Modernization Tools. Gartner's inaugural quadrant splits the field between agentic rewriters like AWS and Microsoft and a deterministic search specialist in Moderne, and AWS alone claims 7.9 billion lines of code analyzed.