August 29, 2026
On August 20, 2026, at its Tricentis Transform event, enterprise testing vendor Tricentis introduced three new AI-powered quality-engineering tools. DevOps.com covered the launch in detail on August 26, 2026, and it's worth being upfront that this is a look at a recently announced product suite, not breaking news -- the announcement is nearly a week old at this point. Still, it's a useful signal of where a big chunk of the software industry thinks testing is headed next. The three tools are: Tricentis Aida, an autonomous agent that explores web and Windows desktop applications on its own to surface defects and coverage gaps, without needing someone to have already written a test suite or scripts; Tricentis AgentScore, which watches how an AI agent behaves in real workflows and produces a composite score along with a recommendation to proceed with or block a deployment; and Tricentis Release Risk Intelligence, which scans a release for coverage gaps, ranks them by severity, and recommends next steps. Tricentis' Chief Product Officer, Eran Sher, framed it as a pace problem as much as a quality problem, saying, as quoted by DevOps.com, that AI is changing not only how software is built, tested, and released, but also how quickly enterprises need to innovate. The timing isn't incidental: this launch comes a month after Tricentis' July 2026 acquisition of AI-coding vendor Tabnine, which added what Tricentis calls an Enterprise Context Engine to its Agentic Quality Engineering Platform. Put plainly, Tricentis now owns a tool that helps AI write code and a tool that checks whether that AI-written code is safe to ship.
Give this tool suite credit for targeting a real problem. When an AI agent generates source code -- or, increasingly, when an AI agent takes actions inside a live application -- a human reviewer often can't just read the output and know whether it's trustworthy. The logic is buried inside generated code, and the agent's behavior in a real workflow can diverge from what the code looks like it should do. AgentScore exists because organizations doing large-scale AI code generation genuinely need something like it: a way to grade an opaque process before it touches production. For a business generating thousands of lines of AI-written code a week, or wiring AI agents into workflows with real financial or customer consequences, a tool like Release Risk Intelligence that prioritizes coverage gaps by severity is a legitimate operational need, not busywork.
AgentScore exists to answer one question: should we ship or block what this AI agent just wrote or did? That question only exists in its current form because the code is opaque by default. The AI wrote it, the AI's output isn't something most people can just glance at and understand, and so a second AI system has to be built to grade the first one's work before a human can trust it. That's a whole new layer of tooling and process a team now has to adopt on top of the original code-generation tool -- and Tricentis' own product roadmap, buying an AI-coding company one month and shipping an AI-code-grading suite the next, is a pretty direct illustration of how quickly that stack is compounding. A visual, no-code builder sidesteps a meaningful chunk of this problem, not because it's magic, but because of how the thing is built in the first place. In ViibeStack's app builder, business logic isn't hidden inside generated source code that needs a separate AI system to interpret -- it's expressed as visible fields, statuses, and workflow steps that the person who built it, or anyone reviewing it afterward, can look at directly and understand. There's no black box between the intent and the result that needs a composite score and a ship/block recommendation from a second AI just to be legible.
It would be dishonest to stretch this into a claim that no-code apps need no testing or review. They absolutely do. A visible workflow can still be built wrong: a status field that doesn't trigger the right notification, an approval step that skips a required check, a report that pulls from the wrong date range. Mistakes are still possible, and verifying that a workflow actually does what you intended is real work -- work that shows up in guides like our own walkthroughs on building a client approval workflow or an inventory and reorder-alert app, where getting the logic right the first time matters. The honest distinction is narrower than "no-code needs no QA." It's that is this AI-written code safe to ship and does this visible workflow do what I expect are two different categories of problem. The first requires a specialized, and increasingly expensive, industry of AI-grading-AI tools because the artifact being shipped is inherently opaque. The second is a review problem that any competent team can do by looking directly at the thing they built, because the artifact was never opaque to begin with. Tricentis is building real, valuable tooling for the first category. A visual builder was simply never exposed to that category in the same way -- which is a meaningful part of the buy vs. build vs. ViibeStack calculation for any team weighing how much new process overhead an AI-generated-code workflow actually costs beyond the code-generation tool itself.
Sources