Tricentis, a company that describes itself as a global leader in agentic quality engineering, used its Transform 2026 conference in Dallas on August 20 to introduce three new AI-powered technologies aimed at a problem that has become one of the biggest headaches in enterprise software: figuring out whether AI-generated code and AI agents can actually be trusted before they go live.
The announcement centers on Tricentis Labs, an incubator the company set up to give customers and partners early access to emerging AI capabilities before they’re finished products. Through it, Tricentis introduced Tricentis Aida, an agent that explores applications on its own to find defects; Tricentis AgentScore, a framework for grading whether AI agents are safe to deploy; and Tricentis Release Risk Intelligence, a tool that flags what’s risky about a software release before it ships.
Of the three, AgentScore is the one worth paying closest attention to — not because it’s the most technically complex, but because it’s trying to solve a problem that has no established playbook yet: how do you know if an AI agent is behaving the way it’s supposed to?
Table of Contents
What Tricentis Announced
The three technologies were unveiled at Tricentis Transform Dallas, the company’s flagship conference, held August 19–20 at the Arlington Convention Center. Tricentis frames Labs as a testing ground rather than a product launch pad: ideas go in, get shaped by real customer use, and only the ones that prove themselves graduate into what the company calls “enterprise-ready solutions.” That framing matters for readers evaluating whether to expect a polished, generally available product — for now, these are previews, not finished offerings.
It’s also worth being precise about the language Tricentis itself uses. The original coverage around this story described three “AI agents,” but that’s not quite how Tricentis characterizes its own work. Only Aida is explicitly called an AI agent in the company’s materials; AgentScore is a “technology,” and Release Risk Intelligence an “intelligent capability.” That distinction matters because Tricentis Labs is explicitly an incubator, not a general release channel — there’s no published pricing, no confirmed enrollment process, and no stated timeline for general availability. Tricentis also hasn’t said whether AgentScore will be sold standalone or bundled into existing licenses, and no customer has been named as an active pilot user of any of the three tools specifically.
What Each Technology Does
| Technology | Category | What It Does |
|---|---|---|
| Tricentis Aida | AI agent | Autonomously explores web and Windows desktop applications, surfacing defects and coverage gaps without an existing test suite or scripts |
| Tricentis AgentScore | Evaluation technology | Observes AI agent behavior in real workflows and generates a composite score with review, block, or ship recommendations |
| Tricentis Release Risk Intelligence | Intelligent capability | Surfaces release-scoped coverage gaps, ranks risk by severity, and offers contextual AI-powered actions for release decisions |
AgentScore is the most conceptually ambitious of the three. Rather than testing an application, it evaluates other AI agents — a shift Tricentis describes as moving “from deterministic testing to probabilistic evaluation.” Traditional software testing checks whether code does exactly what it’s supposed to every time; AI agents can behave differently run to run, which is why a scoring and confidence-based approach is needed instead of a simple pass/fail.
What none of the three announcements include is the technical detail a more rigorous evaluation would need. Tricentis hasn’t disclosed the specific benchmarks AgentScore uses to compute its scores, nor which AI models — proprietary or third-party — power any of the tools. Readers should treat AgentScore’s grading system as a stated concept rather than a documented methodology.
Why It Matters — The AI Agent Trust Gap
The timing of this announcement lines up with a real industry pain point. According to a survey by the research firm Futurum Group covering 820 enterprises in the first half of 2026, 55.4% of organizations named AI agent reliability and hallucination management in production as one of their top challenges in adopting generative AI. That’s a company-independent data point, and it helps explain why a tool like AgentScore — built specifically to answer “is this AI agent ready for production?” — has obvious market appeal.
Tricentis has also pointed to its own research showing that confidence in AI-driven release decisions dropped to 34% in its 2026 Quality Transformation Report. That figure comes from Tricentis itself, not an independent source, so it should be read as the company’s own diagnosis of the problem it’s trying to solve rather than an externally verified statistic.
Put together, the picture is consistent even if the numbers come from different places: enterprises are moving fast on AI adoption and are nervous about whether they can trust what they’re shipping. AgentScore is a direct response to that anxiety, even if its internal workings remain undisclosed for now.
Part of a Bigger Platform Push
This announcement doesn’t stand alone — it’s the latest step in a strategy Tricentis has been building for the better part of a year.
| Date | Development |
|---|---|
| October 2025 | Tricentis previews a unified AI Workspace at Transform 2025, linking its AI agents and services for a planned 2026 release |
| March 2026 | Tricentis launches its Agentic Quality Engineering Platform, orchestrating test creation, automation, performance testing, and quality intelligence agents |
| May 2026 | Tricentis and SAP announce SAP Enterprise Continuous Testing, an SAP-specific AI test-generation product |
| July 30, 2026 | Tricentis acquires Tabnine, an AI-coding platform, to add its “Enterprise Context Engine” to the platform |
| August 19–20, 2026 | Tricentis Transform Dallas conference |
| August 20, 2026 | Aida, AgentScore, and Release Risk Intelligence announced via Tricentis Labs |
The Tabnine acquisition is the most relevant piece of context. Tabnine’s Enterprise Context Engine builds a structured map of an organization’s systems from code, documentation, and infrastructure records, giving AI agents system-level understanding rather than generic retrieval. It addresses the same underlying problem as AgentScore from a different angle — one gives agents better context, the other grades whether agents are behaving well once they have it. Performance figures Tricentis has cited elsewhere, such as up to 60% automation of regression test grids, relate to the existing platform and Tabnine integration — not to Aida, AgentScore, or Release Risk Intelligence, which are too new to have performance data of their own.
How Tricentis Compares to Other AI Testing Tools
Tricentis operates in a crowded field that includes Applitools, Mabl, Sauce Labs, BrowserStack, LambdaTest, and Katalon. Independent comparisons in 2026 generally rank Applitools strongest in visual regression testing and Mabl among the more advanced platforms for autonomous test agents, while Tricentis is typically grouped with Katalon as an enterprise-scale option. AgentScore’s focus — evaluating other AI agents rather than testing an application directly — appears to have no directly comparable competitor offering based on available reporting, though this is an early and unverified differentiation, since no independent analyst comparison of AgentScore exists yet.
What’s Still Unknown
A number of basic questions remain open, and Tricentis hasn’t answered them in its announcement materials:
- What specific benchmarks or metrics AgentScore uses to compute its composite scores
- Which AI models — proprietary or third-party — power any of the three technologies
- Pricing and whether AgentScore will require a separate license or come bundled with existing Tricentis products
- A timeline for when any of the three technologies might reach general availability
- Any named enterprise customer currently piloting Aida, AgentScore, or Release Risk Intelligence specifically
None of this is unusual for an early-access preview, but it does mean the announcement should be read as a statement of direction rather than a fully documented product launch.
Conclusion
Tricentis’s latest announcement is less about a finished product and more about where the company thinks quality engineering needs to go next: toward tools that can evaluate AI agents with the same rigor traditionally reserved for testing conventional software. AgentScore, in particular, is a direct answer to a well-documented industry concern — that enterprises are adopting AI agents faster than they can verify them.
What isn’t yet clear is how well any of this works in practice. The technologies are in early access, the grading methodology behind AgentScore hasn’t been published, and no independent party has evaluated any of the three tools. Whether AgentScore becomes a meaningful new standard for AI agent trust, or one of many competing attempts to solve the same problem, will depend on details Tricentis hasn’t shared yet — including how transparent the company is willing to be about the scoring system once it moves beyond the preview stage. Readers evaluating these tools for their own organizations should watch for a general availability announcement, published pricing, and — most importantly — independent scrutiny of how AgentScore’s scores are actually calculated.
FAQ
What is Tricentis AgentScore? AgentScore is a Tricentis Labs technology that evaluates AI agents by observing their behavior in real-world workflows and generating a composite quality score with a review, block, or ship recommendation, intended to help enterprises judge whether an AI agent is ready for production use.
Is Tricentis AgentScore available now? It’s in early access through Tricentis Labs, an incubator for testing emerging technology with customers before it becomes a fully supported product. No general availability date, pricing, or public enrollment process has been disclosed.
What’s the difference between Tricentis Aida and AgentScore? Aida is an AI agent that autonomously explores applications to find defects and coverage gaps. AgentScore doesn’t test applications directly — it evaluates the behavior and reliability of other AI agents.
How does AgentScore evaluate AI agent behavior? By observing how an agent behaves in real workflows and recommending what should be measured, then generating a composite score. Tricentis has not disclosed the specific benchmarks or metrics behind that scoring process.
Is AgentScore a standalone product or part of an existing Tricentis license? Not publicly disclosed. Tricentis hasn’t specified a licensing or pricing structure for AgentScore.
What AI models power Tricentis’s new agents? Not publicly disclosed. Tricentis hasn’t said whether Aida, AgentScore, or Release Risk Intelligence rely on proprietary models, third-party large language models, or a combination.
How does this relate to Tricentis’s Tabnine acquisition? Tricentis acquired Tabnine on July 30, 2026, to integrate its Enterprise Context Engine — a system that builds a structured map of an organization’s software environment — into its broader Agentic Quality Engineering Platform. That acquisition and the new Labs technologies both address the same underlying issue: giving AI agents the context and oversight needed to operate reliably in enterprise environments.