Testlio has launched LeoCore, a software testing platform that combines AI-led test creation with human validation, alongside its 2026 Software Quality Report.
The report found that 34% of Testlio clients uncovered at least one critical or high-severity defect in the first testing cycle, while 77% of failures involving AI assistants were serious enough to damage user trust and product integrity.
The findings highlight persistent weaknesses in software releases as companies push products out more quickly. Functional bugs made up the largest share of problems identified in initial testing cycles, at 44%, including broken logins and onboarding flows. Payment failures followed at 22%, ahead of UX and usability issues at 20%, and other areas such as localisation, accessibility and regression at 14%.
LeoCore is designed to manage the testing lifecycle in one platform. It uses AI to create tests quickly, while human testers assess issues automated checks may miss, particularly around usability, tone, local market behaviour and edge cases on specific devices.
Quality risks
Testlio's research focuses heavily on AI assistants, where failures often went beyond technical errors. According to the report, these issues were frequently significant enough to affect product integrity or weaken users' confidence in a service.
Darin Brown, Chief Product and Technology Officer at Testlio, said this exposed a gap in software checks.
"The report is a wake-up call. Teams are shipping faster than they can verify, and AI is widening that gap, not closing it," Brown said.
He added: "A passing test suite tells you what you asked it to check. It won't tell you the onboarding flow is confusing, or the chatbot's answer is technically right but tone-deaf. That's what LeoCore is built to catch."
Testlio has spent more than a decade providing testing services to companies in sectors including fintech, eCommerce, media and software. Those industries often face heightened commercial and regulatory risks when products fail in live environments, especially where payments, customer identity or user communications are involved.
LeoCore draws on more than 13 years of testing data to assign testers, run test cycles and analyse patterns over time, with the aim of identifying release risks before products reach users.
Human layer
A central element of the platform is Testlio's global network of testers and specialists in areas such as payments, AI and localisation. The network operates across more than 600,000 real devices, 800 payment methods, 150 countries and 100 languages.
That breadth reflects a wider industry challenge: software may pass standard automated checks yet still fail in real-world conditions because of device combinations, local wallets, network differences, accessibility gaps, or cultural and language issues.
Dean Hickman-Smith described that distinction in practical terms.
"Automation alone can only validate what it's designed to detect. Testlio's global network of expert testers uncovers the real-world issues that matter most: the payment that fails with a local wallet, the AI response that offends or erodes customer trust, the accessibility gap, or the edge case that only appears on a specific device or in a particular market. That's the difference between software that passes tests and software customers trust," Hickman-Smith said.
Testlio also cited Strava as an example of how its testing model has been used in practice. According to Testlio, the fitness app increased release velocity sixfold and set a six-to-nine-hour turnaround for insights gathered by athletes testing in 13 locations.
Benjamin Wolak, Mobile QA Engineer at Strava, commented on that work.
"Testlio knows how important speed is to our growth. I love Testlio's flexibility and willingness to accommodate our needs. When I send Testlio a release, I know it will be ready," Wolak said.
Market pressure
The launch comes as software teams face growing pressure to shorten development cycles while integrating generative AI tools into customer-facing products. That shift has created tension between faster production and the slower, more nuanced work of checking how software behaves for end users.
Testlio's report is based on client engagements over the past 12 months and AI testing experiments using thousands of real-user prompts across multiple AI assistants. The results are intended to show patterns across its testing programmes and connect them to broader software industry trends.
The report's central message is that automated testing alone is not enough when products increasingly depend on AI systems and operate across fragmented global markets. In Testlio's figures, more than a third of clients found severe defects in the first round of testing, and AI-related failures were especially likely to threaten user trust and product integrity.