AI can generate and run tests, but it still needs people who understand the system. Two QA leaders explain why manual testing is evolving, not disappearing.
Every few months, the software industry seems ready to declare manual testing obsolete. With AI now capable of generating test cases, writing automation scripts, and executing tests overnight, the argument sounds convincing: if machines can do the work faster, why do QA teams still need manual testers?
The answer lies in a distinction the debate often overlooks. Executing tests is not the same as understanding what needs to be tested.
AI can automate repetitive actions, but it still depends on people to define expected behavior, identify potential risks, and recognize when a result that looks correct on paper is wrong in practice. That expertise, built through years of working with applications, business processes, and real world scenarios, remains fundamental to software quality.
As AI accelerates both software development and test creation, experienced testers are becoming an increasingly important source of context and judgment.
This was a central theme in the first episode of Katalon's QA Leaders Live series, hosted by Alex Martins, VP of Strategy and AI Transformation at Katalon. The conversation featured two QA leaders with different perspectives on testing and automation.
Lindsay May, a senior user acceptance tester for the Colorado Benefits Management System, helps validate eligibility for public assistance programs such as SNAP and Medicaid. Richie Yu, Managing Director of RYU Solutions Inc., brings 25 years of QA experience, much of it leading testing initiatives at major banks in Toronto. Today, he helps organizations adopt AI-powered testing.
Although their approaches to technology differ, both arrive at the same conclusion: AI is changing how testing gets done, but human expertise remains essential to getting it right.
AI Is Increasing the Demand for Testing
AI powered development tools are changing the pace of software delivery. Developers can now generate code, implement features, and make changes faster than traditional workflows allowed. Yet faster development does not automatically translate into faster, safer releases.
AI generated code can introduce subtle defects, overlook existing business rules, or behave unexpectedly when integrated with complex systems. As the volume of code increases, QA teams face a growing challenge: validating more changes without necessarily having more time, people, or resources.
Organizations often respond in one of two ways. Some release software with limited validation and risk discovering problems through customer feedback. Others increase testing efforts, placing additional pressure on QA teams that may already be stretched thin.
Expanding the team is not always realistic, particularly for organizations operating under tight budgets or strict regulatory requirements.
Lindsay's experience illustrates the challenge. Her team has just 14 testers responsible for validating a benefits management system serving the entire state of Colorado. A single agile project can run for five to six months, requiring sustained collaboration and careful validation throughout development.
When development produces more changes without expanding QA capacity, every hour of testing becomes more valuable.
This is why the idea that AI will simply eliminate manual testing misses the bigger picture. AI may eventually reduce the amount of repetitive work required, but it also creates a greater need to validate what is being built.
The real challenge for QA leaders is not eliminating testers. It is making better use of their expertise.
AI Cannot Test What It Does Not Understand
One of the biggest limitations of AI powered testing is its dependence on context.
As Richie explains, "AI only knows what you tell it."
That limitation becomes particularly significant in organizations where documentation does not fully reflect how a system actually works.
Requirements typically focus on the changes planned for a particular release. They may not capture the complete history of the application, including legacy functionality, previous design decisions, undocumented exceptions, and workarounds that have become part of everyday operations.
In some organizations, original specifications are scattered across outdated SharePoint sites, email conversations, Slack messages, and documents that were never updated after implementation. The system evolves, but its documentation does not always evolve with it.
When AI generates test cases from incomplete requirements, it can produce tests that appear reasonable while overlooking important scenarios. It may validate the latest change without accounting for a business rule introduced several releases earlier.
For Richie, the central problem is not necessarily that AI invents information. More often, it lacks the information needed to make a reliable decision.
"In most cases, it just doesn't have the right information," he explains.
Experienced testers help bridge that gap because they understand the history and behavior of the systems they test.
Lindsay, for example, previously worked as a frontline user of the systems she now tests. That experience gives her practical knowledge of how the application supports real business processes and how its outcomes affect the people who depend on it.
When a result looks unusual, she can draw on her understanding of the underlying policies to determine whether the application has a defect or is behaving as intended. When requirements are unclear, she investigates the live system or consults the business analyst rather than making assumptions.
As she puts it, "I don't like to just assume things."
That mindset is essential in QA. A good tester recognizes uncertainty, asks questions, and seeks evidence before deciding whether something is correct.
AI can help investigate those questions, but it cannot reliably compensate for missing context simply by generating a plausible answer.
Where AI Adds Value and Where Human Judgment Still Matters
Recognizing AI's limitations does not mean resisting its adoption. Instead, it means identifying which activities benefit most from automation and which require human oversight.
Automate Repetitive Execution
AI is particularly useful for tasks that follow a defined process.
Once a test scenario has been reviewed and approved, AI can help transform it into detailed test cases and executable scripts. Work that previously took weeks can, in some situations, be completed in hours.
AI can also reduce the effort involved in analyzing test results. Rather than requiring testers to inspect hundreds of logs individually, AI tools can help categorize failures, identify patterns, distinguish potential defects from test data or environment issues, and summarize findings for stakeholders.
These capabilities allow testers to spend less time on repetitive activities and more time investigating meaningful risks.
Richie captures the distinction simply: "Running a test is just clicking the buttons. Let the AI do that part."
Keep Risk Assessment and Test Design in Human Hands
The more difficult questions come before and after execution.
Which scenarios deserve the most attention? What happens when a requirement is incomplete? Could a change affect a seemingly unrelated workflow? Is a technically valid result actually correct for the customer?
Answering these questions requires an understanding of business priorities, system behavior, historical defects, and the consequences of failure.
AI can support risk analysis and suggest additional test scenarios, but experienced testers must evaluate those suggestions against the realities of the application.
This is especially important when requirements are ambiguous or when the system contains undocumented dependencies.
Make Reviewing AI Output Part of the Testing Process
AI generated tests and scripts should never be treated as automatically trustworthy.
AI output can vary, and a script that looks technically sound may still contain incorrect assumptions, omit critical scenarios, or validate the wrong outcome. Without sufficient domain knowledge, those problems can be difficult to recognize.
Richie compares AI tools to "very smart junior staff." They can work quickly and learn new tasks, but they still need guidance, supervision, and review.
That comparison has practical implications for QA teams. Every deliverable should have a clearly identified person responsible for reviewing its quality and determining whether it is fit for use.
The goal is not to remove human involvement from testing. It is to direct human attention toward the decisions that have the greatest impact on quality.
Why Banking and Insurance Face a Bigger Challenge
The relationship between AI and manual testing becomes particularly important in banking, financial services, and insurance (BFSI).
These organizations often depend on large, interconnected systems that have evolved over decades. New platforms are integrated with legacy applications, business rules are modified to meet changing regulations, and operational processes accumulate exceptions over time.
The result is a complicated environment in which the documented requirements may tell only part of the story.
Consider a banking application that calculates interest or processes loan payments. The underlying logic may depend on historical product rules, previous system migrations, customer agreements, and regulatory requirements. Similarly, an insurance platform may determine premiums or claims using conditions that have been refined over many years.
Testing a new feature in isolation is not enough. QA teams must understand how the change interacts with existing functionality and whether it creates unintended consequences elsewhere.
The consequences of missing a defect can also be significant.
For Lindsay's team, an error in the benefits management system could affect whether someone receives essential public assistance. In banking and insurance, a comparable failure could result in an incorrect payment, an improperly priced policy, or a claim assessed using the wrong rules.
These are not merely technical problems. They can affect people's financial security, create compliance risks, and undermine customer trust.
This is where experienced testers provide a distinct advantage. Their knowledge helps them identify unusual scenarios, question unexpected results, and recognize when the application is behaving differently from the way the business actually operates.
AI can generate tests for a financial workflow, but it needs reliable information to understand which edge cases matter and why. If that context is missing, automation alone cannot guarantee meaningful coverage.
Adoption Takes More Than Buying an AI Tool
Even organizations that recognize the potential of AI testing may struggle to implement it effectively.
In regulated industries, introducing new technology can involve procurement, legal reviews, security assessments, compliance checks, and approval from multiple stakeholders. These processes take time, while AI tools and capabilities continue to evolve rapidly.
Richie notes that "six months is a lifetime" in the current AI landscape. By the time a tool completes an organization's approval process, its capabilities or the available alternatives may have changed considerably.
Teams must also adapt their workflows, establish governance, train employees, and determine how AI generated results will be reviewed and approved, all while continuing to deliver software.
For these reasons, Richie expects the transition to take more than 12 months for many organizations.
The lesson for QA leaders is that AI adoption should be treated as an organizational transformation, not simply a technology purchase.
The Manual Tester Role Is Evolving, Not Disappearing
Ask whether manual testers still have a future, and Lindsay's answer is unequivocal.
"Manual testers are extremely important. AI has its place. Automation has its place. But we're the ones that know the program."
She sees AI as particularly valuable for regression testing after testers have manually validated a change and established the expected behavior. Once the team understands how a feature should work, automation can help verify that subsequent changes do not break existing functionality.
Richie takes a more direct view of how the profession must evolve.
The traditional approach of executing hundreds of predefined test cases in the weeks before a release is increasingly suited to automation. As AI becomes more capable of handling this work, following scripts alone will offer less long term career value.
As Richie explains, "It's no longer enough from a career perspective to say, I know how the system works, and I know how to follow the instructions on what buttons to click."
However, the underlying knowledge remains valuable. What changes is how testers apply it.
Richie expects domain expertise to become increasingly important across roles such as product ownership, business analysis, and subject matter expertise. He describes this knowledge as "the super glue between what's documented and what's in reality."
Together, these perspectives point toward a new model for QA.
The tester of the future will spend less time manually executing predefined steps and more time designing meaningful tests, evaluating AI generated output, investigating unexpected behavior, and identifying risks that automated processes may overlook.
This evolution creates opportunities for testers who are willing to expand their skills. Those who combine deep system knowledge with AI literacy, analytical thinking, and risk based testing will be better positioned to contribute to software quality.
Conversely, testers whose work consists primarily of repetitive script execution will face increasing pressure to adapt as those activities become automated.
What QA Leaders Should Do Next
For QA leaders, the transition to AI powered testing requires a deliberate approach.
First, automate the work that does not require constant human judgment. Repetitive test execution, script generation, log analysis, and reporting are strong candidates for AI assistance, provided the outputs are validated appropriately.
Second, preserve the knowledge experienced testers have accumulated. Document business rules, historical defects, unusual scenarios, system dependencies, and the reasoning behind important testing decisions. This information can improve both human collaboration and the quality of AI generated tests.
Third, establish clear accountability. AI can generate and execute tests, but people must remain responsible for reviewing results, investigating uncertainty, and deciding whether the evidence supports a release.
Finally, invest in the next generation of QA skills. Encourage testers to develop expertise in their business domains, understand how AI testing tools work, evaluate generated test artifacts, and prioritize testing based on risk.
The organizations that benefit most from AI will not necessarily be those that automate the largest number of tests. They will be the ones that combine automation with the knowledge and judgment needed to make testing effective.
Manual Testing Is Not Dead. Its Value Is Changing.
The debate about whether manual testing is dead often focuses on the wrong activity.
Clicking through screens, repeating predictable steps, and executing large suites of predefined tests are increasingly tasks that AI can help automate. But these activities have never represented the full value of an experienced tester.
The real contribution of QA professionals lies in understanding the application, questioning assumptions, interpreting requirements, recognizing unexpected behavior, and deciding what needs to be tested before a defect reaches production.
As AI generates more code and more tests, that expertise becomes even more important.
For QA leaders, the priority is clear: use AI to reduce repetitive work, preserve the knowledge that makes testing meaningful, and ensure that human judgment remains central to quality decisions.
Manual testing is not disappearing. The work is shifting from executing tests to understanding, validating, and improving them.
Want to hear more from QA leaders? Explore upcoming episodes of the QA Leaders Live series, hosted by Alex Martins, VP of Strategy at Katalon.
Frequently asked questions
No. AI is increasingly capable of automating repetitive test execution, but human expertise is still needed to determine what should be tested, interpret unexpected results, and verify that software meets real business requirements.
AI is likely to automate a growing share of the step by step execution traditionally performed by manual testers. However, understanding business processes, interpreting requirements, identifying risks, and reviewing AI generated tests remain important human responsibilities. Testers who develop these skills can adapt to the changing role.
AI testing tools can help generate test cases and automation scripts, execute repetitive scenarios, analyze test logs, categorize failures, and summarize results. They are less reliable when critical context is missing or when a task requires nuanced judgment about business rules and expected outcomes.
Banking and insurance systems often contain complex business rules, legacy integrations, and regulatory requirements that have accumulated over many years. Because this knowledge may be spread across multiple systems and documents, experienced testers play an important role in identifying edge cases and distinguishing genuine defects from unusual but valid behavior.
Manual testers should deepen their understanding of business processes and application behavior while developing skills in AI assisted testing, test design, risk assessment, and reviewing AI generated scripts and results. They should also learn to document business rules, historical decisions, and system dependencies so that both people and AI tools can use that context effectively.

Test everything. Ship with confidence.
From generating test cases to spotting failures, Katalon's AI works across your entire pipeline, automatically.



