In an Asia Insights conversation in Tokyo, MagicPod CEO Nozomi Ito explains why AI-powered test automation is becoming critical as software teams ship faster, rely more on AI-generated code and face new quality risks
TOKYO, July 11, 2026 — AI is changing the way software is written. But as more developers and non-engineers use AI to generate code, a new question is becoming urgent: who is checking whether that code actually works?
For Nozomi Ito, CEO of MagicPod, the answer lies in the next phase of AI-powered test automation. Speaking to AsiaBizToday as part of the Asia Insights series at MagicPod’s office in Tokyo, Ito said software testing is moving from a repetitive manual function into a strategic layer of software quality, reliability and business risk management.
MagicPod is a test automation cloud service for mobile app and browser-based web app testing. It helps software teams automate repetitive quality assurance tasks such as entering data, clicking buttons, checking values, validating workflows and confirming whether a website or mobile application behaves as expected.
“When software engineers create a website or mobile app, they need to check whether it works correctly. That is called testing,” Ito said. “Usually, engineers input various values, click buttons and confirm whether the value is correct. This is manual work, and this is very troublesome.”
MagicPod’s role, he said, is to bring AI and automation into that process. “MagicPod helps this testing process with AI and automation. By using MagicPod, software testing activity becomes very efficient and automated,” he added.
From manual QA to AI autopilot
Ito said AI is making test automation easier to use, including for people who do not write code. In earlier versions of test automation, users had to scan a webpage, select interface elements and manually create automated test steps. Today, MagicPod’s AI autopilot allows users to give natural language instructions.
“Users just need to go to the test target webpage and tell MagicPod AI autopilot, ‘Please check this login page works as expected,’” Ito said. “Then MagicPod analyses the user’s intention and automatically generates the necessary steps.”
Once those steps are created, the user can run the test and MagicPod performs the checks automatically. This shift matters because traditional test automation tools have often required programming skills. Tools such as Selenium helped engineers automate browser testing, but they were difficult for many QA teams, testers and business users who understood the product but did not write code.
Ito has seen this transition closely. He has worked in test automation for more than a decade and was involved in Japan’s Selenium community long before AI became a dominant technology theme. “When I established this company 14 years ago, test automation was not common at all in Japan,” he said. “It was common outside Japan, but not common in Japan.”
At the time, many Japanese companies released software infrequently, sometimes only once a year. In that environment, test automation often seemed too expensive or cumbersome because the setup cost was high and testing was not repeated frequently.
That changed as agile development became more common. “More and more people started agile development. Software improved more frequently, and new versions were released more frequently,” Ito said. “Testing happened every month or even every week. They needed to do the same repetitive manual testing every week. So it did not make sense to continue this manual activity.”
Why no-code test automation matters
The rise of agile development created demand for automated testing. But coding-based automation tools still excluded many users. “In Japan, testers or QA people mainly test products. They frequently do not have coding skill or programming skill,” Ito said. “For them, Selenium is not the best solution. But manual testing is also not the best solution.”
That gap led MagicPod toward no-code test automation. The goal is to allow engineers, QA teams and non-technical users to create and run automated tests without writing scripts. This is increasingly important because software quality is no longer only an engineering issue. Product teams, marketing teams, system integrators, founders and business teams are also becoming responsible for digital product quality.
Ito said MagicPod’s customers broadly fall into two groups. The first is companies that operate their own websites, SaaS products or mobile apps. The second is system integrators and contract development companies that build and maintain software for clients.
Both face the same pressure: release software faster while maintaining quality and controlling cost.
The speed, quality and cost challenge
For software teams, quality assurance has always involved a difficult trade-off. If teams test extensively by hand, cost rises and release speed slows. If they reduce testing to move faster, reliability suffers. If they automate, they reduce repetitive work, but traditional automation itself requires effort to create and maintain.
Ito said AI test automation helps companies balance speed, reliability and cost more effectively. “If they do tests manually, it is very difficult to satisfy these three requirements,” he said. “If they are testing a lot, cost becomes high and speed becomes very slow. But if they emphasise speed, testing cost becomes low, but reliability is sacrificed.”
Automation allows the same tests to be run repeatedly without adding proportional human effort. AI pushes this further by reducing the cost of creating and maintaining automated tests. “AI reduces the cost of test automation further and improves the speed further,” Ito said. “Because it improves the speed of both development and testing.”
Where AI adds the most value
Ito sees AI creating value in two major areas: test creation and test maintenance. For companies that have not yet adopted test automation, the biggest barrier is the effort required to create tests. AI lowers that barrier by allowing users to describe what they want to test in natural language.
“For users who are still hesitant to automate software testing, the biggest barrier is the cost of test creation or setup cost,” he said. “AI reduces the barrier of test creation.”
For more experienced automation users, the bigger pain point is maintenance. Software products change continuously. User interfaces are updated, fields are added, workflows change and buttons move. Every change can break existing test scripts, forcing teams to spend time maintaining them.
“Many people gave up test automation due to this high maintenance cost,” Ito said. “But AI helps this maintenance flow. For these experts, test maintenance is the biggest advantage of AI.”
This is one of the most important shifts in AI-powered testing. Traditional automation is powerful, but it can be brittle. AI can make automated tests more adaptive, helping them respond to interface changes and reducing the burden of manual script correction.
Ito also believes AI can improve reliability by checking for problems that humans or traditional scripts may miss. Manual testers can overlook issues. Traditional automated tests only check what they are explicitly instructed to check. If a test script does not include a specific validation, the automation engine will not perform it.
“But AI is much smarter,” Ito said. “We can just ask AI to check if there are no broken images or there are no issues. With this simple instruction, AI tries to check all kinds of problems.” That could make software quality assurance more comprehensive, especially as applications become more complex and release cycles shorten.
AI-generated code creates new QA risks
Ito said AI coding tools can improve speed, but they can also introduce problems if developers rely on them too heavily. “Thanks to AI testing tools and AI coding tools, it becomes easier to guarantee quality,” he said. “Many quality assurance activities can be automated. But at the same time, AI is another quality issue.”
The risk is that developers may ask AI to generate large amounts of code and then publish it without fully understanding or checking the output.
“Engineers publish it without checking it because it is too complicated or too large to check everything,” Ito said. “Then after publishing it, they face an unfamiliar issue, an unexpected issue.”
If engineers did not write the code themselves and cannot understand it fully, debugging becomes harder. Asking the same AI to fix a bug it failed to detect in the first place may not always work.
“Developers are facing another quality issue,” Ito said. “Maybe currently, developers are depending on AI too much.” His advice is pragmatic: AI should not be used beyond a team’s ability to verify and control the output.
“If engineers cannot check code by themselves, maybe it is too much for them to use AI,” he said. “The industry is now searching for the best balance.”
Japan’s AI adoption is moving faster than SaaS
Ito believes Japan’s adoption of AI is moving faster than its earlier adoption of SaaS. Japan has historically been slower than some markets in adopting SaaS, but he said the AI trend is different. Startups, IT companies and large enterprises are actively discussing AI tools such as Claude, Gemini, Copilot and ChatGPT.
“Already, almost all startups and IT companies began to adopt AI,” he said. “Even big enterprise companies are seriously thinking about AI.”
Many CEOs and executives in Japan are now pushing their companies to become AI-first or AI-native. “I think in the next one or two years, almost all companies will seriously adopt AI,” Ito said. “The speed is much faster than ever. This is clearly different from the SaaS trend.”
For MagicPod, this creates an opportunity to expand the market for AI test automation. Historically, test automation in Japan was limited partly because a large share of development budgets were still tied to waterfall development and one-off testing. AI could help automate a wider range of enterprise testing activities that were previously difficult to automate economically.
What happens to SaaS in an AI-first world?
Ito expects AI to change the SaaS market, but he does not believe general AI models will replace every specialised software tool. Instead, he sees general AI models such as Claude, Gemini and Copilot becoming central hubs, while specialised tools perform tasks that general AI cannot do well.
“It is difficult to do everything using a general AI model,” he said. “They don’t know industry knowledge, or they cannot deeply penetrate into customers’ office. Also, they don’t touch real-world things.”
In software testing, for example, general AI may understand instructions, but specialised tools such as MagicPod can interact with real devices, browsers, test environments and workflows.
“If customers tell AI what they want to do, AI calls MagicPod, and MagicPod is in charge of necessary tasks,” Ito said. This points to a future where SaaS companies must integrate more seamlessly with AI systems. The category itself may change, but specialised tools that solve specific operational problems will remain relevant.
“Some SaaS companies cannot survive, but new SaaS companies also emerge,” Ito said. “Their name may not be SaaS anymore, but different IT companies or software companies surely appear.”
Beyond engineering: QA for marketers and vibe coders
MagicPod’s customer base is still mainly in Japan, but the company is increasingly looking beyond its domestic market. Ito said the company is exploring overseas opportunities in the United States and Asia. It is also developing products for broader quality assurance use cases outside traditional engineering.
One such product is SiteRover, aimed at helping marketing and web teams check websites, landing pages and product pages for issues such as broken links, spelling errors and accessibility problems.
Another product, Kaiba, is aimed at the emerging world of “vibe coding”, where non-engineers use AI tools to create websites or applications. “Many non-engineers started to create websites using vibe coding tools,” Ito said. “But it is not easy for them to fix all problems.”
This reflects a broader shift. As AI makes software creation more accessible, quality assurance must also become more accessible. The future users of testing tools may not only be engineers or QA specialists, but marketers, product managers, founders and business teams using AI to create digital products.
Why software testing is becoming strategic
Asked whether the AI boom is a bubble, Ito said the market may currently show bubble-like behaviour, but the underlying trend is real.
“Currently this is kind of bubble, so maybe this bubble could be broken at some point,” he said. “But AI trend is true. AI will be more and more widely adopted.” The larger message from Ito’s conversation is that software testing is no longer a back-office technical activity. It is becoming strategic.
As companies ship software faster, rely on AI-generated code, expand digital products and depend on web and mobile experiences for customer engagement, quality assurance becomes central to business performance.
A bug is not just a technical defect. It can affect customer trust, revenue, security, brand reputation and operational continuity. AI-powered test automation offers a way to reduce repetitive manual work, support non-coders, improve test maintenance, accelerate release cycles and identify issues more comprehensively.
But Ito is also clear that AI is not a cure-all. It creates new quality risks even as it solves old ones. The challenge for companies is to use AI in ways that improve speed without losing control.
For MagicPod, that balance defines the next phase of software testing: AI should help teams build and test faster, but human judgement, verification and accountability must remain central. In an AI-first software economy, code testing will not become less important. It will become more important than ever.