Two and a half years after the release of ChatGPT, the evolution of LLM (large language model) technology is driving what some call the “second internet revolution” in enterprise digital transformation. One notable trend is that large corporations are beginning to move away from traditional SaaS solutions, aiming instead to internalize operations through the use of AI.
The news that Klarna is phasing out its use of Salesforce and Workday sent shockwaves through the industry. This shift isn’t limited to third-party tools, however. Companies are also undergoing radical internal changes.
For example, Amazon CEO Andy Jassy has openly stated that AI will lead to workforce reductions. The impact is already apparent. In the U.S., tech giants like Amazon, Google, and IBM have conducted massive layoffs—primarily among white-collar workers—while ramping up investments in generative AI infrastructure.
In this ongoing LLM paradigm shift, where anything seems possible, what strategies should startups pursue to survive?
This pressing question was the focus of the first session of the BRIDGE × MUFG Innovation Partners (MUIP) seminar series held on June 12, 2025. The event brought together representatives from LayerX, MNTSQ, and MUFG Bank to discuss real-world approaches for startups facing off against the big tech goliaths.
How enterprise in-house AI is reshaping the playing field
Takashi “Taka” Sano, Chief Investment Officer of MUFG Innovation Partners (MUIP), opened the seminar with the following remark:
We’ve entered an era where you can’t go a day without interacting with generative AI.
He cited a U.S. report published about 18 months ago, which explored the impact of AI on employment, highlighting its particularly profound effect on knowledge workers.
According to the report, the higher the income level of a profession, the more susceptible it is to disruption by generative AI. The legal, computer science, mathematics, engineering, financial analysis, and business operations fields were identified as especially vulnerable.
One striking finding from the report was that among the top 30 companies with a high concentration of such roles, 11 were financial institutions, with tech and telecom companies also heavily represented.
Just as robot automation once revolutionized manufacturing, generative AI is now bringing even greater impact to knowledge-based industries. I think many of you can relate to that. (Mr.Taka)
AI has continued to evolve rapidly since 2024. “We’ve now entered the era of Agentic AI,” Taka added, referring to the next stage beyond generative AI—autonomous agents that can carry out complex tasks independently. This shift is fundamentally changing the traditional relationship between companies and software.
Until now, people have had to adjust their work processes to fit the software implemented under corporate IT budgets. But going forward, software will adapt to the way people work. (Mr.Taka)
This marks a pivotal transition from an era in which businesses adapt to software to one in which software adapts to business.
This transformation is especially significant for Japanese companies, where a declining and aging population has brought them to a crossroads: whether to hire people or hire AI.
This trend is already playing out overseas. Swedish fintech company Klarna recently announced that it would stop using major SaaS platforms like Salesforce and Workday in favor of building its own AI-driven internal systems. Klarna has canceled around 1,200 SaaS contracts and is shifting toward fully in-house AI infrastructure.
Retail giant Walmart is another example, having deployed its internal AI assistant platform “My Assistant” to 75,000 employees to streamline tasks such as drafting and summarization. These in-house AI initiatives at enterprises reflect a dramatic shift in market dynamics—one that startups can no longer ignore.
Ryuhei Itaya, CEO of legal tech startup MNTSQ, captured the situation using a metaphor: “It feels like we’re facing a dark overlord now.”
Just two years ago, the AI space was a playground dominated by startups. But that landscape has changed dramatically. Unfortunately, the overlord armies have arrived—especially in the enterprise domain. And there are some ‘overlords’ that simply cannot be defeated: Microsoft, Google, Salesforce, SAP, and Workday. These giants are now leading in the AI era. (Mr. Itaya)
Microsoft is pioneering with Copilot, Google is countering with Gemini, and Salesforce has launched its first AI-powered SaaS offering with Agentforce. The competitive environment for startups has become vastly more difficult with these well-resourced giants entering the fray.
“It’s a brutally difficult era to survive,” Itaya admitted. So how can startups navigate and endure in this age? In the next section, we’ll look at case studies of startups that are forming strategic collaborations with large enterprises to not only survive but to thrive.
MUFG Bank × LayerX: Building company-specific AI
What specific strategies can startups adopt to compete against the tech goliaths? One answer lies in the collaboration between MUFG and LayerX. For the past year and a half, the two companies have been jointly developing an AI-driven sales knowledge-sharing project, showcasing how advanced AI technology can address the complex challenges unique to large enterprises.
Ryuya Nakamura, Executive Officer and Head of the AI/LLM Division at LayerX, described their approach as follows:
We realized there’s an opportunity to create a new kind of market—one that’s closer to a ‘high-mix, low-volume’ environment where traditional software hasn’t been able to operate effectively. Especially in the enterprise space, the wide variation in client needs made conventional SaaS models difficult to implement. But with LLMs, we believed we could overcome that.
Deep-rooted challenges in enterprise sales
Tomohiko Kimura, Senior Manager at the Corporate Banking Planning Department of MUFG, explained the background behind their AI adoption as follows:
Our department handles sales to large corporations, and our proposals are often highly customized. While some of our top salespeople are extremely skilled, their expertise tends to be very personalized and siloed. There’s little cross-communication—even within adjacent teams in the same division. This made knowledge sharing a major issue.
In enterprise sales, it’s not just about loans. Teams are expected to propose sophisticated solutions that cover governance, ESG, investor relations, and other strategic domains. However, such expertise often stays with individuals and isn’t shared across the organization.
Initially, the bank tried to address the issue using standard tools like Microsoft Teams and SharePoint. But the results turned out underwhelming.
They didn’t stick at all. People weren’t motivated to share. There were issues with the effort required, reluctance to share carefully crafted proposal materials, and poor usability—you had to run full-text searches to find anything. (Mr.Kimura)
It was during this time that MUFG Bank discovered LayerX, whose AI-driven solution offered the right mix of speed, scalability, and flexibility. Developing a similar system in-house would have been costly and time-consuming, so the decision was made to adopt LayerX’s platform as an external SaaS solution.
LayerX’s platform AI Workforce is designed specifically to handle long documents, supporting tasks such as extracting and summarizing data from materials like sales proposals or financial reports that can span hundreds of pages. A key strength of the platform is its focus on designing user experiences that help correct AI hallucinations (i.e., factual errors in generated content).
Copilot is designed for broad use cases, so its UI is inherently a ‘lowest common denominator.’ In contrast, our platform is tailored for long-form documents and offers an experience optimized for catching and correcting mistakes made by AI. (Mr.Nakamura)
A differentiation strategy: AI that excels within the company
According to Ryuya Nakamura, the key differentiation between LayerX’s solution from general-purpose tools like Microsoft’s Copilot lies in the distinction between portable and non-portable skills.
Some human skills are portable—they can be applied across different companies—while others are non-portable and only make sense within the context of a specific organization. Solutions that support portable skills will likely become globally popular with little customization or support. (Mr.Nakamura)
Tasks such as development and research are relatively portable; they follow universal logic and workflows, making them well-suited for general-purpose AI tools.
In the same way that new employees can contribute quickly in these roles, AI tools that assist with such tasks also quickly become useful. (Mr.Nakamura)
However, many jobs in the real world are deeply rooted in company-specific knowledge. These tasks require an understanding of the internal context, culture, and cross-functional dynamics unique to the organization. Just like new hires need structured onboarding and training to be effective in such roles, the same applies to AI.
This is exactly where LayerX sees its opportunity.
Our vision with AI Workforce is to build AI that thrives specifically within your company’s unique environment. That vision is what drives the functional differentiation of our product, especially when compared to more generic tools. (Mr.Nakamura)
Keys to success in enterprise adoption
Technical excellence alone does not guarantee success in the enterprise market. As Kimura pointed out, meeting the specific requirements of large corporations is critical.
The biggest hurdle is security compliance.
Unfortunately, for large corporations like ours, there are a lot of rules and constraints,” said Kimura. “From the perspective of a financial institution, we have to adhere to strict security standards. Even when startups have excellent solutions, they’re often not considered simply because they can’t meet these security requirements. (Mr.Kimura)
Another major challenge is integration with legacy systems.
Some of our core host systems date back to when I was born. We still have parts of those systems running, so compatibility with existing infrastructure is essential. (Mr.Kimura)
Yet, meeting these requirements can lead to substantial rewards. According to Kimura, the startups that succeed are not only technically advanced but also those that “have deeper knowledge in areas where we want to grow, are highly specialized in those domains, and can still move quickly and flexibly while meeting enterprise-grade requirements.”
Currently, MUFG and LayerX are working together on what comes after simple knowledge sharing. While the system currently focuses on improving the accessibility of stored information, their next step is more ambitious: enabling AI to generate proposal documents that rival those created by the bank’s top salespeople.
This collaboration highlights a crucial insight: the most promising business opportunities lie in company-specific operational areas that general-purpose AI tools cannot effectively address.
By offering highly customized AI solutions—optimized over a two to three-month tuning period—startups can meet the complex demands of large enterprises while carving out differentiation from Big Tech companies.
MNTSQ’s strategy: competing in whitespace
While LayerX is pursuing a strategy focused on customizing AI for company-specific operations, another player is tackling the enterprise market from a completely different angle. Legal tech company MNTSQ is taking on large organizations—including MUFG Bank—by deliberately targeting what it refers to as whitespaces.
CEO Ryuhei Itaya, a former attorney turned entrepreneur, exemplifies this strategic shift. The founding of MNTSQ was sparked by an invitation from his university classmate, AI engineer Takahiro Yasuno. Though Itaya had been diligently practicing law, a deep sense of frustration with the structural issues of the legal industry ultimately led him to leap into the world of AI.
Through his legal work, Itaya came to believe that much of what lawyers do is not truly essential. Tasks like drafting favorable clauses in intentionally complex language, or generating risk-avoidance language, often form the core of the job—but lack real substance.
This discontent became the foundation for MNTSQ’s mission: transforming the legal industry in the age of AI. Itaya began to question the very nature of contracts and became convinced that the industry had to evolve.
Moreover, legal services are among the most susceptible to disruption by AI.
After becoming a lawyer, I found myself doing things like making sure contracts used 10.5-point Mincho font, or confirming that Article 6 was followed by Article 7. I mean—are those tasks meaningful? (Mr.Itaya)
He grew increasingly disillusioned with how such mechanical tasks had become central to the legal profession.
On top of that, the legal sector lacks structured knowledge systems. According to Itaya, the reason legal knowledge is so difficult to standardize is that the data itself is inherently unstructured—primarily consisting of files and documents. To handle such data efficiently and effectively, large language models (LLMs) are essential.
Differentiation strategy: MNTSQ’s advantage in a whitespace
What makes MNTSQ’s business strategy particularly compelling is its focus on contracts, a whitespace domain without strong competition.
When it comes to contract data, I can’t think of any dominant player. Companies like SAP and Salesforce aren’t rolling out highly specialized features for contracts. So in that sense, we’re fortunate to be in an enterprise market segment that’s still open. (Mr. Itaya)
This strategy reflects a deep understanding of product architecture. Itaya breaks down products into three layers: database, application, and UI/UX. In the past, having a sleek UI and calling a tool “AI-powered” was often enough to launch a product. But now, Itaya argues, we’ve entered a new era where survival depends on answering deeper questions: What data do we uniquely own? What tasks can only be done on our platform?
MNTSQ is building precisely this kind of structural strength in the domain of contract data. As it continues to expand use cases with major enterprises such as MUFG, the company is amassing proprietary contract data and developing applications on top of that foundation. The result: a platform where all contract-related tasks—from drafting to negotiation—can be completed entirely within MNTSQ, any time of the day.
Another key differentiator is MNTSQ’s use of prompt engineering by legal professionals.
I believe all lawyers should become prompt engineers. The best lawyers in Japan are writing prompts, and those prompts are used to generate and negotiate contracts. When regulations change, we work with law firms to update the prompts.
No matter what kind of contract is uploaded, it benefits from prompts written by top-tier lawyers. This is not only a value-add for users—it’s also a survival strategy for the legal profession. (Mr.Itaya)
MNTSQ has formed partnerships with three of Japan’s Big Four law firms—a highly unusual move, given their typically competitive dynamics. This collaboration reflects a shared sense of urgency by these firms that their profession is at serious risk of disruption.
MNTSQ also invests in the creation of highly reliable data for AI to reference. For example, when laws change, top lawyers from these firms write official legal commentaries. These documents become trusted sources that generative AI can rely on.
According to Itaya, producing such high-quality data—and teaching AI to reference it—is both a way to raise product quality and a strategic lifeline for lawyers.
By continually generating expert-level, trustworthy insights, MNTSQ is creating a flywheel where their content earns social credibility, leading to AI beginning to reference it, and continued
AI usage reinforces credibility.
We are very enterprise-focused. Among Japanese startups, I think we’re quite unique in this regard. Of the top 100 companies in Japan by revenue, we’ve already secured a significant market share. (Mr.Itaya)
The case of MNTSQ highlights the importance of identifying whitespaces and building structural strength within them. By focusing on highly specialized tasks that general-purpose AI tools struggle to address, and by collaborating with domain experts, MNTSQ has created differentiation through both the data and application layers—demonstrating a viable path to competitive advantage in the age of AI.
The practical examples of LayerX and MNTSQ reveal a clear strategy for surviving the brutality of the AI era. They differentiate themselves from general-purpose tools through deep technical understanding; increase replacement costs through structurally robust products; address and meet enterprise-grade requirements; and build moats by collaborating with experts in specialized domains.
To survive in the era of Big Tech goliaths, startups need a comprehensive strategy that combines all of these elements.
Even as large corporations accelerate internal development and AI adoption, their examples suggest a path forward: focusing on company-specific operations and highly specialized fields can still unlock new business opportunities in the age of enterprise AI.
