SaaS Is Moving from Chosen to Called — Built-in AI, MCP, and the Shifting Ground Under Software

最終更新日: 2026-05-28

The way software gets chosen is quietly changing. Until now, SaaS competed by getting you to come to its own dashboard and do your work inside it. But since the start of 2026, some of the largest SaaS companies and financial institutions in the world have begun rebuilding their systems so that external AI agents can call them. SaaS is moving from being chosen to being called. In this article, I want to walk through that structural shift, and what I'm thinking about it as a participant — as honestly as I can.

The reason SaaS exists is being rewritten

In May 2026, at ServiceNow's annual conference "Knowledge 2026" in Las Vegas, there was one declaration.

The CEO, Bill McDermott, said this from the stage: "ServiceNow goes beyond the platform of platforms to become the AI agent of agents. It will be the thing that connects any model, any cloud, and any data source."

The weight of those words becomes clear once you know what kind of company ServiceNow is. It's worth around 20 trillion yen. It sits deep inside the IT, HR, finance, legal, and procurement operations of large enterprises around the world. Until now, the company grew by polishing its own dashboard and workflows relentlessly, by building an experience where you "do your work inside ServiceNow."

That company announced it would open its entire set of workflows, approval chains, and business logic to external AI agents. Through a mechanism named "Action Fabric," AI agents — Claude, Microsoft Copilot, or ones a customer built themselves — can now drive ServiceNow's business systems directly, over MCP (the Model Context Protocol, a common standard for connecting AI to outside services).

I don't think this is a feature addition. It's the moment a giant SaaS company let go, on its own, of the strategy of "get the user to come to our dashboard." When McDermott said "AI agent of agents," it was a declaration that the company would become an execution layer called by any AI agent — not its own UI.

Around the same time, a similar move was happening in a different industry.

On May 5, 2026, JPMorgan Chase CEO Jamie Dimon took the stage at an Anthropic briefing for the financial industry in New York. At the same event, Anthropic CEO Dario Amodei announced the new model "Claude Opus 4.7" and a set of AI agents built for Wall Street. The customer list shared at the venue included JPMorgan, Goldman Sachs, Citi, AIG, and Visa — among the most heavily regulated financial giants in the world. The fact that these firms already run Claude in production was said out loud, by the people themselves.

The audit and consulting world followed too. On May 14, 2026, PwC, one of the four largest accounting firms in the world, announced a partnership to integrate Claude across the firm. Five days later, on May 19, KPMG announced it would integrate Claude across a workforce of about 276,000 people.

These may look like scattered, separate news items. But read along a single axis, you notice you're looking at the same phenomenon from different angles.

Until now, the royal road of SaaS was "get the user to come to our dashboard and operate inside it." Vendors polished their dashboards, competed on ease of use, and invested in retention. But since the start of 2026, that premise has begun to crack. The largest enterprise SaaS, the largest financial institutions, and the largest audit firms in the world have all started rebuilding their systems so external AI agents can call them.

The very reason SaaS exists is being rewritten right in front of us. That's how it looks to me.

Let me put the question differently. Why did an industry that, until now, treated "completing everything inside its own dashboard" as its strength change course in roughly a single year?

Looking for that answer is the purpose of this article.

A tectonic shift in user experience

Before we look for the answer, there's a moment I want to pause on. The strategic turn in the SaaS industry isn't happening only because of the industry's own circumstances. What's moving underneath it is a change in user behavior.

Think about yourself for a moment. Over the past year or so, when you want to look something up, decide something, or talk something through, where does your first move go?

Let me tell you about myself. I now run a large part of my work inside a chat with Claude. When I make a form, when I write, when I think through strategy — I open the chat first. I only log into a SaaS dashboard afterward, when I want to check something. This isn't a special way of working. Many of FORMLOVA's overseas users do the same. About 70% of FORMLOVA's users are outside Japan, and many of them create and run their forms through a chat screen like Claude, over MCP. Users in Japan often enter from the dashboard, but overseas users already take "the first point of contact is the chat" for granted.

This behavior is spreading across professions and ages. And among younger people, it's reaching into everyday life.

In May 2026, there was an event that drew attention across Japan. A professional baseball team's manager ended up resigning over trouble at home. I won't go into the details of the incident here. What I want you to notice is one shape of behavior that became visible in the course of it. The person involved, an 18-year-old daughter, when she faced difficulty at home, first turned to ChatGPT. She told the AI her situation, the AI pointed her to a public agency she could consult anonymously, and following that guidance she contacted a child consultation center.

The journalist Yukiko Kishida, commenting on the case, pointed out that turning to AI first is genuinely common among children today — that many of them ask the chat before they ask their parents.

What this observation shows isn't the unusual choice of an unusual household, I think. When people are in trouble, the first place they consult is becoming an AI chat — not family, not friends, not a search engine. That's a tectonic shift in behavior.

When a person sets out to do something, the starting point is now moving to the chat screen.

Is this a temporary boom that will swing back, or a structural change that won't? I believe it won't swing back. There are three reasons.

The first has to do with how the brain works. Once it has learned the experience of getting an instant answer, the brain strongly resists going back to a more effortful method. This is something behavioral economics has observed again and again; it's not a special claim. It's the same as how, when search engines replaced newspapers and encyclopedias, almost no one went back to paper. Asking AI for an instant answer puts far less load on your head than spending time on search. Once you've known the lower-load option, you don't go back to the higher-load one on your own. Even when people learn that AI answers sometimes contain errors, they don't return to search — they move toward "a more trustworthy AI."

The second reason is generational turnover. People in their teens now have grown up with generative AI close at hand. For them, "asking AI" isn't a learned behavior; it's closer to something natural and built in. Ten years from now, when they make up the core of society, the mainstream of SaaS users will be people for whom "starting from the chat" is obvious. This is a matter of demographics, not taste, so there's almost no room for it to swing back.

The third reason is the speed at which AI improves. Even in the daughter's case earlier, ChatGPT did the thing it was designed to do — it pointed her to a public support agency. Every time a risky response becomes a problem, AI providers respond and keep raising reliability. Meanwhile, a person's ability to "look it up and weigh it myself" naturally weakens the longer it goes unused. AI keeps getting better, and people keep letting go of that ability. This asymmetric relationship makes swinging back structurally difficult.

Behavior won't go back, I think. Users will start every kind of action from inside the chat screen, as an extension of their own mind.

Here, let me return to the SaaS industry.

The starting point of user action moved from the SaaS dashboard to the chat screen. That means the conditions for a SaaS to exist have changed. Until now, SaaS was the one being chosen. Users compared several SaaS products, judged the dashboard design and ease of use, and logged into the one they chose.

From now on, SaaS won't be the one being chosen; it will be the one being called. A user says "I want to make a form," "I want to enter an expense," "I want to issue an invoice" in their own chat screen, and an AI agent calls the right SaaS. In that moment, a SaaS that can't be called from inside the chat can't even exist at the starting point of the user's action.

SaaS has gone from being chosen to being called.

The ServiceNow turn we saw at the start, the moves by the financial giants on Wall Street, the firm-wide integrations at KPMG and PwC — I think they're all responses to this change. The SaaS industry didn't suddenly lose its mind. User behavior rewrote the conditions for SaaS to exist. So SaaS had to remake its own role.

But here a fork appears. Even saying "become the one that gets called," there are two completely different roads to getting there. The choice between these two roads is the real nature of the bifurcation happening in the SaaS industry right now.

The essence of the bifurcation — where you place control

Even while heading in the same direction of getting called, SaaS companies are now moving on two completely different design philosophies.

One is to place a powerful AI agent inside your own dashboard, and give the user the experience of "come to us, and talk to our AI." Let's tentatively call this built-in.

The other is to open your own functions through the common standard called MCP, and have them called from the chat screen the user already uses — Claude or ChatGPT. Let's call this open.

The difference between the two isn't about which features exist or how pretty things look. The essence comes down to a single point: where you place control.

Built-in keeps control inside your own house. The user comes to your site, logs in with your account, and talks to the AI inside your dashboard. The AI looks at your data, drives your functions, and everything is completed inside your worldview. The user's gaze, their actions, and their trust all gather to you.

Open hands control to the chat screen the user uses. The user keeps talking in the Claude screen if it's Claude, the ChatGPT screen if it's ChatGPT, and your functions get called within that as needed. From the vendor's side, it's a choice to give up keeping the user inside your own worldview, and instead participate as a part within the flow of the user's actions.

This difference touches the root of strategy. Built-in is the strategy of placing your own UI at the center of the work. Open is the strategy of placing not your own UI but the user's AI client at the center of the work. That's how you can put it.

What's interesting here is that the world's major SaaS companies aren't choosing only one of the two. Many of them hold both mechanisms. They hold both — but where they place the axis is clearly different.

As an example, let's look at Salesforce's moves.

Salesforce has a powerful built-in AI agent platform called "Agentforce." Deeply integrated with its CRM (customer management) data and its own workflows, it's a mechanism that supports sales inside the Salesforce dashboard. Yet in December 2025, Salesforce announced an integration that runs Agentforce Sales inside ChatGPT — a mechanism where the user can operate Salesforce data without leaving the ChatGPT screen.

Why did Salesforce, which had been polishing built-in, deliberately step out into ChatGPT? The motive an executive, Kris Billmaier, described was candid. This is the same as the past Outlook and Teams integrations: you need to go out to where the user's eyes are.

There's another background to this move. Even before Salesforce officially stepped out, third-party developers had started, on their own, building unofficial MCP servers that connected Salesforce data to ChatGPT. Leave it alone, and your data gets used by AI agents outside your control, slipping past your billing and governance. So they had no choice but to build the official exit themselves. This is less an offensive strategy than a defensive reaction to the change in user behavior.

And one more — the ServiceNow move we touched on at the start makes this structure even clearer.

ServiceNow, too, had powerful built-in AI features called "Now Assist" and "AI Agents." Then in May 2026 it announced Action Fabric and opened its entire system to external agents over MCP. Where Salesforce makes a trip out into ChatGPT, ServiceNow invites any external agent into its own business systems. The directions look opposite, but the essence is the same. They accepted the reality that "the UI the user operates is no longer our own UI."

Here, the words that freee's co-founder and CAIO Takashi Yokoji spoke in early 2026 carry weight: "SaaS has become something that is used not by people, but by AI." That single line names, in the words of someone inside the industry, the reality the giants are facing right now.

There's a caveat, though. Neither Salesforce nor ServiceNow gave up their own UI or built-in AI. They hold both. The accurate way to put it is: they hold both, and clearly shifted where they place the axis. This isn't an either/or of "built-in or open." It's a more fundamental choice: do you keep control in your own house, or hand it to the user's chat screen?

And here is the real fork in the SaaS industry.

One company keeps its own UI as the main axis and adds MCP support as a supplement. Another places "being called over MCP" as the main axis and keeps its own UI as a supplement for checking status. The former is a strategy of extending life along the line of past SaaS. The latter is a strategy of rebuilding on the premise of structural change.

The two look very similar. Both support MCP, both have a dashboard, both have AI features. On the surface you can't tell them apart. But once where you place the axis differs, the details of design decisions, the allocation of investment, how the organization is built, how marketing is done — everything starts to differ.

Which way you swing the axis depends on how you read where users will move from here.

The UI the user operates is no longer your own UI. If you stand on that premise — the one ServiceNow accepted — then the strategy of keeping your own UI as the main axis will eventually grow stale. Every SaaS is now facing the same question ServiceNow already answered for itself.

But a doubt remains. Does the strategy of opening up over MCP actually hold, technically and legally? Especially in industries that handle sensitive data — finance, medicine, audit, HR — there should be serious concerns about handing data to an external AI at all. The reason the SaaS industry had been split in two is exactly that this concern hadn't been resolved.

And yet that wall, too, is now crumbling fast.

The walls are starting to fall

Why had the SaaS industry been split in two until now?

The reason is simple. Opening up over MCP carried serious concerns. There were two big ones in particular: the data sovereignty problem, and the safety problem.

The data sovereignty problem is the question of whether you may hand sensitive data to an external AI. By the nature of MCP, the result of calling a SaaS function always passes once through the user's chat screen — that is, through an external AI vendor's processing infrastructure. Financial transaction data, patient medical information, corporate secrets, HR information. Data that was supposed to never leave leaves your control and passes through an external AI's processing. Under contracts and under law, that wasn't easily allowed.

The safety problem is that letting an external AI agent operate your systems creates a new entry point for attacks. The one that has been watched most closely is the attack called prompt injection — a trick that quietly feeds an AI malicious instructions to make it perform operations it shouldn't. OWASP, an information-security body, lists this as the number-one AI-related security risk. Handing an external AI an entrance to your system meant adding one more target for this attack.

Because of these two walls, industries that handle sensitive data were cautious about opening up over MCP. Instead, they chose built-in, completed inside their own dashboard. This is the real cause of the bifurcation.

And yet these two walls are now crumbling fast.

Let's start with the data sovereignty wall. Remember the facts from the start of this article. JPMorgan, Goldman Sachs, Citi, and Visa — the largest firms in the most heavily regulated financial industry in the world — already run Claude in production. PwC and KPMG, of the four largest accounting firms, have integrated Claude firm-wide. These are the organizations most careful about handling sensitive data. They took the step of adopting it. That fact shows that the data sovereignty concern is no longer an unscalable wall.

But this is a place I want to think through carefully. The largest financial firms using Claude in production is a demand-side story: they accepted AI as a trustworthy counterpart. Opening your own systems to the outside over MCP, on the other hand, is a supply-side design decision. The former doesn't directly guarantee the latter. Even so, the most demanding organizations starting to bring AI into their work is a clear sign that, across society, the resistance to passing data through an external AI is beginning to ease.

So how is the supply side getting over it? The key lies in building out control on the server side. Opening up over MCP doesn't mean letting everything spill out. On the server side, you tightly narrow who can access which data, require multi-step approval for certain operations, and record every call in a form you can verify afterward. The industry is starting to put this layer of control in place as a system.

There's a concrete example. In March 2026, Japan's Money Forward and freee published MCP servers in accounting — the dead center of finance. What both companies share is that they tightly limit the range an AI agent can operate to within the permissions of the logged-in user. freee states explicitly that "no operation beyond the logged-in user's permissions is performed." Rather than handing an external AI unlimited authority, they bind it on the server side so it can only move within the permission frame that user already has. This means the permission-management systems existing SaaS spent a long time building can be reused, as is, to control AI agents.

On the safety wall, too, the layer of countermeasures is maturing fast.

On May 6, 2026, Auth0, a major authentication-infrastructure company, officially launched an MCP-dedicated authentication feature. It guarantees the identity check and permission management for AI agents accessing SaaS, using an industry-standard mechanism. Similar moves are progressing at many companies, and gate-like mechanisms — standing between the AI agent and the SaaS, watching for unauthorized operations, controlling permissions, and recording everything — keep appearing one after another.

ServiceNow's Action Fabric, which we touched on at the start, is built on exactly this idea. ServiceNow governs every action by which an external agent operates its systems, by passing it through identity verification, permission narrowing, operation logging, and approval chains. It opens to the outside and, at the same time, places operations from outside fully under watch. It's a design that achieves both openness and control.

Here, one way of thinking comes into view. A design that controls the AI's movements on the server side — holding the reins, so to speak. Even when you open up over MCP, if you hold the reins firmly on the server side, the AI agent can only move within the permitted range. The output range of data and the authority over operations are all decided on the server side. The maturing of this reins design is the single biggest factor crumbling the data-sovereignty and safety walls.

The industry's understanding of how to build tools has also deepened. It used to be thought that you could just thinly wrap an existing system's functions and turn them into MCP. But it turned out that doesn't work well. You need to reorganize the functions around the user's goal, so the AI agent can handle them easily — that design idea is spreading. Leading companies like the payments service Stripe, and Block, which has run more than 60 MCP servers, have started publishing this know-how. When the reins design and the tool design click together, MCP works safely, and usably.

Read this far, and it might sound like every wall has fallen. But there's something I want to write honestly: walls that haven't fallen still exist, for sure.

For example, there are domains where, by contract or by law, data may not pass through a specific AI vendor's processing infrastructure at all. Some extremely sensitive medical data, information tied to state secrets, data under specific geographic data-sovereignty rules. No matter how firmly you hold the reins on the server side, the data after narrowing passes through an external AI's processing. When that passage itself isn't allowed, the reins design can't solve it. This is a contract and institutional problem, not a technical one. Solving it needs another layer — letting only specific AIs an organization has approved connect, or controlling even where the AI processes things — to settle as an industry standard. That will probably take years.

Still, this remaining wall isn't large enough to stop the whole flow of SaaS. The majority of the world's business data falls into the domain that, with an appropriate reins design, can be handled safely over MCP. The walls that haven't fallen are a limited corner of the whole. And that corner, too, will move as institutions catch up over time.

With the walls fallen, the reason for SaaS not to open up over MCP is being lost fast. So the next thing to ask is: what happens on the other side of opening up? Where will SaaS value be created from here?

The place where value is created is moving

When the walls fall and SaaS can open up over MCP, what happens? To put the conclusion first: the very place where SaaS value is created moves.

Until now, much of SaaS value lived in the dashboard — that is, in the UI. An easy-to-use screen, organized functions, a refined feel of operation. Users chose a SaaS by the quality of this UI, and stayed with it by getting used to the UI. So vendors put their biggest investment into the UI. They hired designers, poured effort into front-end development, repeated screen improvements, and set up support to help people learn the operations. This was the source of SaaS competitiveness.

But once users start their work from the chat screen, that premise breaks. Users no longer look closely at the SaaS dashboard. They convey their request inside the chat, and the AI agent calls the SaaS behind the scenes to process it. What appears before the user's eyes is only the chat screen they're using. The SaaS UI nearly vanishes from the user's view.

What does this change mean? The market already showed it in a violent form. In late January 2026, in the several trading days right after Anthropic released a certain mechanism, an estimated 285 billion dollars — about 40 trillion yen — of market capitalization disappeared from SaaS-related stocks. It wasn't a new AI model, nor a groundbreaking service. Just the release of a mechanism that makes it easier for AI agents to directly operate existing systems, and this much value was lost within a few days.

What did the market react to? It was a premonition of structural change: SaaS value gets pulled up into the AI-agent layer above and the data layer below, and the UI layer in the middle thins out. Value moves from the middle to both ends, and the middle gets squeezed. The industry calls this the squeeze on the middle layer.

So what specifically changes for SaaS vendors? Let me go through three changes in order.

The first is a change in where you invest. The investment that until now went into building out the UI — screen design, improving the feel of operation, support to help people learn the operations — its return falls. Of course it does: users stop looking at the UI. What rises in return instead is the reins design we saw earlier, the design of tools that are easy for AI agents to handle, and getting your data in order. Can you offer tools that the AI can call accurately, that are unlikely to cause mistakes, and that don't misread the user's intent? That becomes the new competitiveness. This isn't just a shift in emphasis — I think it's a change at the level of who is inside the SaaS company and how the organization itself is built.

The second is a change in how you charge. Until now, many SaaS charged a monthly fee based on the number of users. But once users stop logging into the dashboard, the very idea of "how many people log in and use it" stops holding. What spreads instead is charging based on how much the AI agent called functions, how much processing it ran, how much value it produced. The research firm IDC predicts that by 2028, per-seat pricing will no longer be mainstream, and that 70% of software vendors will rebuild how they charge.

The third is the most overlooked, and the most important, change. It's a change in where the brand gets recognized.

Think about it. When the user only ever touches a SaaS through the chat screen, where does that SaaS's brand get recognized? The places SaaS invested in for brand-building until now — the dashboard design, the logo exposure, the worldview of the whole screen — stop entering the user's eyes. The user may not even be aware of which SaaS they're using. Because the AI agent is just calling it behind the scenes.

So does investment in the brand lose its meaning? It doesn't lose its meaning — the place you invest moves.

There are three new brand assets, I think. One is the quality of the tools. When the AI agent calls them, do they run accurately, with few mistakes, accurately grasping the user's intent? AI agents avoid hard-to-use tools and choose easy-to-use ones. The quality of the tools themselves becomes the reason to be chosen. The second is the words you offer to developers. Do you have words to tell what design philosophy this SaaS was built on? The third is the broadcasting of thought. In the industry-wide conversation, are you broadcasting what stance you take and what future you picture?

In a world where users don't look at the UI, SaaS can't speak for itself through the UI. Instead, it shows who it is through the quality of its tools and the words that tell its thinking.

There's an irony here. The place where value is created has already begun to move, yet many SaaS vendors keep investing in the old place. A prettier dashboard, a more elaborate feel of operation, more attentive support on the UI. Until a few years ago these were the source of competitiveness, but from here they become investments with thin returns. Only the vendors who could change the direction of their investment survive where value has moved. That's how I see it.

So when, and to whom, does this change arrive? Here we need to think about Japan's market and its particular circumstances.

When does it come, and the matter of Japan's market

The place where value is created moves. SaaS goes from being chosen to being called. This structural change has already begun. The question is when, and how far, it spreads.

From here on it's a matter of prediction. Predictions can miss, so I'll avoid declarations. Instead, I'll share a few signposts for judging for yourself how far the change has come.

The first signpost is whether ordinary users can use external services from the chat screen without being aware of the word MCP. Right now, using MCP still takes the trouble of setting up the connection yourself. When that trouble disappears and users reach a state where they use it without being aware of anything, that's a big signpost.

The second signpost is what share of business-facing SaaS in Japan has become MCP-ready. Right now, Money Forward and freee in the accounting field are ahead, but across the whole industry, the companies that support it are still a handful. Once this passes 30%, you can judge that the flow has begun in earnest.

The third signpost is how far the AI from giant IT companies — Google's Gemini, Microsoft's Copilot — has made connecting to external services a standard feature. Once these can use MCP by default, the number of users jumps at once.

The moment several of these signposts line up becomes the turning point where the change advances all at once. Whether that turning point is half a year away or two years away, I honestly don't know. But that it's a not-too-distant future, I hold as a sense from the field.

The basis for thinking so lies in a small service I run, called FORMLOVA.

FORMLOVA is a service for making and running forms from a chat. When I put it out on Product Hunt — a place that introduces new services to the world — I couldn't win the badge that draws attention. But from users who actually touched it, spontaneous appreciation spread. In particular, word of mouth started in Indie Hackers, a community where developers gather, and it rose to be displayed at the top there. Overseas tech articles picked it up, too, and the inflow from there continues.

What I want you to notice here is that the origin of the word of mouth wasn't product promotion. What I did was write and publish a number of articles explaining how to use MCP. I didn't sell the product — I shared a way of thinking. And developers responded to it. The earlier point — that a new brand asset is the broadcasting of thought — is something I experienced myself, on a small scale.

And there's one more important fact. About 70% of FORMLOVA's users are overseas. The main countries are the US, Spain, and European countries — places where AI use is advanced. Use from Japan is about 30%. Compared by number of site visitors, there's about a tenfold gap between overseas and Japan. The user ratio and the visitor ratio look different because the way visits settle into actual use differs by country, but by either number, it's clear that overseas is far ahead.

This imbalance can't be explained by the amount of effort I put into marketing alone. I broadcast a fair amount in Japanese too — on my blog, on X, on Note. And still the gap appears. It's natural to think of this as a difference in the market's readiness. Overseas, especially in AI-advanced countries, a certain number of people who are used to using external services over MCP already exist. In Japan, that layer is still thin.

Let me add this so there's no misunderstanding. The reason there are fewer users in Japan isn't that the technology is inferior, nor that there's no need. From what I see, this is a difference in the thickness of the layer that tries new things first, and a difference in the culture of adoption. Japan has a tendency to value going along with others — if the person next to you starts using it, you use it too. Put the other way, the first few people don't easily start moving. In the US and parts of Europe, there's a constant number of people who'll jump in if they think it's good, even when the person next to them isn't using it. So the fire catches first.

MCP right now asks three active moves of the user: to know, to choose, to connect. In a market with a thick first-try layer, people who clear these three appear, so those who follow imitate them. In a market with a thin first-try layer, the first roll doesn't happen, so those who follow don't move either. It's the difference of whether that first small clump forms.

This fact holds an important suggestion for SaaS developers in Japan, I think. Many people judge that MCP is still too early. The basis is usually their own felt sense of the Japanese market around them. But that felt sense is biased. The fact that MCP isn't yet popular in the Japanese market doesn't reflect the quality of the technology called MCP; it merely reflects Japan's particular way of adopting things. Look globally, and a layer that's ahead clearly exists. FORMLOVA's number of 70% overseas tells that story.

So will Japan's SaaS be left behind? I have no intention of stoking a sense of crisis. Because Japan's market has another face.

Japan's market is slow to get a new technology off the ground, but once the fire catches, it has the property of spreading fast. The spread of smartphones, of QR-code payments, of SaaS itself — the initial moves lagged, but the spread after the turning point was actually faster than other countries. MCP will likely follow the same path.

If so, what to consider isn't the impatience of "it's too late, hurry." It's how to position yourself before the fire catches. To stand, while you still can, in the place you'll be remembered first as the most natural choice the moment the fire catches in Japan's market. That's the preparation that pays off over the next year or two. Precisely because it's a slow market, those who move first take the good positions. This isn't a crisis; it's an opportunity, I think.

What I'm doing with FORMLOVA is exactly this preparation. Acquire users overseas first, keep broadcasting the thinking, and build the state of already being there when the fire catches in Japan's market. This is a bet. But I think it's a bet with a basis.

Finally, as the one making that bet, let me tell you honestly what I thought about, and struggled with, over the past year.

Where FORMLOVA stands now

Up to here, I've talked about the big structural change in the SaaS industry from as wide a view as I could. But the view from the field is more complex. At the end, I want to tell you that complexity honestly, as a participant.

I run FORMLOVA, a service designed on the premise of opening up over MCP. You make a form from a chat, and run it afterward inside the chat too. I've kept the dashboard, but it's only a supplement for checking the situation at a glance. When you want to see how many responses a form has gathered, or how the whole thing looks, it's still handier to glance at a screen. The chat is the lead; the screen is a supporting role. I decided this master-servant relationship from the start. I write more about the background of this design in the MCP Form Service Guide and in Why Form Operations Need an MCP Layer.

In the future, I think a world will come where even this situation-checking gets shown inside the chat screen by the AI. But it's not that time yet, so I keep the screen as a supplement. Rather than throwing away one or the other, I place the axis clearly on the chat side. This is the design I chose.

Building with this design, there's something I struggled with more than I expected. It's the wall of awareness.

Honestly, this is the hardest part. When someone who wants to make a form searches, what comes up at the top are existing form-creation services. Everyone is looking for information centered on the act of making a form. Meanwhile, almost no one is searching with the pinpoint words for the idea of running forms using MCP. People search starting from what they're troubled by. The very idea of running forms over MCP hasn't yet become the words for many people's troubles.

So FORMLOVA often gets mistaken for an ordinary form-creation service. This frustrates me, because FORMLOVA isn't a service just for making forms. Starting from an application or an inquiry, it builds out the flow of work that follows, inside the chat. I think of this as something like a concierge for the web. You enter through the doorway of a form, and it clears away the work beyond it while talking with you. That's the position.

That said, I have no intention of fighting head-on with the existing tools that automate workflows. Rather than fight them, I bundle them well and deliver a higher experience. That's the worldview I hold.

But getting this position — "not a form-creation service, and not a workflow-automation tool" — understood correctly is genuinely hard. The work of getting a new category recognized takes far longer than I imagined. There's no fast-acting move. Patiently, I write articles from many angles and keep broadcasting across outlets. I'd also like to aim for spreading at once on social media, but here I hit a wall. FORMLOVA's goodness doesn't come across just by looking.

This is another point I'm struggling with. You actually connect it to a chat and use it, and only then feel "this is different." But before using it, even showing screenshots, the value is hard to convey. In fact, from people who have used it, I get good responses. The problem is the hurdle of that very first step.

On top of that, the more someone is used to a dashboard, the more they feel discomfort at first. For a person who has always done the work of making and running a form while looking at a screen, doing it inside a chat feels different. The experience of making while looking, and the experience of confirming while talking, are very different. This is partly a matter of getting used to it, but it's friction that always arises in a transition period.

Written out like this, it might sound like nothing but hardship. But I think these hardships are necessary ones.

As we've seen so far, the flow of opening up over MCP is advancing for certain. The largest SaaS in the world turned the wheel, the financial giants moved, the two accounting leaders supported it. The direction of the flow won't change anymore. But in the transition period before that flow reaches all of society, friction like this always arises. The awareness of a new category. The move from a familiar experience. Explaining value that doesn't come across just by looking. Unless someone crosses these walls first, those who follow can't cross either.

Those who move first take on this friction. In return for taking it on, they get to stand in the place they'll be remembered first, as the most natural choice, the moment the fire catches in the market. What I'm doing now is taking that on.

Where FORMLOVA stands now is still on the way. I'm stopped in front of the wall of awareness, getting mistaken, struggling to explain. But this standing point isn't mine alone, I think. Because it's also a miniature of the wall that anyone trying to build SaaS in the age of MCP has to cross.

SaaS has gone from being chosen to being called. To be called means to vanish from the user's view. To keep being chosen even after vanishing from view — what do you invest in, what do you speak, what thinking do you hold? To that question, I'm trying to give my own answer with a small service called FORMLOVA.

It isn't only the SaaS industry that's shaking. The ground under my own feet, as one who builds it, is shaking just the same. To decide, within that shaking, where to plant your weight — that, I think, is what it means to build SaaS in this era.

Disclosure and Verification

This is a FORMLOVA owned-blog article. The author is the developer of FORMLOVA. The other companies' announcements and figures mentioned in the body (about ServiceNow, Anthropic, Salesforce, PwC, KPMG, Money Forward, freee, Auth0, IDC, OWASP, and so on) were written after checking primary and secondary sources available as of May 2026. Because reported market-capitalization swings and the contents of each company's announcements vary across reports and estimates, please confirm with each company's official announcements if you rely on them for important decisions. Pricing, features, and limits are subject to change, so please check each service's official pages for the latest information.