How to Build a Business in the AI Era | When "Building It" Got Cheap, What Became Valuable?

Hello, this is Ryuta Hamamoto from TIMEWELL.

The other day, an acquaintance who wanted to start a business asked me, "I want to build an app. I should hire an engineer first, right?" Without hesitating, I answered, "I don't think you need to." I wasn't being cold. Right now, in this moment, not being able to write code is no longer a reason not to start a business. If anything, the people who begin by looking for an engineer are the ones who tend to put off the most important question of all.

Over the past few years, the fundamentals of making things have flipped. What used to take millions of yen and several months can now take shape as a working product over a single weekend, at almost no cost. And precisely because of that, something paradoxical is happening: "being able to build it" is no longer, in itself, a strength for a business. So what does become valuable? Let me give you the conclusion of this article up front. The answer is exactly one thing: the heat of the field, that is, raw conversation with the customer. Knowing in your bones what a specific person is struggling with, and how much they are willing to pay for a solution. That is the only scarce resource left that AI cannot get its hands on.

This article is written as a hands-on guide for anyone who wants to use AI to advance a new business or start a company. Whether you have been handed a new-business mandate at a large company, want to strike out on your own or start a side business from zero, or run your own shop in a regional town, I've made it into something you can act on tomorrow. If you want to follow how to build a new business more systematically, I've also prepared The Complete Guide to New Business Frameworks as a map of the whole terrain. If you first want to measure how much room your own business has for using AI, try the AI Literacy Check before you read on, and the way you use the prompts in the second half will click into place.

The Era When "Building It" Got Cheap: The Assumption Vibe Coding Broke

The numbers make the scale of this change clear. Around 2018, when I first got involved in building businesses, making a single minimum viable product, an MVP, to put out into the world cost roughly three months and about 12 million yen. You gathered a few engineers, locked down the requirements, designed and implemented, and a quarter of the year was gone before anything that worked appeared. That was normal.

Today, a validation prototype of the same scope stands up in a week, for under 150,000 yen. Roughly one-eightieth of the cost. I introduced this figure in my book, How to Build a Business in the AI Era, Walking With the Customer, as well. It isn't a magic trick. When you talk to tools like Cursor or Claude Code in plain language, saying "build me a screen like this," something that works comes back even if you can't write code. When an error appears, you paste the message in as is and say "fix it," and it gets fixed. This way of building came to be called Vibe Coding. Building by feel and by dialogue, that sense of the thing became the name itself.

A prototype in a weekend, for almost nothing. It's wonderful. But it's so wonderful that many people miss the crucial part. That the speed of building went up a hundredfold means that, if you aim wrong, the wrong thing gets finished a hundred times faster. You build something nobody wants at furious speed. Speed becomes valuable only once it's pointed in the right direction.

That is why I believe the center of gravity for investment in building a business has shifted. It used to be normal to spend four months on development and to wrap up the upstream conception in a few weeks. Now it's the reverse. Because the time to build has shrunk dramatically, you can proudly spend two months upstream deciding "for whom, and what, do I build." In fact, that is exactly where you should spend your time. In a world where anyone can build, the only thing that sets your business apart from the rest is the ability to discern what to build.

So What Became Scarce? The Heat of the Field, and the Skills Whose Value Moved

If building became a commodity, where did the value flee to? My answer, as I wrote at the start, is the heat of the field.

What Became Scarce Is the Heat of the Field

For example, what wears out the wife of a craftsman at a small town factory most, in the daily back-and-forth of orders and deliveries? That is information that exists only in that household. It's written nowhere on the internet, and it won't come out if you ask an AI. If you know that struggle in vivid detail and start building from there, then even if the machinery behind it is just a simple setup that calls a generative AI, that business becomes one of a kind. Not because of the technology, but because the source of the problem is unique.

This is the biggest opportunity for non-engineers. Someone who has stood in the elder-care field for ten years, someone who has run a shop in a regional town, someone who has kept doing gritty work in a specific industry. The firsthand information these people hold doesn't become a business on its own, but the moment they pick up the tool of Vibe Coding, it turns into a unique fuel. Knowing the field is far harder to come by than being able to write code. The order has reversed.

The Skills That Fell, and the Skills That Rose

Over these ten years, value has clearly moved. As I see it, skills split into those that fell and those that rose.

Skills whose value fell (AI can stand in)	Skills whose value rose (they exist only between people)
Writing code	Deciding what to build
Making documents look clean	Seeing who to build for
Memorizing knowledge	Designing how much to delegate to AI
Translating languages	Telling apart what someone really means and what they say to be polite
Gathering and summarizing firsthand information	Taking responsibility for the numbers you put out

The left column used to be the source of a professional's salary. Now AI handles it in seconds. This isn't something to be sad about. It means you can now pour all of your time and attention into the skills on the right. The judgment to decide who to build what for, the nose to tell whether someone saying "sounds great" means it or is just being courteous, and the resolve to own the sales forecast you put out. These skills, which arise only in the space between people, become the true measure of the businessperson from here on.

Putting into words which customer's struggle your business concept is rooted in is work where the thinking behind designing customer value with the Value Proposition Canvas helps. If you make clear, before you build, "whose pain, and which pain, am I taking on," the decisions that follow get startlingly faster.

AI-Native, or Just a Wrapper? Stalls and Trees

When I take consultations about new businesses, I often see ideas in the shape of "AI × ◯◯." AI × HR, AI × contracts, AI × recipes. The angle is interesting, yet many fall into the same pit. They've become what's known as an AI wrapper, nothing but a thin skin of prompts over a generative AI.

I often explain it with the metaphor of a stall and a tree. The company providing the foundation model is, so to speak, the ground. An AI wrapper is like a food stall set up on that ground. The location is borrowed, and no roots are put down. The moment the owner of the ground puts out a stall with the same menu, or starts handing out a similar feature officially for free, the business is over. An AI-native business, on the other hand, is a tree. It has driven roots deep into that soil: proprietary data, proprietary workflows, proprietary customer relationships. Even if the ground shakes a little, it doesn't topple easily.

You can tell which one your idea is with the following five questions. If you can say yes to three or more, you're on the tree side.

• If you removed the AI, would more than 70% of the value disappear?
• Is the customer's proprietary data the fuel for the AI?
• Has the entire workflow been rebuilt around AI as a premise?
• Is there a learning loop that gets smarter the more it's used?
• Is it something a customer could not replace just by using ChatGPT directly?

To be honest, I expect about 80% of the "AI × ◯◯" businesses out there today to quietly disappear within 24 months. This is my read, not a statistic. But watching the speed of foundation-model progress up close, it feels like a forecast that isn't far off. That's exactly why, before you start building, I want you to ask yourself, mercilessly, "If OpenAI or Anthropic shipped the same feature officially three months from now, what would be left of my business?" Only businesses with something left can put down roots. How to design that discernment is the very point we check together first, every time we take a business consultation, at WARP, our AI consulting service.

Before You Build, Go and Meet People: The Mom Test and Job Stories

Where do the roots of a tree-like business grow from? Not the whiteboard in a meeting room. From conversation with customers. That's why I recommend, more strongly than anything, going to meet people before you start building.

What helps here is the famous questioning technique from The Mom Test. The name comes from a warning: you can't learn the truth about your business idea by asking your mom. Family is kind, so they'll usually say "that's nice." Everyone wants to make the person in front of them feel good. So you must not ask about hypothetical futures. If you ask "would you use an app like this if it existed," almost everyone answers "sure, I would." Those words are worth nothing. It's the same as getting 100 likes on social media and not a single yen landing in your account.

Instead, you ask only about specific past behavior. "When was the last time you did that task?" "How many minutes did it take that time?" "What did you do afterward?" "How are you making do with the inconvenience now?" Past facts have no flattery mixed in. Listen 80% to them, 20% to yourself, and don't fear three seconds of silence. Real feelings come out after the silence. And on your way out, always ask: "Could you introduce me to two people who might have a similar problem?" Repeat this, and the conversations snowball. Talk to three people a week and that's 150 people in a year.

The voices you gather can't be used as is. You have to translate the customer's words into the job they truly want done. This is where JTBD, Jobs to Be Done, comes in. The complaint "the month-end Excel work is exhausting" is the surface. Beneath it hides the job they really want done: "what I actually want is to leave on time and have dinner with my daughter." What you should sell isn't faster Excel, it's the dinner time with the daughter. Once you can make this translation, you escape the competition on features. How to re-frame the customer's job is something I plan to cover in detail in Thinking with JTBD (the jobs customers want done), but for now, get into the habit of unraveling the customer's words across three layers: situation, motivation, and outcome.

Let me state one thing plainly. This validation must not be outsourced to AI. I occasionally hear, "I had AI play the customer and sparred with it," but judging whether a business is a hit or a miss that way is dangerous. AI is good at returning plausible answers but bad at returning unexpected ones. And business discoveries are almost all born from the unexpected. AI is superb as training wheels for widening your hypotheses. But the final call on whether you're right can only be confirmed by meeting a flesh-and-blood human.

Making AI Your Sparring Partner: Five Prompts and a Fictional Business Called "Tsugumi"

So far I've covered the reason to spend time upstream, how to spot an AI-native business, and the technique of going to meet people. From here, I'll hand you the tools to actually run it. Paste the following five prompts straight into Claude or ChatGPT and use them. Just replace the parts in 【 】 with your own business and they become a sparring partner for each step.

To make it concrete, I'll use a fictional business called "Tsugumi" as an example. Tsugumi is a service for regional home-care providers that helps aides create visit records and communicate with families. The founder is a former aide who worked in the field for ten years and can't write code. After each visit, aides type up care records on their phones in their parked car, and handle their LINE reports to families by hand too. Fifteen minutes of admin per visit, eight visits a day, about two hours. They really want to warmly tell the family "your mother smiled while drinking her tea today," but, buried in admin, it turns into a boilerplate message. This is a struggle only the family and the aide understand. That very heat becomes Tsugumi's one and only fuel.

First, designing the customer interview. Have the AI create questions that don't invite flattery and ask only about past behavior.

You are an expert in designing customer interviews for new businesses. Follow the principles of "The Mom Test" strictly. Do not ask about hypothetical futures ("if there were a ...") or "would you want to use it" at all; ask only about specific past behavior, time, and money.

I am about to interview 【target customer: e.g., regional home-care aides】 about 【assumed problem: e.g., the admin burden of creating records and contacting families after a visit】. Please create the following.

1. Ten "ask-only-about-past-behavior" questions to use in the first 30 minutes (all in the form "when / how many minutes / what did you do afterward, the last time you did ◯◯"; hypotheticals and opinions are forbidden)
2. Three ways to dig for the truth when the person says "that sounds great"
3. One sentence to get two introductions at the end
4. Five "flattery-inducing questions" I tend to ask by accident, with why each is bad and a reworded example

Output as a table. Don't hold back: if any flattery-inducing questions have slipped in, point them out without hesitation.

Next, translate the voices you gathered into the real job. This is the prompt that turns "Excel is exhausting" into "I want to have dinner with my daughter." For Tsugumi, you start to see that the product isn't "automating records" but "five minutes that delivers peace of mind to a family living apart."

You are an expert in Jobs to Be Done. Help me translate a customer's surface-level complaint into the job they truly want done.

What the customer said: 【e.g., the month-end Excel work is exhausting】
The customer's situation: 【role, position, and the scene where that task happens】

First, give me the questions I should confirm with the customer, plus example expected answers, across these three layers:
1. What were they trying to do at that moment (situation)
2. What did they want to do next once that was done (motivation)
3. What would make them feel it "went well" (outcome)

Next, three Job Stories in the form "In situation A, I want to do B, because C." Finally, in one line, what can really be sold to this customer (not a feature, but the message you should deliver). Don't retreat to surface-level answers like "faster Excel."

Once the skeleton of the business comes into view, have it diagnose, sharply, whether it's a tree or a wrapper. Run Tsugumi through this prompt and you find that proprietary data (visit records that accumulate per provider), workflow integration (records flowing into family contact, handover notes, and care-software linkage in one continuous stream), and a learning loop (aides' edits nudge the report tone closer to the provider's own) are all in play, making it a tree with yes on all five questions.

You are a product strategist well-versed in AI-native businesses. Diagnose my idea harshly.

The idea: 【whose problem, what problem, and how AI is used to solve it】

Judge the following five questions with Yes / No and a reason.
1. If you removed the AI, would more than 70% of the value disappear?
2. Is the customer's proprietary data the fuel for the AI?
3. Has the entire workflow been redesigned around AI as a premise?
4. Is there a learning loop that gets smarter the more it's used?
5. Is it something a customer could not replace by using ChatGPT directly?

On top of that, give an overall verdict (AI-native = tree / in-between / the wrapper trap = stall), and write mercilessly "if OpenAI or Anthropic officially released an equivalent feature three months from now, what would be left?" Finally, name two barriers to entry I should start building right now to escape being a wrapper, each with a concrete 90-day action. I don't need generous scoring.

If the diagnosis reveals weak points, design your barriers to entry (moats). A single thin moat is easily filled in, so plan on combining at least two. For Tsugumi, the two pillars are report text optimized to each provider's tone (the learning loop) and workflow integration.

You are an expert in designing competitive advantage (moats) for AI businesses. For my business 【overview】, score the following seven from 0 to 5, and rank them in the order I can grow them over 24 months.

1. Proprietary data  2. Workflow integration  3. A learning loop that improves with use  4. Industry expertise  5. A human-and-AI network  6. Getting ahead on distribution channels  7. Regulatory compliance and trust

On top of that, pick the two weakest / most-growable barriers, and show a concrete 90-day action for each. Next, write three scenarios where "OpenAI or Anthropic comes down into the same area," and a countermeasure for each. Finally, a 10-second answer (script) for when a customer asks "why not just use ChatGPT?" A thin moat on a single axis is fragile, so design on the premise of combining at least two axes.

Last is the instruction sheet for building a throwaway minimum mock. This is where you finally move your hands. The trick is to ask it in advance to stop you when you try to make things lavish. Tsugumi's founder took this instruction sheet to Cursor and, in a week, stood up a screen where "if you just speak for 30 seconds by voice, the care record, the report for the family, and the handover note for next time come out automatically."

You are a mentor supporting the MVP design of a non-engineer entrepreneur. I can't write code. I want to stand up a throwaway minimum mock in a week without over-building it, to validate a single hypothesis only.

Business: 【overview】
Hypothesis to validate: 【e.g., do aides really want to finish record entry with 30 seconds of voice?】

Give me the following.
1. A "what we won't build this time" list for this mock (the call on which features to cut, and why)
2. List exactly three minimum screens needed for validation, with each screen's role in one line
3. Japanese instructions you can paste straight into Cursor or Claude Code (one screen at a time)
4. A script for showing the mock to three customers that draws out behavior, not opinions

Don't propose excessive features. Just the bare minimum needed to validate the hypothesis. If I try to make it lavish, stop me.

Just running these five in order turns a passing thought into the outline of a business you can validate. Tsugumi's founder showed the mock to three working aides and asked only about past behavior: "How many minutes did last week's records take you?" After reflecting just three of their requests, two of them started paying a monthly subscription. All of this happened within three weeks of writing the first line of code.

In Closing: Three Things You Can Do Starting Tomorrow

In an era when building got cheap, value moved to conversation with the customer. I've written at length, but what you need to move starting tomorrow is actually very simple.

• Today, write down on paper five acquaintances whose stories you want to hear. Next to each, add their industry and the date you last met them.
• Tomorrow, message one of them in a single sentence, "not to sell you anything, but to learn from you," and ask for a 30-minute conversation.
• In the conversation, seal away your own idea for the first 30 minutes and ask only about specific past behavior. On your way out, always get two introductions.

After that comes translating the voices you gather into Job Stories, running the AI-native diagnosis and moat design with the five prompts, building a throwaway mock in a week, and showing it to three customers. Keep to the order. Building comes dead last.

Let me write just one honest thought at the end. Now that anyone can build, the work left for a businessperson hasn't actually changed since long ago. To understand, more seriously than they do themselves, what the person in front of you is truly struggling with. That's all it is. AI has become a partner that dramatically shortens the distance to giving that understanding a form. That's precisely why the person who doesn't get drunk on the tools, and goes to meet people, wins. In this era when the speed of building went up a hundredfold, the strongest weapon is still a 30-minute conversation over a single cup of coffee.

If you want to make AI a partner in building your business, or you want someone to walk alongside you through the launch of a new business itself, we work on this kind of customer-resolution design and validation together at WARP, our AI consulting service. When you want to talk through where to start based on your own situation, reach out from WARP's individual consultation. When you reach the stage of estimating market size, How to Calculate TAM, SAM, and SOM is worth keeping at hand, and when you reach the stage of drawing the whole business design on a single page, How to Write a Business Model Canvas is too, so you can move forward without getting lost.

References

• Ryuta Hamamoto, How to Build a Business in the AI Era, Walking With the Customer (TIMEWELL Inc.)
• Rob Fitzpatrick, The Mom Test (2013)
• Anthony W. Ulwick, Jobs to Be Done: Theory to Practice (2016)
• Hamilton Helmer, 7 Powers: The Foundations of Business Strategy (2016)