The Day I Couldn't Buy My ¥168 Hydrogen-Roasted Iced Coffee: "Expectation Lock-In" in the Age of AI

Hello, this is Koji, CEO of Flagship. A while back, I wrote a column here called "Learning from Hydrogen-Roasted Coffee: Innovation, Survival Strategies in the AI Era, and Redefining Value." A convenience store coffee for around ¥168 that delivers satisfaction approaching a specialty café. How does "inexpensive yet more than good enough" quality change the value of services that used to command a premium? And isn't the same thing happening in software development because of AI? That was the question behind that piece.

Recently, though, that same hydrogen-roasted coffee taught me a second lesson. This time, the iced version I drank every day disappeared from the shelves.

The Morning the Hydrogen-Roasted Iced Coffee Disappeared

Almost every morning, I bought Seven Café's "Hydrogen-Roasted Coffee, Iced (R)." At ¥168.48 including tax, it had very little bitterness and a clean, clear finish that I had grown to love. It had become part of my daily routine. Seven-Eleven itself describes this iced coffee as having "a clean taste with little harshness."^[1]

Then one day, it vanished from my usual store. At first I assumed it was simply sold out. But it wasn't there the next day either. Nor at another store. After several days of searching, I finally lost patience and asked a staff member. The answer: "This year's supply of beans has run out, and we won't be getting more for a while." That was a real blow.

To be precise, hydrogen-roasted coffee itself hasn't disappeared. The hot version is still on sale. In fact, as of this writing, Seven-Eleven's official menu still lists both the hot (R) at ¥160.92 and the iced (R) at ¥168.48, while also noting that some stores may not carry them and that they may be unavailable due to circumstances such as raw material supply.^[2]

So this is not a story about hydrogen-roasted coffee vanishing nationwide. It's a story about the iced coffee I drank every morning becoming unavailable in my own daily orbit.

"It's the same hydrogen-roasted coffee. Just drink the hot one." Fair enough, perhaps. But for me, that wasn't the point. The cold cup I had every morning was gone, and it hurt more than I expected.

Of course, there was another option: Seven Café's regular iced coffee, at ¥140.40 for the R size.^[2] It's perfectly good. Or, a short walk away, there are cafés serving coffee for around ¥500. Simplified, the choice looked like this:

• Go back to the ¥140 regular iced coffee.
• Go to a café and pay around ¥500.

On price alone, the answer is obvious. The ¥168 product is gone, so pick the ¥140 one. Yet what my hand actually reached for was the ¥500 option. And that puzzled me a little.

The regular iced coffee hadn't suddenly gotten worse. Before, I would have drunk it without a single complaint. But once you have experienced the hydrogen-roasted iced coffee as your everyday cup, going back inevitably feels like a step down. Five hundred yen is not cheap for a daily coffee. Even so, as long as it was within my means, my honest feeling was: "If the alternative is lowering the quality, I'd rather pay more."

That's when it hit me. What I had liked wasn't the price of ¥168. It was the quality I had been getting for ¥168.

And I'm not the only one. One of our employees got hooked on the hydrogen-roasted iced coffee just like I did. Flagship's office is in Yotsuya, Tokyo. As it happens, Seven-Eleven Japan's headquarters is also within walking distance of Yotsuya Station, in Nibancho.^[3] This employee figured, "Surely they'll have it near the head office," and went all the way to the Nibancho area to look for it. No luck there either. Disappointed, they ended up buying a coffee for around ¥500 and coming back.

Both of us had a perfectly rational ¥140 option right in front of us, and both of us drifted toward ¥500. It looks like a story about coffee lovers, but I think it points to something rather important.

Disruptive Innovation Has a "Second Act"

In the previous column, I looked at hydrogen-roasted coffee as an example of disruptive innovation. Quality that used to require a ¥600 visit to a specialty café suddenly became available in a ¥168 convenience store cup. Convenience stores already win overwhelmingly on accessibility; when technology adds quality on top of that, the products and services caught in the middle face a very hard time. That was the market disruption through "good enough" quality that I wrote about last time.

This time, having become a user of that product myself and then having lost it, I realized that disruptive innovation has a "second act." In the first act, value that used to be expensive is delivered at a dramatically lower price. In the second act, that experience rewrites the consumer's own sense of "normal."

This is the crucial part. Innovation doesn't only change market prices. It moves the baseline inside us for what level of quality counts as ordinary. Research in behavioral economics and consumer behavior shows that people don't evaluate price and quality in absolute terms; they judge them against a "reference point" formed by past experience and expectations. Some studies find that a drop below the reference point, in either price or quality, is experienced as a loss.^[4][5]

Before I knew the hydrogen-roasted iced coffee, the ¥140 iced coffee was a perfectly satisfying product for me. But drinking the ¥168 version every day updated my "normal." And the ¥168 product disappearing did nothing to reset my expectations. The result: I started paying ¥500 to maintain that level of quality.

It's a curious phenomenon. Having experienced the price disruption of a ¥168 product, my spending after that product disappeared ended up higher than before. And this structure closely resembles what we are all experiencing with AI right now.

Today's AI Is Extraordinarily Underpriced

Consider the value that today's generative AI delivers, and it's clear we are living in an unusual moment. Writing. Translating. Researching mountains of material. Writing code. Drafting designs. Analyzing data. Bouncing ideas around. Work that once required hiring a specialist or hours of human effort can increasingly be done at a very low marginal cost.

Stanford HAI's 2026 AI Index reports that the estimated consumer surplus from generative AI grew 54% in a single year, reaching the equivalent of $172 billion annually in the United States by early 2026. Meanwhile, many generative AI tools are offered for free or close to it.^[6] From the user's perspective, the price feels extraordinarily low relative to the value received.

And competition keeps intensifying. The same AI Index notes that performance gaps between leading models are narrowing, and that competition is broadening to cost, reliability, and task-specific performance.^[7] If one model raises its prices, users may move to another. Precisely because no single company can easily hold overwhelming pricing power, we currently get to use extremely capable AI at fairly low prices.

But this is not a state guaranteed to last forever. Running advanced AI requires enormous computing resources, starting with GPUs. Data centers require massive capital investment and electricity. Stanford HAI also points out that even as AI companies' revenues grow, spending on compute and infrastructure investment is climbing to record levels.^[6]

Competition pushing prices down; compute and infrastructure pushing prices up. Today's AI market has both forces at work at the same time. That's why I'm not trying to predict that "AI prices will inevitably rise." What really matters is a different question: if AI prices did rise, could we go back to the world before AI?

The "Normal" That AI Raised Won't Go Back Down

When an organization adopts AI in earnest, it doesn't simply gain one more tool. The "normal" of work itself changes. A deck that used to take three hours is done in thirty minutes. Research that took days is ready for the same day's meeting. Engineers write more code per day. PMs process more information. Designers explore more ideas.

At first there is surprise: "Look how fast this got done with AI." But keep it up for six months, a year, and the surprise fades. "This speed is normal." "Getting to this level of quality on the first pass is normal." "Doing this much with this many people is normal." And once client deadlines, headcount, and even business plans are built around that baseline, there is no easy way back.

If AI usage fees went up, the response would rarely be "Then let's stop using AI and go back to taking three times as long." Clients, executives, and employees already know the new speed.

Just as those of us who lost the hydrogen-roasted iced coffee chose to pay ¥500 rather than go back to ¥140, companies are likely to decide: "Rather than lose the productivity we've gained, we'll keep using AI even if it costs somewhat more." That, I believe, is where the real lock-in of AI lies.

"Expectation Lock-In" Is Scarier Than Vendor Lock-In

The IT world has long had a term: "vendor lock-in." Depend too heavily on a particular cloud, SaaS, or database, and you can't easily move elsewhere when prices rise or specifications change. AI naturally has the same problem. But I suspect AI has a deeper layer of lock-in beneath it. Call it "expectation lock-in."

You may be able to move from ChatGPT to Claude. From Claude to Gemini. Swapping the API over to a different model may well be technically possible. But deadlines that assume AI. Headcount that assumes AI. Costs that assume AI. Deliverable quality expected on the assumption of AI. Escaping from those is far harder than switching models.

In other words, before we get locked into any particular AI vendor, we are already being locked into a world that presupposes AI. People do not easily go back from a level of quality and productivity they have come to know.

This is not an argument to stop adopting AI. Quite the opposite. I believe we should use AI thoroughly, with a clear understanding of this irreversibility.

What Stripe's Acquisition of OpenRouter Means

Seen from this angle, a recent piece of news is highly symbolic. On August 19, 2026, Stripe announced that it had agreed to acquire OpenRouter.^[8] OpenRouter is a model marketplace and gateway that lets you access a large number of AI models through a single interface. Stripe's announcement describes it as a service that routes and optimizes enterprises' token usage across more than 400 models from over 80 providers.^[8] OpenRouter itself says it processes more than 10 trillion tokens a day and serves a community of over 10 million developers and businesses.^[9]

Neither company has officially disclosed the price, but Reuters, citing people familiar with the matter, reported it at slightly over $8 billion.^[10]

Why would Stripe, a payments infrastructure company, see such value in "routing" AI models? I think this news captures where the AI market is heading. Going forward, what matters won't only be "which model is the smartest" but increasingly "which task should go to which model, at what price."

Use an expensive, high-performance model for advanced reasoning. Use a cheap model for simple classification. Switch to another model if one is slow. Consider alternatives if prices go up. Keep choosing the best model by watching quality, price, speed, and availability.

The more AI becomes infrastructure, the more the value shifts from "picking the single strongest company" to keeping your options open.

What Companies Need in the AI Era Isn't the "Best AI" but "Options"

So how should companies prepare for this shift? I see three broad areas.

1. Design models to be interchangeable

Getting the most out of a specific AI model is one thing; building a system you can never leave is another. Evaluate each model's strengths, weaknesses, and pricing, and keep a structure that lets you switch when necessary. Whether or not you use OpenRouter itself, what matters is adopting the mindset that you can route.

Of course, using OpenRouter creates a dependency on OpenRouter. So the point is not to pick a particular service. The point is to build systems that never lose their options.

2. Keep the ability to decide for yourselves what "good quality" means

The more AI produces the deliverables, the more the human role changes. More than being able to make everything by hand, what matters is the ability to judge whether something is good or bad.

For writing, what counts as good writing? For code, which quality standards must be met? For design, what must the brand never compromise? Hand those criteria to AI as well, and even moving to a different AI becomes difficult. Using AI to raise quality is one thing; outsourcing the definition of quality to AI is another.

3. Don't let human "muscle" drop to zero

This is not a call to "skip AI and rely on humans alone." AI should be used to the fullest. But you need people who can understand what's happening when AI stops. People who can tell when AI's output is wrong. People who can switch to a different technology or approach when needed. That kind of "muscle" must never be allowed to disappear entirely from the organization.

The "Playground" Has a Second Meaning

In the previous column, I wrote that surviving the AI era requires investment in education and a "Playground." Try new technology in real work right away. Fail. Try other tools. Compare. Build up experience from there.

I've since realized this idea has a second meaning. The Playground isn't only a place for adopting new AI early. It's also a place for knowing that other ways exist.

If you've only ever touched one AI, that AI becomes your whole world. If you've handled multiple models, tried different technologies, and run small experiments of your own, you can think: "If this one goes away, we have that one."

In other words, investing in a Playground is an offensive investment in new technology and, at the same time, an investment in building the company's escape routes. I feel that meaning more strongly now than before.

Thinking About What Comes "After" Disruptive Innovation

Last time, having just tried hydrogen-roasted coffee, I asked: "If something like this shows up for ¥168, how is a ¥600 specialty café supposed to compete?" This time, with the ¥168 hydrogen-roasted iced coffee gone, I found myself looking at the same problem from the other side.

What I learned is that disruptive innovation has an "after." The first thing that happens is price disruption: expensive value becomes dramatically cheaper. But the truly big change comes afterward. Affordable, high-quality things become our "normal."

And once a level of quality has become "normal," it doesn't disappear just because the product or service does. When it's gone, rather than returning to the old quality, we may pay even more than before to protect the new standard we've acquired. AI is probably no different.

Right now, we are enjoying the "cheap, very high-quality AI" that competition has produced. That's a wonderful thing. But even as we enjoy the convenience, which technology to use, how much to pay, what counts as good quality, and whether we can move to another approach: it's essential not to hand that control to anyone else.

Perhaps the true power of disruptive innovation isn't simply making existing products cheaper. It's irreversibly rewriting what people take for granted. That, I think, is the greatest disruption of all. And with AI, our "taken for granted" is being rewritten at tremendous speed right now.

That's exactly why we shouldn't run from the change, but master it while never letting go of our options. That may be one survival strategy for the AI era.

Incidentally, the hot hydrogen-roasted coffee is still on sale. Even so, I want the iced version back. And I suspect that employee feels the same. Human expectations are a stubborn thing.

References

1. Seven-Eleven Japan, news release: "A 'Refreshingly Crisp' Iced Coffee Joins Seven Café Hydrogen-Roasted Coffee" (March 31, 2026, in Japanese)
2. Seven-Eleven, "Seven Café Product Menu" (in Japanese) — Iced Coffee (R) ¥140.40, Hydrogen-Roasted Coffee Hot (R) ¥160.92, Hydrogen-Roasted Coffee Iced (R) ¥168.48 (as confirmed at the time of writing) / Hydrogen-Roasted Coffee Iced (R) product page
3. Seven-Eleven Japan Co., Ltd., "Company Profile" (in Japanese) — Head office: 8-8 Nibancho, Chiyoda-ku, Tokyo / Seven & i Holdings, "Access" (in Japanese) — 4–6 minutes on foot from Yotsuya Station
4. Hardie, B. G. S., Johnson, E. J., & Fader, P. S. "Reference Dependence, Loss Aversion, and Brand Choice." Marketing Science, 1993. (Columbia Business School page)
5. "Loss aversion, price and quality." The Journal of Socio-Economics, 2007. — A model in which both price and quality are evaluated against reference points.
6. Stanford Institute for Human-Centered AI, "Economy | The 2026 AI Index Report." — Consumer surplus from AI use; AI companies' compute and infrastructure costs.
7. Stanford Institute for Human-Centered AI, "Technical Performance | The 2026 AI Index Report." — Convergence in performance among leading models and the broadening of competition to cost, reliability, and task-specific performance.
8. Stripe, "Stripe agrees to acquire OpenRouter to help businesses optimize token routing and usage," August 19, 2026. — Describes OpenRouter as a gateway spanning 400+ models from 80+ providers.
9. OpenRouter, "OpenRouter is Joining Stripe," August 19, 2026. — 400+ AI models, 10+ trillion tokens per day, a community of 10+ million developers and businesses.
10. Reuters, "Payments firm Stripe to buy marketplace OpenRouter in AI push," August 19, 2026. — Terms not officially disclosed; reported at slightly over $8 billion, citing people familiar with the matter.

Information and prices in this article are as of August 2026. Seven Café availability and prices may vary by store and over time.