A company ships something. People pay for it. The company shuts it off. Nobody gets their money back. That's it. That's the whole story of what happened to players of Aliens: Fireteam Elite on Nintendo Switch this week. A $60 game, cloud-dependent, now gone. A reminder that "purchased" and "owned" are not the same word when someone else controls the infrastructure.
We've been watching this dynamic for 25 years. And what strikes us isn't the gaming angle. It's that the same pattern is playing out across every industry, all at once, in ways that most founders aren't tracking.
Here's the thread: the closer you get to a capability you didn't build and can't control, the more fragile your business becomes. Doesn't matter if it's a cloud game server, an AI model, a shopping platform, or the language a tool was trained on. The dependency is the vulnerability. And right now, everyone is adding dependencies as fast as they can and calling it "moving fast."
The AI Dependency Nobody Is Measuring
The GitHub Copilot SDK for Java is genuinely interesting. Enterprise Java developers can now drive AI assistance directly from idiomatic Java code, using annotations and virtual threads, embedded right in the tools they already work with. Powerful. Real. We like it.
But here's what nobody says out loud: the moment you build your workflow around an SDK owned by a single vendor, you've signed a lease you can't read. The terms will change. The pricing will change. The model underneath it will change, quietly, and your output will shift in ways you won't notice until a client does. This isn't a knock on GitHub. It's a structural observation about any capability you can't swap out in an afternoon.
And the problem goes deeper than vendor risk. New research measuring what's called the Cross-Lingual Comprehension Gap shows that the same AI model, given the same question in a different language, often performs significantly worse than it does in English. The capability you thought you had isn't consistent. It shifts based on variables you didn't know were variables. If your product serves a non-English-speaking market, or if you work with multilingual data, you may be sitting on a comprehension gap you've never measured because nobody told you to look.
This is what dependency blindness looks like in practice. You adopt the tool. You trust the output. You build on top of it. And the floor quietly isn't where you thought it was.
The Live Shopping Lesson
Whatnot just raised $545 million, reportedly the largest single raise in live shopping history, with revenues reportedly on course for more than a billion dollars in 2026. That's a real business, not a PowerPoint. And the core insight behind it is ancient: people buy from people they can see, in real time, when they trust the seller and feel the urgency.
But look at what Whatnot actually built. They didn't just build an app. They built the room. They own the platform. Every seller on Whatnot is a tenant. They're building audience, reputation and revenue on someone else's land. If Whatnot changes the fee structure, changes the algorithm, or gets acqui-hired and shut down, the seller loses everything except whatever email list they thought to collect on the side.
This isn't an argument against selling on Whatnot. It's an argument for knowing exactly which parts of your business you own and which parts you're renting. Most founders we talk to can't answer that question cleanly. They say "our audience" when they mean "our Instagram following." They say "our platform" when they mean "our Shopify store." They say "our AI capabilities" when they mean "our API key."
The Hike Nobody Did Until Someone Did
Here's the angle nobody expects: Bradford Bartels just completed the first solo thru-hike of the proposed 500-mile Alaska Long Trail, Seward to Fairbanks, through terrain that doesn't care about your plan. No cloud backup. No refund. No vendor support. Just conditions you read right or you don't.
What makes a route like that survivable is the same thing that makes a business anti-fragile: you have to understand what you're actually carrying, what you can swap out, and what happens when something fails in a place where you can't call for help. Bartels didn't survive 500 miles of Alaskan wilderness by assuming the terrain would stay cooperative. He built a model that accounted for the terrain being hostile.
Most founders build assuming the terrain stays cooperative. They build assuming the API stays up, the platform keeps converting, the model keeps performing, the vendor keeps the terms the same. They optimize for the case where nothing changes. That's not a business. That's a bet.
What We Actually Tell Clients
When we audit a product, one of the first things we map is the dependency chain. Not just the technical dependencies, but the business ones. Which parts of this operation stop working if a single vendor changes one thing? Which capabilities live entirely inside a tool you don't own? Which audience segments only exist inside a platform that could decide tomorrow you're in violation of community guidelines?
The founders who weather the shakeouts, and there will always be shakeouts, are the ones who built the actual relationship, not just the distribution channel. They collected the email. They built the SDK wrapper they could swap. They ran the multilingual test before the product shipped to a Spanish-speaking market. They own the data, the customer record, the contract, the relationship.
The others get the Aliens: Fireteam Elite treatment. Not because they were stupid. Because they assumed "purchased" meant "owned" when it never did.
Three Questions to Run Right Now
- If your primary AI vendor doubled its prices next quarter, what breaks and how fast?
- If your top distribution platform banned your account tomorrow, what customer relationships would survive?
- If the tool you're building on top of changed its core behavior without announcing it, how long before you'd notice?
If any of those answers scared you, that's the thing to work on. Not the next feature. Not the next integration. The thing you actually own.
We build the parts of products that belong to the people paying for them. That's the whole job. If you're not sure which parts of your business are really yours, that's a conversation worth having before someone else makes the decision for you.