Two stories landed within a day of each other, and together they answer a question small business owners keep asking.
Anthropic reported preliminary revenue above $11.5 billion for the quarter along with its first operating profit, a genuinely rare thing in frontier AI. The next day, Stripe finalized a deal to buy OpenRouter for more than $7 billion. OpenRouter's entire product is a single access point to more than 400 AI models that routes each job to the one that does it best and cheapest.
One company proved an AI lab can be a real business. The other proved that not being tied to any single lab is worth billions.
The question owners keep asking is "which one should I bet on." The better question is "how do I build so I do not have to."
The bet nobody should be making
Picking the eventual winner of the AI industry is not a small business's job, and it is not a solvable problem. The labs leapfrog each other every few months. Prices fall in steps. Products get retired. Terms change.
None of that has to reach you. It only reaches you when your business is welded to one vendor in ways that are expensive to undo.
The three places lock-in actually hides
It is almost never a contract. It is these:
Your data lives only in their system. Customer records, conversation history, lead notes, call logs. If the only copy is inside a vendor's dashboard and the export is a mess or does not exist, you are not a customer, you are a hostage with a monthly bill.
Your process exists only as clicks. Nobody wrote down what the automation actually does. It was configured in an interface eighteen months ago by someone who has since left. Rebuilding it elsewhere means rediscovering it first, which is why so many businesses stay on tools they have outgrown.
Your workflow is hardwired to one model. The system does not just use a model, it depends on one specific model's quirks. When that model changes, gets more expensive, or gets deprecated, the whole thing needs redoing rather than repointing.
Each of these is cheap to prevent at the start and expensive to fix later. That asymmetry is the whole reason to care now.
What to do instead, in order
1. Get your data out on a schedule. Not once, on a schedule. Contacts, conversations, bookings, and notes exported in a format something else could read. If a tool makes this hard, that is your answer about the tool.
2. Write the process down outside the tool. A page describing what triggers the automation, what it does, what it sends, and where it hands off to a person. This is the artifact that makes switching an afternoon instead of a project.
3. Prefer systems that route. Ask any AI vendor a direct question: can you change which model runs this job without rebuilding my workflow? Systems that route give you every industry price cut automatically. Systems that cannot leave you paying yesterday's prices for a task that has since gotten cheaper. Workhorse model prices have been falling in steps all year, and that trend has one direction.
4. Keep the customer-facing edge yours. Your website, your phone number, your booking flow, and your customer list should belong to you and live somewhere you control. Tools plug into that spine. When a tool changes, you swap the tool, not the spine.
5. Put the limits in the platform, not the prompt. Whatever a system is allowed to touch should be enforced by real permissions, not by instructions asking it politely. Anything irreversible, money moving, contracts, deletions, stays behind a human approval.
Vendor durability is still a fair question
None of this means the vendor's health is irrelevant. A provider that can grow fast and still turn an operating profit has less pressure to raise prices abruptly or kill a product line, and that is a legitimate input into a buying decision, especially for anything touching customer communication.
Just keep it as one input among several, and keep the caution attached: one preliminary quarter is a signal, not a guarantee. Buy the best tool for the job today, and make sure leaving it later is boring.
The version of this that actually pays
Do this and something useful happens. The AI industry's drama, the funding rounds, the price wars, the acquisitions, stops being something you have to track and starts being something that quietly works in your favor. Prices fall and your bill follows. A better model ships and you point at it. A vendor stumbles and you move.
That is what a durable setup buys: the ability to benefit from every change without being exposed to any of them.
If you want a straight look at where your current tools sit on those three lock-in points, and which one to fix first, we will walk through it with you. More on how we build automations to be portable is on our operations automation page, and if you are earlier than that, our guide to where small businesses should actually start with AI is the better first read.