Open your phone and count the apps that have added "AI features" in the last year. Then count how many of them want a separate paid subscription for those features. The numbers are uncomfortable when you line them up.
A meal-planning app I tried last year wanted $9.99/month for AI suggestions. A note-taking app wanted $10/month for AI summarisation. A photo-editing app wanted $8/month for AI background removal. A productivity tool wanted $20/month for an AI assistant. I already pay $20/month for ChatGPT Plus and $20/month for Claude Pro, both of which would happily do every one of those things if they could just see my data.
That last clause — "if they could just see my data" — is the whole story. For years, the only way an app could connect AI to your stuff was to bake the AI into the app itself, which meant the AI was captive to that one app and you paid the markup. Then MCP happened and that constraint went away. The apps that haven't noticed yet are about to look very expensive.
The double-billing problem
Here's the maths. A typical "AI-powered" consumer app bundles three things into its subscription:
- The actual app (the UI, the data store, the workflow)
- API calls to OpenAI or Anthropic, marked up with margin
- The app company's own profit on top
You're already paying #2 separately if you have a Claude or ChatGPT subscription. You're paying for it againthrough the app. And the app is paying retail token rates that are higher than what you get with your consumer subscription. This is the "double-billing" problem in one paragraph.
You already pay for the brain. Why should every app you use also charge you to rent it?
Some apps acknowledge this with a "bring your own API key" option, where you paste in an OpenAI key and pay token costs directly. That helps, but it's clunky — you're managing keys, watching usage dashboards, and the app still owns the conversation experience. Most users won't do it. It's a workaround, not a solution.
BYOAI is the actual solution
Bring Your Own AI is a simple inversion: instead of the app embedding an AI inside itself, the app exposes its data and tools through MCP, and you connect it to whatever AI you already use.
This sounds like a small architectural difference and is actually a massive economic one:
- You pay one AI bill, not many.Whatever you spend on Claude or ChatGPT covers everything the AI does, across every BYOAI app you connect it to. There's no second invoice.
- The app stops being an AI company.It's a kitchen app, or a notes app, or a calendar app. It doesn't need to manage token costs, prompt engineering, model upgrades, or AI customer support. That's your AI provider's problem.
- You keep your AI relationship. If you switch from ChatGPT to Claude tomorrow, every BYOAI app you use comes with you. No data migration, no learning a new AI personality.
- Apps get cheaper, often free.Without a token-cost line item to cover, BYOAI apps don't need a $10/month subscription. Surprise Chef is free. So is most of the public MCP server registry. The economics genuinely change.
Who's losing money under BYOAI?
Three groups, in roughly this order:
- Apps whose main differentiator was "we have AI". If your only moat was that you wired up a chat box to GPT-4, BYOAI eliminates your moat. Anyone's ChatGPT can do what you do now, assuming the data is reachable.
- Bundlers who marked up token costs. If you were buying tokens at wholesale and selling them at retail, BYOAI removes your margin entirely.
- Apps that hoarded user data to feed their own AI. The old playbook was: lock up the user's data, train models on it, sell those models back. BYOAI lets users extract their data through their own AI, which is a problem if your business model assumed they couldn't.
Most existing "AI-powered" consumer apps fall into one of those three categories. Which is why most of them won't move to BYOAI voluntarily. They'll wait until users start asking why they can't connect their existing Claude account, and then they'll ship an MCP server in six months as a defensive move.
Who's winning?
Three groups:
- Users. One AI bill, every app smarter, no data lock-in.
- AI providers. Anthropic and OpenAI both make more money when their assistants are more useful. Every new MCP app that ships is a free upgrade to Claude and ChatGPT, paid for by someone else.
- Small app builders.If you can ship a focused MCP server, you don't need a $50k OpenAI bill or a venture round to launch a useful product. Surprise Chefis built by one person in regional NSW for his own family. The economics simply weren't possible before.
The honest counterargument
There are two real ones, and they're worth taking seriously:
1. BYOAI assumes you have an AI subscription.Most BYOAI apps target people who already pay for Claude or ChatGPT — that audience is in the tens of millions and growing fast, but it's not everybody. If you don't have an AI subscription, a bundled-AI app is sometimes the only way in. Free tiers help (Claude and ChatGPT both have generous ones, plenty for daily kitchen use), but it's a legitimate gap.
2. The UX of bouncing between an app and a chatbox is sometimes worse than just opening the app.True for some workflows. The answer here is widgets — embedded UI that renders inside the AI client, so you don't have to leave the conversation to see the result. MCP supports this, Surprise Chef uses it for recipe cards and pantry audits, and the experience is genuinely better than either pure-chat or pure-app. But it's newer and not every MCP app has caught up yet.
Neither of those is a reason to keep paying double. They're reasons the BYOAI ecosystem will keep getting better.
What to do
If you're a user: next time an app asks you for a separate AI subscription, ask whether they have an MCP server first. If they do, connect it to your existing AI and skip the upsell. If they don't, you now have a much better question to ask their support: why not?
If you're an app builder: the apps that ship MCP servers in 2026 are going to look like the apps that shipped mobile apps in 2009. The early movers get a big head start. Anthropic publishes the spec and SDKs, OpenAI now supports MCP-compatible tools through ChatGPT Apps, and the work to ship a basic server is genuinely smaller than building a chatbot from scratch.
And if you want to see what BYOAI feels like in practice — connect Surprise Chefto your Claude or ChatGPT account, ask it to plan your week, and notice how it never once asks you to swipe a credit card for the AI part. That's the model.