Every App Wants to Sell You Its Own AI. Your Files Only Need One Agent.

Per-app AI add-ons exist because each app's data is trapped inside it. Each app sells its own AI seat because that AI is the only thing licensed to read what you put there. When your notes are open files you control, a general agent you already pay for can read them directly, and the per-app add-on becomes redundant.

Microsoft 365 Copilot launched at thirty dollars per user per month in 2023, and it is still thirty dollars per user per month today ^{1 2}. Notion bundles its AI into a Business plan at twenty dollars per member per month ³. Google retired its standalone Gemini add-ons and folded standard AI into most Workspace plans instead ^{4 5}. Three different vendors, one identical move: the AI lives where the data lives. That is not a coincidence. It is the shape of the silo.

What everyone is feeling: every tool now sells an AI seat

The pattern is real and priced. In two years the AI feature stopped being a bonus and became a line item, a per-seat charge bolted onto the tool you already paid for. The question knowledge workers now ask is not whether the AI is good. It is whether they need to buy it again, once per app.

This is a live demand signal, not a hunch. On Hacker News in June 2026, a discussion titled "Uber's $1,500/month AI limit is a useful signal for AI tool pricing" drew 624 points ⁶. The number itself comes from Bloomberg: Uber is "limiting all employees to $1,500 in monthly token spending per AI coding tool," a company spokesperson told the outlet ⁷. When an enterprise starts capping AI spend per tool, the rest of us start counting our subscriptions too.

Communities are asking the blunt version out loud. In note-app forums, a recurring question runs roughly: if I already pay for Claude or ChatGPT, do I still renew my note app's AI? Treat that as a community demand signal rather than a verified statistic. The demand itself is unmistakable. People are tired of paying for the same capability in a dozen places.

Why it happens: the AI is priced where the data is locked

Each app charges for its own AI because its data is a silo, and only the app's AI is wired to read that silo. The add-on is not a markup on intelligence. It is the toll for reaching data that cannot leave. The more thoroughly an app traps your content, the more it can charge to act on it.

Watch the pricing move and the architecture reveals itself. Microsoft's Copilot price held steady from its 2023 launch — "$30 per user, per month for Microsoft 365 E3, E5, Business Standard and Business Premium customers, when broadly available" ¹ — through to the current enterprise listing ². Notion folded full AI into its paid Business tier at twenty dollars per member per month ³. Google went further, announcing that "the following add-ons are no longer available for purchase," including Gemini Business-Legacy and Gemini Enterprise-Legacy, and moving the capability inside the plan instead ⁴.

The Google blog states the logic plainly: "we'll continue to include standard access to these advanced capabilities in most Workspace Business and Enterprise plans at no additional cost" ⁵. Read that as an architecture decision, not a discount. The AI belongs to the plan because the data belongs to the plan. Once content lives inside a closed system, the AI that reads it has to live there too — and gets billed accordingly.

The alternative architecture: open files plus one general agent

Keep your notes as open files on your own device, and a single general agent can read all of them. The data is no longer captive to any one app, so no app has to sell you a private AI to reach it. The question shifts from how many AI subscriptions to which agent, and most people already have one.

The agents already exist, and they already converge on a shared way of reaching into files. The Model Context Protocol — what one MCP explainer calls "a USB-C port for AI applications," a standardized way to connect AI to external systems ⁸ — has crossed real scale. Anthropic reported "more than 10,000 active public MCP servers" and "97M+ monthly SDK downloads across Python and TypeScript" when it donated the protocol to a new foundation in December 2025 ^{9 10}. One agent reading many sources is no longer hypothetical plumbing. (For how that connection actually works, see how agents reach into your files.)

Plain files are the most directly readable target of all. A folder of Markdown notes needs no connector, no permission grant, no per-app billing relationship — it is just text an agent can open. That is the quiet economic consequence of an old idea: when your content is a file rather than a database row, the AI that reads it is a choice you make once, not a tax you pay per tool. (On why plain Markdown is the format agents read, see why your notes are already AI-ready.)

The honest balance: what the add-on actually buys

In-app AI genuinely wins on deep, vendor-specific integration, and pretending otherwise would be dishonest. The add-on sees the app's live structure, its permissions, its relations between records, its real-time state. An external agent reading a folder of files simply does not have that context. The per-seat price is the price of that depth, and that depth is not nothing.

So the claim here is narrow and worth stating precisely: the add-on is redundant for your own open data, not worthless in general. If your work lives inside a richly structured workspace — linked databases, shared permissions, automations — the app's own AI will understand that structure better than any outside agent. The toll buys a real road. The honest question is how much of your content needs that road, and how much is just notes that would read fine as files.

For most personal note-taking, the answer leans toward files. A second brain is mostly prose, links, and lists: content that survives perfectly well outside any one app and reads cleanly to any agent. The deeper the integration you actually use, the more an in-app seat earns its keep. The simpler your notes, the more an agent you already pay for covers the same ground without a second bill.

What to do tomorrow

The move is not to cancel everything in a fit of frugality. It is to notice which of your AI subscriptions exist only because your data cannot leave the app that holds it, then decide each one on purpose rather than by inertia. Three concrete steps make that audit fast.

1. List your AI seats. Write down every tool charging you separately for AI. Most people are surprised by the count.
2. Ask the silo question for each. Could a general agent you already pay for do this job if the content were a plain file? If yes, that seat is a silo toll, not a capability you lack.
3. Move the loose content out. Keep the notes that do not need deep integration as open Markdown files you control. The ones that genuinely rely on an app's live structure can stay there; pay for that depth knowingly, not by default.

This is downstream of two older choices: keeping your notes as plain text and keeping them on your own device. Get those right and the AI question answers itself. (See the plain-text love letter for why the format outlives the tool.)

Frequently Asked Questions

Do I need my note app's AI if I already use ChatGPT or Claude?

If your notes are open files, a general agent you already use can read them directly, so the app's bundled AI is redundant for that data. The exception is deep, vendor-specific integration: in-app AI sees the app's live structure and permissions that an outside agent reading a folder does not. For plain notes, one agent is enough.

Why does every app charge extra for AI now?

Each app's data is locked inside it, so each app must ship and price its own AI to act on that content. The add-on is the price of reaching trapped data, not a markup on intelligence itself. Microsoft, Notion, and Google all moved the same way: the AI is billed where the data lives ^{1 3 4}.

How much is Microsoft 365 Copilot per user?

Microsoft 365 Copilot launched in 2023 at thirty dollars per user per month for eligible Microsoft 365 plans, and the enterprise price is still listed at thirty dollars per user per month today ^{1 2}. The stability of that number over two years is itself the point: the per-seat AI add-on has settled in as a permanent line item.

Is Notion AI included in the Business plan?

Yes. Notion's Business plan is twenty dollars per member per month and bundles its AI suite into that tier rather than selling it as a separate add-on ³. Folding the AI into a paid plan is the same architectural move as a standalone seat — the capability is priced where the data sits.

Is Gemini free in Google Workspace now?

Google retired its standalone Gemini add-ons — stating they are "no longer available for purchase" — and now includes standard AI capabilities in most Workspace Business and Enterprise plans "at no additional cost" ^{4 5}. Higher-usage tiers remain paid. The standalone add-on did not vanish so much as move inside the plan.

Can one AI agent read all my apps?

Increasingly, yes, for open data. A standard like the Model Context Protocol lets one agent connect to many sources — Anthropic reported more than 10,000 active public MCP servers in late 2025 ⁹. Plain files on your own device are the most directly readable of all: no connector, no permission grant, just text an agent can open.

Is paying for multiple AI subscriptions worth it?

It depends on how locked-in your data is. The more your content lives in open files, the fewer per-app AI seats you need, because a general agent you already pay for can read those files directly. The more your work relies on an app's deep internal structure, the more its in-app AI earns the separate charge.

Per-app AI add-ons are the price of the data silo. Keep your notes as open files and the AI question stops being how many subscriptions you can afford — it becomes which single agent you already trust. This builds on Steph Ango's "file over app" principle: "File over app is a philosophy: if you want to create digital artifacts that last, they must be files you can control, in formats that are easy to retrieve and read" ¹¹. Apps are ephemeral, but your files have a chance to last ¹² — and the agent that reads them is yours to choose. To keep notes as open Markdown on your own device, where any agent you choose can read them, there is mnmnote.com.