The Best AI in Your App is Invisible

Right Now, Using AI Means Making a Special Trip

I’ll be honest about how I used AI today. I stopped what I was doing. I opened a tab, typed my request into a box, waited, copied the answer, and carried it back to wherever my actual work lives. Then I closed the tab and got on with my day.

And yes, I can already hear a few of you reaching for the reply button: “Why didn’t you wire it up with an MCP server so the model could reach your tools directly?” Fair, and I do. I wire tools up through MCP, and I even build skills to handle the repetitive parts. But that is exactly the point: I am still sitting in a separate AI surface, pointing it at my work from the outside. The plumbing gets better. The starting move is still “go to the AI first.”

That is the dominant way most people use AI in 2026, and there is nothing wrong with admitting it. Hundreds of millions of us open a dedicated AI app every day. It is, right now, the entire center of gravity of the industry. So let me clarify, because the title of this piece might suggest otherwise: I am not claiming nobody uses AI as a destination. They obviously do. I did, to write this article.

But look closely at what just happened. I left my work to go get the AI, and then I carried the AI back to my work. The intelligence lives in one place and the actual job lives in another, and the human becomes the courier shuttling between them. That round trip is friction. It is a tax we pay on every single task, and we have gotten so used to paying it that we barely notice the toll booth anymore.

Here is something to consider. People don’t want AI because it is AI. They want what it does, the capability it gives them, and they want it delivered in the places they already work. The writer wants the sentence finished inside the document. The designer wants the layout fixed on the canvas. The product manager wants the status update written inside the tool where the project already lives. The “go somewhere else to get intelligence” step was never the product. It is scaffolding. And scaffolding is the kind of thing you take down once the building can stand on its own.

We Stopped Saying “It’s an Internet App.” We’ll Stop Saying “It’s an AI App” Too.

We have seen this before. Every general purpose technology arrives wearing a name tag, gets treated as the headline feature, and then becomes so common that we quietly retire the name tag. The technology does not disappear. The label does. In the 90s, it was “.com” bolted onto every company name to make it sound like the future. In 2025 and beyond, it is “AI” bolted onto every product for the exact same reason. The suffix changed. The reflex did not.

Start with electricity. There was a stretch when “electric” was the selling point itself. You bought an electric toaster, an electric fan, an electric washing machine, and the word “electric” was doing the marketing. A 1917 Sears advertisement urged customers to “use your electricity for more than light,” because plugging things in was still novel enough to sell.¹ Nobody writes that ad today, and you do not shop for an “electric” blender. Not because appliances stopped using electricity, but because all of them do. The word stopped adding information.

Then the internet did the same thing, and this is the one every builder remembers. In the late 1990s, “internet-powered” and “web-enabled” were features you printed on the box. Being online was the pitch. Then the internet won so completely that almost every app became an internet app, and the moment that happened, we stopped saying it. Nobody announces “it’s an internet app” anymore, because that now describes basically everything. The label did not become false. It became pointless. The internet did not disappear into irrelevance. It disappeared into ubiquity.

Here is the pattern, stripped down:

AI is on the exact same track. Right now we say “it’s an AI app” because the label still sets a product apart. But AI is heading toward the internet’s fate, not gone, but everywhere. When AI runs underneath every tool you touch, calling something “an AI app” will describe everything, which is another way of saying it will describe nothing. We are not going to stop building with AI. We are going to stop announcing it. The badge is not a sign of maturity. It is a sign of how early we still are.

The Label Was Never the Point

So if AI is not the headline, what is? The result.

When you put “AI-powered” on the marquee, you are advertising your ingredients instead of the meal. Nobody adopts a tool because of the technology inside it. They adopt it because it does something they need, faster or better or for the first time.

So to be clear, the move is not to hide your AI. Hiding implies there is something to be ashamed of, and there is not. The move is to stop making AI the headline and start making the result the headline. Be loud about what your tool can suddenly do that it could not do a year ago. Be quiet about the machinery that makes it possible, because nobody really cares about the how. They care about the what and the why. Your users will feel the difference long before they think to ask what is running underneath.

The Companies Winning Already Work This Way

You do not have to take my word for where this is going. The companies with the most to lose are already building exactly this way, and the tell is what they are not doing. None of them shipped a standalone “AI app” and asked you to move in. They poured AI into the tools you already live in.

Look at Apple. And yes, before anyone tells me “they absolutely launched an AI app,” I know. At its developer conference in June 2026, Apple did announce a standalone Siri app, a place where you can continue a conversation you started with Siri. But look at where it sits in the flow. It is not where you go to begin. You start inside the work, by asking Siri on your phone, and the app is only an optional place to pick the thread back up later if you want it. What Apple did not do is build a destination and ask everyone to move in. Instead it threaded Apple Intelligence through the software hundreds of millions of people already use every day: Messages, Mail, Calendar, Photos. Apple’s own framing is the whole argument in one sentence: “Truly helpful AI must be centered on our users’ needs, deeply integrated into the products they rely on every day.”² You are rarely sent somewhere else to “use the AI.” It is just there, in the apps, making them a little sharper.

Look at Figma. This week, at Config 2026, Figma did the same thing on the canvas. The new AI does not live in a separate panel labeled “AI.” It shows up as design material. You describe a shader and it appears. You animate and adjust without leaving the canvas. An agent reaches into the tools your team already uses.³ A reporter covering the launch summed up the strategy plainly: Figma is “embedding AI capabilities directly into its existing design platform rather than creating separate standalone products.”⁴ The designer never leaves the canvas to go get help. The help comes to the canvas.

Look at Notion. Same instinct, different corner of the market. Notion AI, in their own words, “isn’t a separate app you have to toggle over to.” It is a built-in assistant that works on the page you are already on, pulling context from the documents and databases around it.⁵

This is not a coincidence. It is a strategy, and the sharpest analysts in the field have given it a name. Rex Woodbury calls it “invisible AI,” and his rule for builders belongs on a sticky note above your monitor:

If you’re introducing people to a new technology, you should change as little as possible about how and where they work.⁶

The goal, he argues, “isn’t creating a new way of working, but rather creating a better version of working in the ways that are familiar.”⁶ That is the entire playbook. Meet people where they already are. Make that place better. Say nothing about the engine.

Apple, Figma, Notion. A consumer operating system, a professional design tool, a knowledge workspace. Three completely different worlds, one identical move: put the AI under the surface of a tool people already trust, and let the work quietly get better.

Invisible Doesn’t Mean Hidden

There is a line in here that is easy to cross, and I want to be precise about it, because “make the AI invisible” can be heard two very different ways.

Invisible should mean the user does not have to think about the machinery. The seams are gone, the friction is gone, and the work just flows. It should not mean the user is kept in the dark about what the software is doing on their behalf. Those are opposite things. One removes effort. The other removes agency.

The difference comes down to a simple test. Invisible is “you do not have to think about it.” Hidden is “you are not allowed to know.” A good embedded AI feature passes the first and refuses the second. The intelligence stays out of the way until you want it, and the moment you ask what just happened, who changed this, can I undo it, the answer is right there.

So bury the controls, not the truth. When AI drafts something, let people see it is a draft and edit it. When AI makes a decision that carries weight, a payment, a deletion, a message sent on someone’s behalf, surface it and ask first. When AI touches private data, say so plainly. None of that requires a flashing “AI INSIDE” badge on your marketing. It requires honesty at the moments that matter, which is a sturdier kind of trust than any label ever bought.

This is also just good product sense. The tools people keep coming back to are the ones that feel calm and in control, not the ones that do mysterious things in the background and dare you to notice. Quiet without trustworthy is how you lose people. Trustworthy without quiet is the special trip we are trying to retire. The goal is both at once.

Our Job Is the Tool, Not the AI

We were never in the AI business. We are in the business of making something people need, and making it better than it was yesterday. AI is the most powerful raw material we have ever been handed for that job, but it is still a raw material. The question that should drive our roadmap is not “how do I build an AI product.” It is “how does AI make my product dramatically better.” Those sound almost the same. They lead to completely different places.

The first question sends us off to build a destination: one more app for people to open, one more special trip for them to make, one more “AI” badge competing in a sea of identical badges. The second question keeps us exactly where we should be, inside the tool people already use, asking what it can finally do now that it could not do before. Finish the sentence in the document. Fix the layout on the canvas. Summarize the project in the place the project already lives. Take the friction out, one task at a time, and never make anyone leave to get it.

Electricity did not win because people fell in love with electricity. It won because the toaster got warm, the lights came on, and eventually nobody had to think about it at all. The internet did not win because we admired the internet. It won because it dissolved into everything we already do. AI is going to win the same way, by disappearing into the tools that earn it. The companies building that future are not the ones shouting “AI” the loudest. They are the ones quietly making their tools so good that their users stop wondering how, and just keep working.

That is the bar. Not the most AI. The least visible AI doing the most useful work. Build that, and you will not need to put it on the box. People will feel it the moment they start, and they will never want to make the special trip again.

About the Author