How it should meet people
Articulation Frame: build a skill
The value comes from saying what you want, or from the act of saying it. Research, writing, planning, analysis, and coaching all fit. It needs no interface of its own.
This frame keeps growing because, as models get smarter, forms, dashboards, and other screens turn out to have been ways of asking all along. A travel site's filters were a way of saying "a hotel near the venue under $200." A real estate app's search screen was a way of saying "three bedrooms, good schools, a short commute." That's exactly why Booking.com, Expedia, and Zillow were among the first companies to put their services inside ChatGPT (OpenAI). The capability survived. The screens became optional.
The test: if a user could get the full value by describing what they want and reading back an answer, you're in the Articulation Frame. Build a skill, or a plugin on top of an MCP server, and meet people in the conversation they're already having.
Presence Frame: build a traditional app
The value is something words can't deliver, however smart the model gets. A traditional app is a place you go, and it has to earn that place. I see six ways it can:
Connection. Feeling with other people: a friend's face, a creator's world, the sense that someone is there. A chatbot can describe your friends; it can't be them. That value lives in places where people show up for each other, not in a text box.
Embodiment. It works through your body: a run tracked by your watch, a breath timed to your pulse. Strava has more than 150 million athletes in 185+ countries. The product starts with GPS and heart rate from a body in motion, not a request.
Play. Mastery inside a world with rules and friction. Describing a game isn't playing it. Roblox averaged 138 million daily users in Q3 2026, all of them choosing to be inside a world rather than ask about one.
Space. Thinking by pointing and moving, as in Figma, a map, or a timeline you drag. Even inside the chat, spatial value forces an interface back in: OpenAI's Apps SDK exists so apps can render interactive UI in the conversation, like Zillow's listings on a map (OpenAI). When the answer is a layout, words aren't enough.
Shared record. One truth everyone sees the same way: a balance, a contract, a shared calendar, a team's project board. Agents can read and write the record, but the people who depend on it need one place to check it and trust it. Even Stripe's agent checkout protocol is designed so the merchant keeps the order and the customer relationship rather than the agent.
Glance. Some things are faster to see than to ask about: the weather, a streak, where your friend is.
If your idea doesn't clear at least one of these bars, the app is probably a costume on a skill.
Ambience Frame: build an ambient app
A traditional app is a place you go. An ambient app comes with you. It runs in the background, hears you, sees your context, and steps in without you switching windows. The value comes from being beside you while you work.
On the desktop. Clicky describes itself as "an AI buddy that lives on your computer." Press a hotkey and it sees your screen, you ask out loud, and it walks you through the tool you're stuck in, drawing on the screen to point the way. You can also say "heyclicky agent" and it goes and does the task. It works with anything on screen, no plugins needed, and it reports 25,000+ users with plans from free to $100 a month. Notice what it isn't: a destination. It has no home screen worth visiting.
In the voice layer. The models that make ambience possible shipped in 2025. OpenAI released gpt-realtime, a production speech-to-speech model, in August 2025. Google made Gemini Live's camera and screen sharing free on Android and iOS in May 2025, so the assistant can see what you see. Voice is becoming a front door to software. I'd treat that as a forecast, not a fact, although the direction is clear because of its accessibility and convenience.
On your face. Ambience is getting its own hardware. More than 7 million Meta AI glasses were sold in 2025, up from 2 million across 2023 and 2024 combined, and AI-glasses revenue almost doubled in Q2 2026. Growth has started to cool: smart-glasses sales slowed in Q3 compared with earlier in the year (Reuters). That's still millions of people wearing an assistant at eye level that sees what they see and answers in their ear.
This is the frame to watch. It's the least settled, and the one where the interface is disappearing fastest.
Most ideas mix frames
Few products live in one frame. The best ones already show the seams:
Strava is Presence at its core, through embodiment and connection. But it added Athlete Intelligence, which turns your workout data into plain-language summaries and trends. That's the Articulation Frame showing up inside the app, and it's the part that could just as well live as a skill.
Duolingo looks like an app, but much of learning a language is articulation. Its Max tier added Video Call with Lily, an AI conversation partner for speaking practice: pure Articulation Frame. The streak, the leaderboard, and the game mechanics are what earn it an interface.
Notion is space and shared record. With Notion 3.0 in September 2025 it shipped agents that can work across your workspace and connected tools for up to 20 minutes at a time. More and more of Notion gets used by asking rather than by clicking.
Meeting notetakers are an ambient app during the call, a skill when you ask about the meeting later, and a glance at the summary afterward. Otter.ai alone reported 35 million users and $100 million in ARR in 2025, and the category is built on not making you open anything during the meeting.
The discipline is to find where the core lives, build it in that frame, and add interfaces only at the edges where they earn it. My mental-models trainer is mostly articulation, and literally so: explaining a model in your own words is how you know you've learned it, so the conversation is the product. That makes it a skill. A streak your friends can glance at is presence, the one piece that might someday earn an app.