Some Apps Explain Themselves

Cal AI is popular because people want to know how many calories they are eating, but they hate logging food.

That sentence explains almost the entire product. You take a photo, the app estimates what is on the plate, and a task that previously required searching, measuring and manually entering ingredients becomes a camera gesture.

The interesting part is not that Cal AI uses artificial intelligence. Plenty of apps now use artificial intelligence. The interesting part is that the product can be understood before you have even used it.

Take a photo instead of logging your meal.

That is the feature, the onboarding, the advertisement and the reason someone tells a friend about it.

Some apps are unusually easy to understand because they replace an annoying behaviour with one obvious action. They do not begin with a large category such as productivity, wellness or personal finance. They begin with a specific thing people already want and a specific process they are tired of performing.

People want the result. They hate the work currently required to produce it.

Shazam is the obvious older example. People hear a song they like, but they do not know its name. Before Shazam, identifying it might mean remembering part of the lyrics, asking someone nearby or hoping the radio presenter mentions the title. Shazam reduced all of this to one action: let the phone listen.

Loom works because people want to explain what is happening on their screen, but hate writing a long message or scheduling another meeting.

PhotoRoom works because people want polished product images, but do not want to learn Photoshop.

Otter works because people want meeting notes, but cannot take good notes while actively participating in the meeting.

Splitwise works because people want shared expenses to be fair, but hate repeatedly calculating who owes whom.

These apps are very different, yet their product logic is almost identical.

Let the phone listen instead of searching.

Record the screen instead of explaining it.

Upload the photograph instead of editing it.

Record the conversation instead of taking notes.

Add the expense instead of maintaining the accounting.

A useful product often appears when a complicated verb is replaced by a simpler one.

From construction to capture

A lot of software asks users to construct information.

To track a meal, the user constructs a nutrition record. To document a meeting, the user constructs a summary. To report a software problem, the user constructs an explanation. To create a product image, the user constructs a visual asset from editing tools, layers and settings.

The final output may be useful, but producing it feels like work.

A newer interaction pattern asks the user to capture the raw situation instead.

Photograph the food.

Record the conversation.

Show the screen.

Scan the document.

Let the software turn the captured material into something structured.

This is one of the clearest opportunities created by AI. It can transform unstructured evidence into a structured output.

The user no longer needs to understand the system's internal format before receiving value from it. They do not need to know which database category a meal belongs to, how meeting minutes should be formatted or which background-removal tool to select.

They provide what they naturally have, and the application translates it into what the system needs.

This is more important than adding an AI chat box to an existing product.

A chat box still asks the user to formulate a request. A capture interaction can remove the need to formulate anything. The user does not have to describe the meal. The meal is already there.

The most natural AI interface may therefore not look like a conversation. It may look like a camera button, a microphone, a screen recorder or a file upload.

The first success explains the product

Products like Cal AI understand something many more complicated applications miss: the user should experience the core value before being asked to understand the system.

Take a customer relationship management system such as Attio. After creating an account, the user may encounter an empty workspace, import settings, data structures and configuration decisions. The product may be powerful, but its value remains theoretical until enough work has been done to make the system useful.

Cal AI has almost the opposite structure.

The user photographs one meal and immediately receives a result. It may not be perfectly accurate, but the product has already demonstrated its promise.

This first success matters because most people do not abandon software after carefully evaluating all its capabilities. They abandon it before those capabilities become visible.

An app that is easy to explain often produces value quickly too. The explanation and the interaction share the same shape.

Shazam does not need a tutorial about audio fingerprinting. You tap the button and discover the song.

PhotoRoom does not need to explain image segmentation. You upload a photograph and watch the background disappear.

Loom does not need to persuade you that asynchronous communication is strategically important. You record your screen and send the link.

The product argument happens through use.

The demonstration becomes the distribution

Simple interactions are also easier to spread because their use can be shown rather than explained.

A Cal AI video can show a plate of food, a camera scan and a calorie estimate.

A PhotoRoom video can show an ordinary product photograph and the finished result.

A Shazam demonstration can begin with an unknown song and end with its title.

The transformation is visible within seconds.

This makes the product naturally compatible with short-form video. The advertisement does not need to make a claim and then support it with abstract benefits. It can show the entire interaction.

The user sees the problem, the action and the result.

This is why some products feel almost designed for TikTok even when their underlying need existed long before TikTok. The product contains its own content format. Each demonstration creates the same small moment of surprise.

It is tempting to call this virality, but that makes it sound more mysterious than it is. The product is easy to communicate because the product itself is already a compact story.

There is something I want.

There is an annoying obstacle.

The app removes it.

These products do not need a separate marketing narrative placed on top of them. Their interaction is the narrative.

Simple on the surface

Being explainable in one sentence does not mean the company or technology behind the product is simple.

Shazam required sophisticated audio recognition. PhotoRoom depends on computer vision. Cal AI must identify foods, estimate quantities and connect them to nutritional data. Splitwise must handle complicated webs of transactions between groups.

The simplicity exists at the boundary between the system and the user.

The company absorbs the complexity so the user does not have to.

This is an important distinction because many products expose their sophistication as if complexity itself were evidence of value. They display dashboards, modes, settings, workflows and AI agents before establishing the one thing the user should care about.

A strong product may have enormous technical depth, but that depth has been compressed into a clear action.

This compression is part of the design.

It is also why copying the visible feature is not always enough. A developer may reproduce the basic calorie estimate, background removal or transcription capability. The harder question is whether the entire experience has been organised around making that capability feel obvious, immediate and repeatable.

The technology produces the result. The product design makes the result feel natural.

Find the behaviour people keep avoiding

A useful way to search for new application ideas is not to ask what AI can generate.

Ask what useful records, documents or decisions people repeatedly fail to produce because the input process is too annoying.

People want an accurate record of what they eat, but hate entering every ingredient.

People want documentation of what was decided, but hate writing meeting notes.

People want their spending organised, but hate categorising transactions.

People want to remember what they read, but hate manually maintaining a knowledge system.

People want to report problems clearly, but hate filling out long support forms.

The opportunity is often hiding inside the neglected input.

This does not mean every manual activity should be removed. Sometimes entering information forces useful reflection. Sometimes automation introduces errors that matter. Sometimes the user needs control more than speed.

But in many products, manual input is not a meaningful part of the experience. It is simply the historical price of obtaining the output.

AI makes it possible to reconsider that price.

The most compelling product idea may not be a new thing people can do. It may be an old thing they can finally do without behaving like a database administrator.

That is why Cal AI is so easy to understand.

It does not ask people to become better calorie trackers.

It lets them take a photograph.

Connected reading

• The First Win Is Not a UI Problem
• The Interface Is Now Part of the Answer
• The End of the Software Request
• The Product Needs a Harness

AI authorship note

This essay grew from Kris Haamer asking why Cal AI had become popular, then identifying a broader pattern across apps that replace a behaviour people avoid with one obvious action. AI supported research, comparison, drafting, editing, source gathering, and implementation. Kris directed the argument toward apps that explain themselves, rejected the language of calling them the best, and chose the final version for publication.