2026
2026
10
10
Systems
Systems
Strategy
Strategy
Why the best AI products have the least visible AI.
Why the best AI products have the least visible AI.
Why the best AI products have the least visible AI.
There is a pattern that emerges when you look at the AI products people actually integrate into how they work, the ones they use without thinking about using them, the ones they'd notice immediately if they disappeared.
There is a pattern that emerges when you look at the AI products people actually integrate into how they work, the ones they use without thinking about using them, the ones they'd notice immediately if they disappeared.
There is a pattern that emerges when you look at the AI products people actually integrate into how they work, the ones they use without thinking about using them, the ones they'd notice immediately if they disappeared.
The AI in these products is almost invisible.
The AI in these products is almost invisible.
The AI in these products is almost invisible.
Not hidden. Not disguised. Not absent. But so deeply integrated into the value the product delivers that it doesn't present itself as a separate layer. It's not the feature. It's the reason the feature works.
Not hidden. Not disguised. Not absent. But so deeply integrated into the value the product delivers that it doesn't present itself as a separate layer. It's not the feature. It's the reason the feature works.
Not hidden. Not disguised. Not absent. But so deeply integrated into the value the product delivers that it doesn't present itself as a separate layer. It's not the feature. It's the reason the feature works.
The products that lead with their AI, that brand every interaction with it, that make sure the user knows at every moment that they are in the presence of artificial intelligence, are almost never the ones people can't live without. They're the ones people try and abandon. The ones that feel impressive for a session and exhausting for a month.
The products that lead with their AI, that brand every interaction with it, that make sure the user knows at every moment that they are in the presence of artificial intelligence, are almost never the ones people can't live without. They're the ones people try and abandon. The ones that feel impressive for a session and exhausting for a month.
The products that lead with their AI, that brand every interaction with it, that make sure the user knows at every moment that they are in the presence of artificial intelligence, are almost never the ones people can't live without. They're the ones people try and abandon. The ones that feel impressive for a session and exhausting for a month.
This is not a coincidence. It's a design consequence.
This is not a coincidence. It's a design consequence.
This is not a coincidence. It's a design consequence.
Why teams over-expose their AI
Why teams over-expose their AI
Why teams over-expose their AI
The pressure to make AI visible in a product is real and comes from multiple directions simultaneously.
The pressure to make AI visible in a product is real and comes from multiple directions simultaneously.
The pressure to make AI visible in a product is real and comes from multiple directions simultaneously.
It comes from product strategy. The AI is the differentiator. It's what justifies the pricing, the pitch, the positioning. If users don't see the AI, how will they understand what they're paying for? How will they attribute the product's value to the right cause?
It comes from product strategy. The AI is the differentiator. It's what justifies the pricing, the pitch, the positioning. If users don't see the AI, how will they understand what they're paying for? How will they attribute the product's value to the right cause?
It comes from product strategy. The AI is the differentiator. It's what justifies the pricing, the pitch, the positioning. If users don't see the AI, how will they understand what they're paying for? How will they attribute the product's value to the right cause?
It comes from engineering pride. The system is genuinely impressive. The model performs well. The infrastructure is sophisticated. The team that built it wants that sophistication to be legible to the people using it.
It comes from engineering pride. The system is genuinely impressive. The model performs well. The infrastructure is sophisticated. The team that built it wants that sophistication to be legible to the people using it.
It comes from engineering pride. The system is genuinely impressive. The model performs well. The infrastructure is sophisticated. The team that built it wants that sophistication to be legible to the people using it.
It comes from marketing. AI is a category signal right now. Products that lead with it attract a certain kind of attention, a certain kind of early adopter, a certain kind of press coverage. Hiding the AI feels like leaving value on the table.
It comes from marketing. AI is a category signal right now. Products that lead with it attract a certain kind of attention, a certain kind of early adopter, a certain kind of press coverage. Hiding the AI feels like leaving value on the table.
It comes from marketing. AI is a category signal right now. Products that lead with it attract a certain kind of attention, a certain kind of early adopter, a certain kind of press coverage. Hiding the AI feels like leaving value on the table.
It comes from a misunderstanding of transparency. Teams that have internalized the importance of trust in AI products sometimes conflate transparency with visibility. If users should understand what the system is doing, shouldn't they see the AI doing it?
It comes from a misunderstanding of transparency. Teams that have internalized the importance of trust in AI products sometimes conflate transparency with visibility. If users should understand what the system is doing, shouldn't they see the AI doing it?
It comes from a misunderstanding of transparency. Teams that have internalized the importance of trust in AI products sometimes conflate transparency with visibility. If users should understand what the system is doing, shouldn't they see the AI doing it?
Each of these pressures is understandable. Together, they produce products where the AI is everywhere: in the labels, the loading states, the feature names, the onboarding, the marketing copy that has migrated into the interface itself.
Each of these pressures is understandable. Together, they produce products where the AI is everywhere: in the labels, the loading states, the feature names, the onboarding, the marketing copy that has migrated into the interface itself.
Each of these pressures is understandable. Together, they produce products where the AI is everywhere: in the labels, the loading states, the feature names, the onboarding, the marketing copy that has migrated into the interface itself.
The result is a product that feels like a demonstration of AI rather than a tool that uses it. Users experience the technology, not the value. Those are not the same thing, and confusing them is one of the most common design failures in AI products today.
The result is a product that feels like a demonstration of AI rather than a tool that uses it. Users experience the technology, not the value. Those are not the same thing, and confusing them is one of the most common design failures in AI products today.
The result is a product that feels like a demonstration of AI rather than a tool that uses it. Users experience the technology, not the value. Those are not the same thing, and confusing them is one of the most common design failures in AI products today.
AI as feature versus AI as infrastructure
AI as feature versus AI as infrastructure
AI as feature versus AI as infrastructure
There is a distinction that clarifies almost everything about how AI should be presented in a product.
There is a distinction that clarifies almost everything about how AI should be presented in a product.
There is a distinction that clarifies almost everything about how AI should be presented in a product.
AI as feature means the AI is the thing the user is interacting with. The chat interface. The generation tool. The recommendation engine that surfaces as a visible component. The user's relationship is with the AI itself. They're aware of it, they're managing it, they're evaluating its outputs as AI outputs.
AI as feature means the AI is the thing the user is interacting with. The chat interface. The generation tool. The recommendation engine that surfaces as a visible component. The user's relationship is with the AI itself. They're aware of it, they're managing it, they're evaluating its outputs as AI outputs.
AI as feature means the AI is the thing the user is interacting with. The chat interface. The generation tool. The recommendation engine that surfaces as a visible component. The user's relationship is with the AI itself. They're aware of it, they're managing it, they're evaluating its outputs as AI outputs.
AI as infrastructure means the AI is what makes the product work, invisibly, in the background. The user's relationship is with the outcome. They're not thinking about how the outcome was produced. They're thinking about what they can do with it.
AI as infrastructure means the AI is what makes the product work, invisibly, in the background. The user's relationship is with the outcome. They're not thinking about how the outcome was produced. They're thinking about what they can do with it.
AI as infrastructure means the AI is what makes the product work, invisibly, in the background. The user's relationship is with the outcome. They're not thinking about how the outcome was produced. They're thinking about what they can do with it.
Most products that lead with their AI are treating it as a feature. Most products that people actually integrate into their workflows are treating it as infrastructure.
Most products that lead with their AI are treating it as a feature. Most products that people actually integrate into their workflows are treating it as infrastructure.
Most products that lead with their AI are treating it as a feature. Most products that people actually integrate into their workflows are treating it as infrastructure.
This distinction is not about hiding what the product does. It's about what level of abstraction the user is asked to operate at. A user who has to think about the AI to use the product is operating at the wrong level. A user who thinks about their goal, and finds the product helping them reach it, is operating at the right one.
This distinction is not about hiding what the product does. It's about what level of abstraction the user is asked to operate at. A user who has to think about the AI to use the product is operating at the wrong level. A user who thinks about their goal, and finds the product helping them reach it, is operating at the right one.
This distinction is not about hiding what the product does. It's about what level of abstraction the user is asked to operate at. A user who has to think about the AI to use the product is operating at the wrong level. A user who thinks about their goal, and finds the product helping them reach it, is operating at the right one.
The best AI products have made a deliberate choice to put the AI at the infrastructure level. That choice shapes everything: the interface, the interaction model, the language, the metrics. It's not a branding decision. It's a product philosophy.
The best AI products have made a deliberate choice to put the AI at the infrastructure level. That choice shapes everything: the interface, the interaction model, the language, the metrics. It's not a branding decision. It's a product philosophy.
The best AI products have made a deliberate choice to put the AI at the infrastructure level. That choice shapes everything: the interface, the interaction model, the language, the metrics. It's not a branding decision. It's a product philosophy.
What products that do this well have in common
What products that do this well have in common
What products that do this well have in common
The AI products that have achieved genuine integration into how people work share a set of properties that are worth examining closely.
The AI products that have achieved genuine integration into how people work share a set of properties that are worth examining closely.
The AI products that have achieved genuine integration into how people work share a set of properties that are worth examining closely.
They define their value in terms of outcomes, not capabilities. The product doesn't help you "use AI to write better." It helps you write better. The AI is the mechanism. The outcome is the product.
They define their value in terms of outcomes, not capabilities. The product doesn't help you "use AI to write better." It helps you write better. The AI is the mechanism. The outcome is the product.
They define their value in terms of outcomes, not capabilities. The product doesn't help you "use AI to write better." It helps you write better. The AI is the mechanism. The outcome is the product.
They remove friction at the point where AI would otherwise require the user's attention. The moments where the system is doing something that could be made visible are instead handled quietly, with the result surfaced cleanly. The user gets the benefit without the overhead of managing the process.
They remove friction at the point where AI would otherwise require the user's attention. The moments where the system is doing something that could be made visible are instead handled quietly, with the result surfaced cleanly. The user gets the benefit without the overhead of managing the process.
They remove friction at the point where AI would otherwise require the user's attention. The moments where the system is doing something that could be made visible are instead handled quietly, with the result surfaced cleanly. The user gets the benefit without the overhead of managing the process.
They treat the AI's involvement as a detail of implementation, not a feature of the experience. When the system does something intelligent, it doesn't announce it. It just does it, and the quality of the result speaks for itself.
They treat the AI's involvement as a detail of implementation, not a feature of the experience. When the system does something intelligent, it doesn't announce it. It just does it, and the quality of the result speaks for itself.
They treat the AI's involvement as a detail of implementation, not a feature of the experience. When the system does something intelligent, it doesn't announce it. It just does it, and the quality of the result speaks for itself.
They invest heavily in the moments where the AI does have to become visible, making those moments precise and purposeful rather than ambient. When the system surfaces its AI-ness, it's because the user genuinely needs that information, not because the product wants credit for it.
They invest heavily in the moments where the AI does have to become visible, making those moments precise and purposeful rather than ambient. When the system surfaces its AI-ness, it's because the user genuinely needs that information, not because the product wants credit for it.
They invest heavily in the moments where the AI does have to become visible, making those moments precise and purposeful rather than ambient. When the system surfaces its AI-ness, it's because the user genuinely needs that information, not because the product wants credit for it.
The common thread is restraint. These products have resisted every pressure to make the AI more visible than it needs to be. That restraint is not passive. It's an active, continuous design decision made against significant organizational pressure.
The common thread is restraint. These products have resisted every pressure to make the AI more visible than it needs to be. That restraint is not passive. It's an active, continuous design decision made against significant organizational pressure.
The common thread is restraint. These products have resisted every pressure to make the AI more visible than it needs to be. That restraint is not passive. It's an active, continuous design decision made against significant organizational pressure.
When showing the AI is a mistake, and when it's necessary
When showing the AI is a mistake, and when it's necessary
When showing the AI is a mistake, and when it's necessary
Showing the AI is a mistake when it adds cognitive overhead without adding value. When the user has to process the fact that they're interacting with AI in order to complete a task that doesn't require that awareness. When the label "AI-generated" creates doubt rather than context. When the visible AI makes the product feel like a technology demo rather than a tool.
Showing the AI is a mistake when it adds cognitive overhead without adding value. When the user has to process the fact that they're interacting with AI in order to complete a task that doesn't require that awareness. When the label "AI-generated" creates doubt rather than context. When the visible AI makes the product feel like a technology demo rather than a tool.
Showing the AI is a mistake when it adds cognitive overhead without adding value. When the user has to process the fact that they're interacting with AI in order to complete a task that doesn't require that awareness. When the label "AI-generated" creates doubt rather than context. When the visible AI makes the product feel like a technology demo rather than a tool.
It's also a mistake when it's used to manage expectations preemptively. The instinct to label everything as AI-generated so that users will forgive quality issues is understandable but counterproductive. It teaches users to distrust the output before they've evaluated it, and it signals that the product team doesn't fully believe in what the system produces.
It's also a mistake when it's used to manage expectations preemptively. The instinct to label everything as AI-generated so that users will forgive quality issues is understandable but counterproductive. It teaches users to distrust the output before they've evaluated it, and it signals that the product team doesn't fully believe in what the system produces.
It's also a mistake when it's used to manage expectations preemptively. The instinct to label everything as AI-generated so that users will forgive quality issues is understandable but counterproductive. It teaches users to distrust the output before they've evaluated it, and it signals that the product team doesn't fully believe in what the system produces.
Showing the AI is necessary when the user needs to make a decision that depends on knowing the output was AI-generated. When the provenance of the content is material to how it should be used. When the system is operating with a level of autonomy that the user needs to be aware of in order to maintain appropriate oversight. When something has gone wrong in a way that requires the user to understand the system's involvement to make sense of it.
Showing the AI is necessary when the user needs to make a decision that depends on knowing the output was AI-generated. When the provenance of the content is material to how it should be used. When the system is operating with a level of autonomy that the user needs to be aware of in order to maintain appropriate oversight. When something has gone wrong in a way that requires the user to understand the system's involvement to make sense of it.
Showing the AI is necessary when the user needs to make a decision that depends on knowing the output was AI-generated. When the provenance of the content is material to how it should be used. When the system is operating with a level of autonomy that the user needs to be aware of in order to maintain appropriate oversight. When something has gone wrong in a way that requires the user to understand the system's involvement to make sense of it.
The distinction is purpose. AI visibility that serves the user's decision-making is necessary. AI visibility that serves the product's positioning is noise.
The distinction is purpose. AI visibility that serves the user's decision-making is necessary. AI visibility that serves the product's positioning is noise.
The distinction is purpose. AI visibility that serves the user's decision-making is necessary. AI visibility that serves the product's positioning is noise.
Every instance of visible AI in a product should pass a single test: does the user need this information to make a better decision? If the answer is no, it shouldn't be there.
Every instance of visible AI in a product should pass a single test: does the user need this information to make a better decision? If the answer is no, it shouldn't be there.
Every instance of visible AI in a product should pass a single test: does the user need this information to make a better decision? If the answer is no, it shouldn't be there.
The transparency paradox
The transparency paradox
The transparency paradox
Here is something that seems counterintuitive but holds up under examination: a product that hides its AI can be more transparent than a product that constantly displays it.
Here is something that seems counterintuitive but holds up under examination: a product that hides its AI can be more transparent than a product that constantly displays it.
Here is something that seems counterintuitive but holds up under examination: a product that hides its AI can be more transparent than a product that constantly displays it.
Transparency is not visibility. Transparency is honest communication about what matters. A product that labels every output as AI-generated but gives users no way to evaluate the output's reliability, no insight into where the system's confidence is high or low, no path to verify or correct, is not transparent. It's performing transparency while withholding the information that would actually make the user more informed.
Transparency is not visibility. Transparency is honest communication about what matters. A product that labels every output as AI-generated but gives users no way to evaluate the output's reliability, no insight into where the system's confidence is high or low, no path to verify or correct, is not transparent. It's performing transparency while withholding the information that would actually make the user more informed.
Transparency is not visibility. Transparency is honest communication about what matters. A product that labels every output as AI-generated but gives users no way to evaluate the output's reliability, no insight into where the system's confidence is high or low, no path to verify or correct, is not transparent. It's performing transparency while withholding the information that would actually make the user more informed.
A product that doesn't foreground its AI but is precise and honest at the moments when the system's involvement is genuinely relevant, when uncertainty is high, when an action is irreversible, when the system is operating outside its reliable range, is more transparent in any meaningful sense. It gives users the information they need, when they need it, without burying it in ambient AI-branding that users have learned to ignore.
A product that doesn't foreground its AI but is precise and honest at the moments when the system's involvement is genuinely relevant, when uncertainty is high, when an action is irreversible, when the system is operating outside its reliable range, is more transparent in any meaningful sense. It gives users the information they need, when they need it, without burying it in ambient AI-branding that users have learned to ignore.
A product that doesn't foreground its AI but is precise and honest at the moments when the system's involvement is genuinely relevant, when uncertainty is high, when an action is irreversible, when the system is operating outside its reliable range, is more transparent in any meaningful sense. It gives users the information they need, when they need it, without burying it in ambient AI-branding that users have learned to ignore.
This matters because the debate about AI transparency in products is often framed as a binary: show the AI or hide it. The real question is more specific: what does the user need to know about the AI's involvement in this specific moment, and how should that information be communicated?
This matters because the debate about AI transparency in products is often framed as a binary: show the AI or hide it. The real question is more specific: what does the user need to know about the AI's involvement in this specific moment, and how should that information be communicated?
This matters because the debate about AI transparency in products is often framed as a binary: show the AI or hide it. The real question is more specific: what does the user need to know about the AI's involvement in this specific moment, and how should that information be communicated?
Answering that question honestly, for every interaction, is harder than labeling everything. It's also the only approach that actually makes users more informed rather than more aware of the branding.
Answering that question honestly, for every interaction, is harder than labeling everything. It's also the only approach that actually makes users more informed rather than more aware of the branding.
Answering that question honestly, for every interaction, is harder than labeling everything. It's also the only approach that actually makes users more informed rather than more aware of the branding.
Invisibility as a measure of maturity
Invisibility as a measure of maturity
Invisibility as a measure of maturity
There is a reliable signal of product maturity in AI: the AI becomes less visible over time.
There is a reliable signal of product maturity in AI: the AI becomes less visible over time.
There is a reliable signal of product maturity in AI: the AI becomes less visible over time.
Early versions of AI products tend to foreground the technology. The team is figuring out what the system can do, and the product reflects that exploration. The AI is prominent because the team is still discovering how to deploy it. The user is invited into that discovery.
Early versions of AI products tend to foreground the technology. The team is figuring out what the system can do, and the product reflects that exploration. The AI is prominent because the team is still discovering how to deploy it. The user is invited into that discovery.
Early versions of AI products tend to foreground the technology. The team is figuring out what the system can do, and the product reflects that exploration. The AI is prominent because the team is still discovering how to deploy it. The user is invited into that discovery.
As the product matures, the team develops a clearer understanding of where the AI adds genuine value and where it's noise. The interface tightens around the real use cases. The AI moves from the foreground to the background. The product stops being about what the AI can do and starts being about what the user can accomplish.
As the product matures, the team develops a clearer understanding of where the AI adds genuine value and where it's noise. The interface tightens around the real use cases. The AI moves from the foreground to the background. The product stops being about what the AI can do and starts being about what the user can accomplish.
As the product matures, the team develops a clearer understanding of where the AI adds genuine value and where it's noise. The interface tightens around the real use cases. The AI moves from the foreground to the background. The product stops being about what the AI can do and starts being about what the user can accomplish.
This maturation is not automatic. It requires a team that is willing to remove things, to resist the pressure to make every capability visible, to prioritize the user's experience of outcome over the product's demonstration of capability. That willingness is rare, and it's one of the clearest signals of a design-led product culture.
This maturation is not automatic. It requires a team that is willing to remove things, to resist the pressure to make every capability visible, to prioritize the user's experience of outcome over the product's demonstration of capability. That willingness is rare, and it's one of the clearest signals of a design-led product culture.
This maturation is not automatic. It requires a team that is willing to remove things, to resist the pressure to make every capability visible, to prioritize the user's experience of outcome over the product's demonstration of capability. That willingness is rare, and it's one of the clearest signals of a design-led product culture.
The products that reach this maturity are the ones that last. Not because they're the most capable, but because they've figured out something that most AI products haven't: the technology is not the point. It never was.
The products that reach this maturity are the ones that last. Not because they're the most capable, but because they've figured out something that most AI products haven't: the technology is not the point. It never was.
The products that reach this maturity are the ones that last. Not because they're the most capable, but because they've figured out something that most AI products haven't: the technology is not the point. It never was.
The best AI products have the least visible AI because they've done the work of integrating it completely. The AI has stopped being something the product has and become something the product is. That integration is the hardest thing to build and the most durable competitive advantage an AI product can have.
The best AI products have the least visible AI because they've done the work of integrating it completely. The AI has stopped being something the product has and become something the product is. That integration is the hardest thing to build and the most durable competitive advantage an AI product can have.
The best AI products have the least visible AI because they've done the work of integrating it completely. The AI has stopped being something the product has and become something the product is. That integration is the hardest thing to build and the most durable competitive advantage an AI product can have.
Raphaël D. - Head of Product Design, designing at the intersection of AI infrastructure and human experience.
Raphaël D. - Head of Product Design, designing at the intersection of AI infrastructure and human experience.
Raphaël D. - Head of Product Design, designing at the intersection of AI infrastructure and human experience.
Design for AI
Thinking through the design problems that AI products create.
Not how to use AI as a designer. How to design for it.
© 2026 Design for AI. All rights reserved.
Design for AI
Thinking through the design problems that AI products create. Not how to use AI as a designer. How to design for it.
© 2026 Design for AI. All rights reserved.
Design for AI
Thinking through the design problems that AI products create. Not how to use AI as a designer. How to design for it.
© 2026 Design for AI. All rights reserved.