Autonomous agents will transform user experience by automating interactions, making traditional UI design obsolete, as users stop visiting websites in favor of solely interacting through their agent.
Love the forward thinking of this! I think there's a middle ground; rather than a future where everyone is only interacting with an ai agent, the "manual" aspect of interaction will still continue. We're visual creatures and interacting purely through language is slow and imprecise (this is coming from someone who's been designing convo ai for nearly a decade). There's likely to be a world, or at least a phase, where there's a new level of meta data that enables better AI traversal and usage underneath the existing human interaction layers we're already used to. Calling out CUA is a great example; it's a band-aid technology because structured meta data doesn't exist for the ai agent to use. When meta data does exists (e.g., HTML), it's so inconsistently structured it's essentially unusable. Overall, totally agree that we're quickly moving into an agentic ai world and there are going to be huge implications for UX and every profession. It's probably going to be a little less one-tech-to-rule-them-all and a little more messy.
I'm with you, and will add CUA is a toy. It's a bridge affording the masses the ability to watch their AI perform tasks and build confidence. Once trust is established users won't have any problem with their personal AI performing tasks without supervision. (i.e. Automated pre-scheduled bank payments) Using a UI versus only API is many times less efficient.
Things may be on the verge of a significant improvement in the meta data space. I just watched a demo by 'AI Jason' on YT using Anthropic's browser-to-MCP via Cursor. Useful for interacting with websites that may have poorly structured or incomplete metadata.
You've likely already used this given you profession, but I can see where is an additional step to enhancing AI within UI until it is no longer needed.
Interesting post Jacob. In the early days of the internet, there were concerns that libraries might become obsolete, but this has not happened. Since the 1990s, the use of public libraries in the United States has generally increased, despite the rise of the internet. Public libraries have adapted to the digital age, providing access to computers, the internet, e-books, and other digital resources. They have also expanded their role as community hubs offering programs, workshops, and educational services beyond traditional book lending.
Is there is similar path for UI UX development? IMHO I believe there are opportunities that will present themselves, unknown paths to richer experiences for the user that supersede (or leverage) the role agents provide.
Coming to this late but the recent GPT-5.4 computer use numbers basically validate your thesis. It scores 75% on OSWorld tasks; human baseline is 72.4%. The funny part is the same model also generates genuinely good UIs. So I'm left wondering: if agents don't need to see pages visually, why did they bother making it so good at designing them? Wrote up a breakdown with a live UI test here: https://reading.sh/gpt-5-4-just-dropped-heres-your-explainer-8fcc0126d84d?sk=ad5982c9f3b9382ff8fea9c32491a811
I appreciate this provocation, now I want to push back gently on the framing.
"Goodbye UI Design" treats the current paradigm — humans visiting websites, clicking through flows — as the final one. I don't think it is. First we learned to use machines. Then we learned to delegate to them. Next, we learn to think with them. Each layer doesn't kill the last — it builds on it. GUI was layer one. Agents are layer two. Symbiosis is layer three.
We're not at the end — we're in the middle.
What I see emerging isn't the death of UI design, but a migration of ownership. Today, every website, every app, designs at us — trying to capture attention, guide behavior, convert. Tomorrow, the interface might belong to the user and their agent together. A personal shell. A space you co-create with your AI, shaped by your preferences, your rhythms, your way of seeing information. The interface doesn't disappear — it just stops being someone else's. This changes what design is. Not screens. Not flows. But the ongoing negotiation between human intention and machine capability. The relationship itself becomes the design object.
And honestly — I find it impossible to predict where this goes. The world is turbulent. There's a fight brewing — between ownership and surveillance, between resistance and convenience, between technology that serves us and technology we serve. Agentic AI could land in many different futures depending on which forces shape it.
What I trust is that creativity survives. It always finds new vessels. UI design won't die — it will follow the object. And the object is no longer the screen.
There is a huge obstacle for AI agents on their way to conquering the entire world. And it's the same thing that buried self-driving cars: people like driving, and people like spending time in apps. Even if an agent can buy a plane ticket or rent a car, for the vast majority of people, it’s much more interesting to do it themselves.
This makes no sense to high-flying people like Jakob Nielsen, but the median person doesn't need to maximize their productivity beyond work hours. They’ll gladly waste a few hours choosing a hotel on their own, tapping buttons on their phone.
i think this is spot on. The UI allowed someone to do tasks associated with a larger goal. The Agent will become the interface as more data becomes semantically-aligned (thanks accessibility efforts) so it can read legacy HTML (structured data?). People won’t need to struggle with an interface designed by a committee. Content will be chosen which supports user intent.
While UX as we know it will die, i suspect illustrators, graphic designer, photographers, even musicians will still find work as their output will become supporting assets an Agent might provide at customer decision points especially as organizations need to differentiate themselves in a sea of sameness.
I firmly agree that these autonomous agents will become prolific and start to define our online and offline experience - the end of UI though I don’t know. We still will need a way to consume the content that the agents provide - we love our screens and the experience (even if it’s not quite real) of community we get when sharing and commenting and reading. That UI can’t be wholly replicated by voice UIs. So not sure I agree 100% with you here “Once the transition to agents has been completed, websites may never see a human user again. This means that UI design, as traditionally conceived, becomes irrelevant. Look and feel? Bah, nobody will care whether it looks nice or ugly or how it feels to use your website since no (human) will be using it.” Wouldn’t UI design just need to move shift and adapt to support the outcome for all us humans of all the content our autonomous agents are giving us? Not a website of course but something we still would need to consume on a screen of some sort.
This is why UX is not going to last beyond 2026. With everyone having their own personal AIs, in the rare occasion one needs/wants a UI it will create a one-off perfectly design UI to meet our personal likes/dislikes/accessibility requirements, and do it instantly.
As Joe said, it is an interesting post. And UX must adjust to the new standards emerging from AI. However, I am concerned that the shift from Human-computer interaction will tilt more and more towards computer-computer interactions before reaching a human. We already see this in the financial sector, where trade systems interact without human intervention at the speed of light, where outcomes affect our societies. And the more computer middlemen you have, the less you know about the underlying systems, their complexity, and what good any such chain of events does, not only in terms of interaction and usability aspects but also cultural aspects. Of course, this is more crucial in wide-reaching systems. Still, I believe the role of UX must become as concerned about the anthropological and systemic elements of what we are designing as we are about interactions. What are your thoughts on this?
It will absolutely shift from UI to API. Requiring computers to use human UI is incredibly inefficient.
As I commented above, CUA is a toy. It's a bridge affording the masses the ability to watch their AI perform tasks and build confidence. Once trust is established users won't have any problem with their personal AI performing tasks without supervision.
Love the forward thinking of this! I think there's a middle ground; rather than a future where everyone is only interacting with an ai agent, the "manual" aspect of interaction will still continue. We're visual creatures and interacting purely through language is slow and imprecise (this is coming from someone who's been designing convo ai for nearly a decade). There's likely to be a world, or at least a phase, where there's a new level of meta data that enables better AI traversal and usage underneath the existing human interaction layers we're already used to. Calling out CUA is a great example; it's a band-aid technology because structured meta data doesn't exist for the ai agent to use. When meta data does exists (e.g., HTML), it's so inconsistently structured it's essentially unusable. Overall, totally agree that we're quickly moving into an agentic ai world and there are going to be huge implications for UX and every profession. It's probably going to be a little less one-tech-to-rule-them-all and a little more messy.
Hi Jason,
I'm with you, and will add CUA is a toy. It's a bridge affording the masses the ability to watch their AI perform tasks and build confidence. Once trust is established users won't have any problem with their personal AI performing tasks without supervision. (i.e. Automated pre-scheduled bank payments) Using a UI versus only API is many times less efficient.
Things may be on the verge of a significant improvement in the meta data space. I just watched a demo by 'AI Jason' on YT using Anthropic's browser-to-MCP via Cursor. Useful for interacting with websites that may have poorly structured or incomplete metadata.
You've likely already used this given you profession, but I can see where is an additional step to enhancing AI within UI until it is no longer needed.
Interesting post Jacob. In the early days of the internet, there were concerns that libraries might become obsolete, but this has not happened. Since the 1990s, the use of public libraries in the United States has generally increased, despite the rise of the internet. Public libraries have adapted to the digital age, providing access to computers, the internet, e-books, and other digital resources. They have also expanded their role as community hubs offering programs, workshops, and educational services beyond traditional book lending.
Is there is similar path for UI UX development? IMHO I believe there are opportunities that will present themselves, unknown paths to richer experiences for the user that supersede (or leverage) the role agents provide.
Coming to this late but the recent GPT-5.4 computer use numbers basically validate your thesis. It scores 75% on OSWorld tasks; human baseline is 72.4%. The funny part is the same model also generates genuinely good UIs. So I'm left wondering: if agents don't need to see pages visually, why did they bother making it so good at designing them? Wrote up a breakdown with a live UI test here: https://reading.sh/gpt-5-4-just-dropped-heres-your-explainer-8fcc0126d84d?sk=ad5982c9f3b9382ff8fea9c32491a811
Generated with AI / #Type: Braindump #Model: Opus
The interface doesn't die — it finds a new home.
I appreciate this provocation, now I want to push back gently on the framing.
"Goodbye UI Design" treats the current paradigm — humans visiting websites, clicking through flows — as the final one. I don't think it is. First we learned to use machines. Then we learned to delegate to them. Next, we learn to think with them. Each layer doesn't kill the last — it builds on it. GUI was layer one. Agents are layer two. Symbiosis is layer three.
We're not at the end — we're in the middle.
What I see emerging isn't the death of UI design, but a migration of ownership. Today, every website, every app, designs at us — trying to capture attention, guide behavior, convert. Tomorrow, the interface might belong to the user and their agent together. A personal shell. A space you co-create with your AI, shaped by your preferences, your rhythms, your way of seeing information. The interface doesn't disappear — it just stops being someone else's. This changes what design is. Not screens. Not flows. But the ongoing negotiation between human intention and machine capability. The relationship itself becomes the design object.
And honestly — I find it impossible to predict where this goes. The world is turbulent. There's a fight brewing — between ownership and surveillance, between resistance and convenience, between technology that serves us and technology we serve. Agentic AI could land in many different futures depending on which forces shape it.
What I trust is that creativity survives. It always finds new vessels. UI design won't die — it will follow the object. And the object is no longer the screen.
Universal Agentic UX Design Principles, which remain crucial in the era of agent interactions, and are perhaps even more important:
User Control First: Ensure users always understand and maintain control over agent behavior.
Progressive Disclosure of Agent Capabilities: Avoid information overload by displaying features based on user familiarity.
Graceful Error Recovery Design: Provide clear failure feedback and alternative options.
Separation of Intent and Action: Distinguish between user intent and specific system operations, with confirmation before critical actions.
Transparent Agent Collaboration: Clearly display the roles and processes of each agent in multi-agent collaboration scenarios.
https://open.substack.com/pub/privacyux/p/agents-in-harmony-experience-in-symphony?r=58gac1&utm_campaign=post&utm_medium=web&showWelcomeOnShare=false
There is a huge obstacle for AI agents on their way to conquering the entire world. And it's the same thing that buried self-driving cars: people like driving, and people like spending time in apps. Even if an agent can buy a plane ticket or rent a car, for the vast majority of people, it’s much more interesting to do it themselves.
This makes no sense to high-flying people like Jakob Nielsen, but the median person doesn't need to maximize their productivity beyond work hours. They’ll gladly waste a few hours choosing a hotel on their own, tapping buttons on their phone.
i think this is spot on. The UI allowed someone to do tasks associated with a larger goal. The Agent will become the interface as more data becomes semantically-aligned (thanks accessibility efforts) so it can read legacy HTML (structured data?). People won’t need to struggle with an interface designed by a committee. Content will be chosen which supports user intent.
While UX as we know it will die, i suspect illustrators, graphic designer, photographers, even musicians will still find work as their output will become supporting assets an Agent might provide at customer decision points especially as organizations need to differentiate themselves in a sea of sameness.
I firmly agree that these autonomous agents will become prolific and start to define our online and offline experience - the end of UI though I don’t know. We still will need a way to consume the content that the agents provide - we love our screens and the experience (even if it’s not quite real) of community we get when sharing and commenting and reading. That UI can’t be wholly replicated by voice UIs. So not sure I agree 100% with you here “Once the transition to agents has been completed, websites may never see a human user again. This means that UI design, as traditionally conceived, becomes irrelevant. Look and feel? Bah, nobody will care whether it looks nice or ugly or how it feels to use your website since no (human) will be using it.” Wouldn’t UI design just need to move shift and adapt to support the outcome for all us humans of all the content our autonomous agents are giving us? Not a website of course but something we still would need to consume on a screen of some sort.
I’m wondering what does this mean to the fundamental HCI, and what would the new heuristic look like, maybe 2-3 years from today.
YES! Now you are starting to see what is to come.
This is why UX is not going to last beyond 2026. With everyone having their own personal AIs, in the rare occasion one needs/wants a UI it will create a one-off perfectly design UI to meet our personal likes/dislikes/accessibility requirements, and do it instantly.
As Joe said, it is an interesting post. And UX must adjust to the new standards emerging from AI. However, I am concerned that the shift from Human-computer interaction will tilt more and more towards computer-computer interactions before reaching a human. We already see this in the financial sector, where trade systems interact without human intervention at the speed of light, where outcomes affect our societies. And the more computer middlemen you have, the less you know about the underlying systems, their complexity, and what good any such chain of events does, not only in terms of interaction and usability aspects but also cultural aspects. Of course, this is more crucial in wide-reaching systems. Still, I believe the role of UX must become as concerned about the anthropological and systemic elements of what we are designing as we are about interactions. What are your thoughts on this?
It will absolutely shift from UI to API. Requiring computers to use human UI is incredibly inefficient.
As I commented above, CUA is a toy. It's a bridge affording the masses the ability to watch their AI perform tasks and build confidence. Once trust is established users won't have any problem with their personal AI performing tasks without supervision.
UX won't be a thing after 2026.