Claude Duet is beautiful. I still forgot to use it

We are currently living through the great consolidation of the AI interface. Every week, another lab drops a new wrapper, a new agent, or a new way to talk to a model. Anthropic recently rolled out Claude Duet, and the initial reaction from the design community was predictable: it is beautiful. It is fluid. It feels like the future of how we are supposed to interact with software.

But after living with it for a while, I found myself hitting a wall that no amount of clean UI can fix. I simply forgot to use it. This isn't a knock on Anthropic's engineering team; it's a diagnosis of a much larger problem facing every founder and builder in the space right now: agent fatigue.

The Aesthetics of Efficiency

From a product perspective, Duet is a masterclass in minimalism. It moves away from the clunky, vertical chat history we have grown accustomed to with ChatGPT and even the base version of Claude. It feels like a workspace. For a builder, this is attractive. We spend our lives in IDEs, Figma boards, and Notion pages. We like surfaces that feel like they can hold a project, not just a conversation.

The promise of Duet is that it acts as a companion. It sits there, ready to assist with the deep work. It is supposed to be the bridge between the raw power of the Opus model and the actual messy reality of our daily tasks. The design language suggests that this isn't just a bot you ask for trivia; it is a tool you use to build. But as any founder knows, a tool is only as good as the habit it creates.

The Habit Gap

The core issue with the current crop of AI agents is the cognitive load of remembering they exist. When I am in the flow of building a product or editing a piece of technical analysis, my brain follows the path of least resistance. Usually, that path leads to a search bar, a terminal, or a known workflow that has been burned into my muscle memory over a decade.

Duet, despite its elegance, requires a pivot. You have to decide to use it. In the high-pressure environment of a startup or a development sprint, that moment of decision is a friction point. If the tool isn't already integrated into where my hands are moving, it might as well not exist. This is the challenge for Anthropic and everyone else building in the 'Personal Agent' category. They are competing with the status quo of how our brains have been wired by thirty years of traditional computing.

Why Builders Should Care About the 'Duet' Model

If you are building in the AI space, you need to look at Duet as a case study in the 'Second Wave' of AI interfaces. The first wave was just the chat box. The second wave is the workspace. We are seeing a shift where the model is no longer the product; the environment is the product.

• Context is the new UI: Duet tries to understand the broader scope of what you are doing. For builders, this means our apps need to move away from being reactive and toward being contextual.
• Reducing the 'Prompt Tax': The goal of these agents is to reduce the amount of explaining a user has to do. If the agent knows your project, you don't have to write a three-paragraph prompt to get a one-sentence answer.
• The Aesthetic Moat: Anthropic is betting that a superior user experience will create brand loyalty that raw benchmarks cannot. They are trying to build the Apple of AI.

However, the 'Prompt Tax' is still there. Even with a beautiful UI, the user still has to initiate. For a builder, the real win isn't a better chat window; it is an agent that acts before being asked. We aren't there yet, and Duet, for all its polish, still feels like a very high-end version of the same conversational paradigm we've been stuck in since late 2022.

The Skeptic's View on Personal Agents

I am skeptical of the term 'Personal Agent' because it implies a level of autonomy that current LLMs simply do not possess. When I hear agent, I think of a system that can go off, execute a series of complex tasks, and report back with a finished result. Duet is more of a 'Personal Assistant' in the sense that it helps you do the work, but it doesn't do the work for you.

For those of us in the crypto and AI trenches, we know that true utility comes from permissionless execution. If I have to babysit the agent, I am still the one doing the heavy lifting. The fatigue comes from the fact that we have a hundred different assistants, all asking for our attention, and none of them can talk to each other. I have my coding assistant, my writing assistant, and now my Claude Duet assistant. It is a fragmented ecosystem that adds to the noise rather than quieting it.

The Lesson for Founders

If you are developing a product right now, do not mistake design for a value proposition. You can have the most beautiful dashboard in the world, but if it requires the user to change their fundamental behavior without providing a 10x return on that effort, they will forget it exists. The 'Forgetting Curve' for new AI tools is steeper than most founders realize.

The success of a tool like Duet won't be measured by how many people sign up for the beta or how many design awards it wins. It will be measured by its retention on a Tuesday afternoon when the user is stressed, behind schedule, and just needs to get things done. In those moments, we don't want beauty; we want invisibility.

The ultimate interface isn't a better chat box; it is the absence of an interface where the work just happens.

Anthropic is closer than most. They understand that the 'vibe' of the software matters. They understand that builders want a tool that feels professional and refined. But they are still fighting the battle of the browser tab. As long as the agent lives in a separate tab or a separate window that I have to consciously navigate to, it will remain a secondary tool.

Where We Go From Here

I expect to see these agents move closer to the OS level. We are already seeing Apple and Microsoft try to bake this directly into the hardware and the operating system. That is the only way to solve the 'I forgot to use it' problem. When the agent is part of the file system, part of the keyboard, and part of the screen, the friction disappears.

Until then, tools like Claude Duet are like high-end stationery. They are beautiful to look at, and they make the act of working feel more intentional, but they don't necessarily make you more productive if you don't have the discipline to reach for them every single time.

My takeaway for the STKR community is simple: watch the design patterns Anthropic is using. They are setting the standard for how AI software should look. But don't be fooled into thinking that look is the same as utility. As builders, we need to focus on integration and automation. If you make me remember to use your tool, you've already lost the battle for my attention.

Read the original at Lenny's Newsletter →