A New Layer of Abstraction
Agentic AI—systems that act with minimal supervision rather than merely responding to prompts—is the latest technology to dominate tech headlines. Offerings from OpenAI, Perplexity, Anthropic, Microsoft, and Google have sparked both excitement and skepticism. Some early adopters have demonstrated the appeal, like Olivia Moore, AI Apps Partner at Andreessen Horowitz, who used an agent to pay a bill from a picture, while others have expressed frustration at the idea of AI agents crawling the web, with one developer building an infinite maze specifically to trap them.
The central question, according to the community, is not whether agents can work, but what they should be used for and how they should surface in user experience.
Beyond the Chatbot Interface
Henry Modisett, Vice President of Design at Perplexity, argues that AI should be how something works, not what a user interacts with. He draws a comparison to the early days of the touchscreen, when Apple's R&D team showed Jony Ive and Steve Jobs a working prototype only to be asked what they should do with it. That same exploratory phase is now repeating with agentic AI.
Users, in Modisett's view, don't want the cognitive load of engaging with a chatbot that has a persona. They want to press a button and have the task completed, much like starting a car without needing to inspect the engine. Agentic AI forces a necessary reconsideration of how much underlying complexity is exposed to users.
Moving Past the Toggle Tax
Beyond individual tasks, Box CEO Aaron Levie has predicted that agentic AI could herald "a new era in web interoperability," functioning across platforms the way a human assistant would. If that bears out, it might offer a long-awaited escape from the "toggle tax"—the cumulative burden of switching between contexts, apps, and browser tabs just to get routine work done.
Manosai Eerabathini, Product Manager at Figma, sums up the motivation plainly: "We weren't born to browse the web. We were born to be curious, create, and connect." Agents, in this reading, are simply the next level of abstraction that lets people spend less time managing interfaces and more time doing the work that matters.
Greg Huntoon, Design Advocates Manager at Figma, also pushes back on dystopian fears of a single all-knowing system. The more realistic model, he says, is not a monolithic polymath like HAL 9000 but a set of multiple agents with specialized expertise that work in concert, like a well-structured team. That design philosophy—distributed competence over a singular omnipotent tool—will likely shape how agentic experiences are built.
The real product opportunity lie in figuring out what a task should look like when the user's role is aspiration and intent, not execution and oversight.
