The App Layer Will Decide Which AI Models Matter

Figma CEO Dylan Field recently described the current state of AI as the “MS-DOS era,” where the prompt is the interface—powerful, but only for those who know how to command it. That analogy cuts deeper than it might first appear. MS-DOS, released in 1981, required users to type commands to run programs and manage files, which made it nearly impossible for non-technical users to operate. What brought personal computing into the mainstream wasn’t the operating system itself, but the graphical user interfaces that came after. Similarly, the internet only became broadly useful once browsers, search engines, and everyday web apps turned an academic tool into something anyone could use.

The same pattern is now playing out with AI. The raw capability of large language models will not be enough; the layer that transforms that infrastructure into usable tools will determine whether the technology sticks. As with the Macintosh in 1984 or the first wave of smartphones, it wasn't the hardware that defined the era—it was the apps that dedicated teams built on top.

Why Interaction Design Decides Adoption

Building that layer is not sufficient on its own. The quality of design details is what separates widely adopted products from those that fade. The early internet saw dozens of browsers and search engines emerge, but the ones that defined the web combined new functionality with clean, intuitive design. In the smartphone era, apps like Uber unlocked new possibilities, but it was the interaction patterns—pinch to zoom, inertial scrolling, live maps—that gave the most successful apps their staying power.

AI apps will need the same attention to interaction. Most people won’t leverage models directly; they'll turn to products that translate raw capability into useful actions. These won't just be wrappers around an LLM, but entirely new ways of interacting with technology, with interaction patterns that make AI feel more natural and enjoyable than typing a prompt.

Early Signs of the Shift

The early evidence of this transformation is visible across fields, from creative work to mental health and parenting. AI apps are emerging that feel less like chatting with a computer and more like using tools native to the problems they solve. One example is the Good Inside app, which trains its parenting advice chatbot on the work of therapist Dr. Becky. Its success doesn't come just from the underlying model, but from the interface: clean fonts, a pale yellow interface, and empathetic responses presented as simple cards. Even small touches like a typing animation add up to an experience that feels calming and reassuring to a parent struggling with bedtime.

The same kind of adaptation will matter across AI apps. Interfaces will need tuning to reflect the needs of the people they serve, whether parents, lawyers, doctors, or designers. Off-the-shelf chatbots could have the same technical discussions, but without that intentional blend of content, tone, and interaction design, the impact is far weaker.

  • Good Inside: a parenting chatbot with quick, accessible guidance and a calming interface designed for frazzled parents.
  • Headspace’s Ebb: an AI companion offering safe, supportive conversations and personalized mindfulness recommendations.
  • Granola: a lightweight AI notetaker that captures meeting audio and delivers summaries and searchable transcripts.
  • Perplexity: an AI-powered search app combining chat with live web results for real-time product comparison and topic exploration.
  • Duolingo’s Lily: an AI video companion for practicing languages through real-time conversation.
  • Harvey: an AI legal assistant offering law firms secure drafting, research, and workflow tools.

When Interface Outweighs Model

The reaction to GPT-5’s launch in mid-August is a telling signal. The most striking change wasn't the model's expanded capability but the simplification of the model picker in ChatGPT—an interaction design choice that sparked a strong emotional response from users. That reaction suggests design decisions at the app layer often outweigh model advances in the eyes of everyday users. Raw capability remains critical, but the way it's packaged is what most people notice first.

The industry is already responding. Atlassian’s acquisition of The Browser Company indicates how even familiar tools might be reimagined as part of the AI app layer, moving the browser from a passive holder of tabs to an active interface that helps apps work together.

This is why the moment is so significant for product teams. The breakthroughs will come not just from models, but from how designers, developers, and product managers turn them into apps people care about. The key question is how these apps make users feel: supported as a parent, inspired as an artist, confident as a lawyer. Success comes less from features and more from emotional resonance—the details of interaction that deliver on a user's needs in the moment.

For those building with this technology, the rise of the app layer represents an opportunity to shape how AI feels to use. Product builders will need to create interactions that surface AI outputs in a seamless and satisfying way, supported by reliable systems that can scale. Expect a flood of new entrants across industries, all aiming to transform daily interaction with AI. Some will stand out through design; most will blend into the pack. A few might become as transformative as the GUI was for computing.