The Unexpected Durability of the AI Chat Interface
Why Conversation May Be Computing’s Next Platform, against all odds.
10/06/2025, just before this went to 'Press':
OpenAI just launched App SDK, which imagines using its ChatGPT distribution strength to allow apps to publish into it. App discovery happens seamlessly in conversation, and app functions function within ChatGPT. Imagine commanding and using TikTok from ChatGPT. This is the world they envision. This would be a sea change as fundamental as the OS or Browser. It would eventually displace operating systems as long as the ChatGPT-style conversational interface endures. Read on to talk about whether it will or not.The Interface Skeptic’s Conversion
There’s a moment in every technological shift when early skeptics find themselves reconsidering their assumptions. That moment arrived for us a few weeks ago. A couple of years ago, on the OpenAI Dev Day stage in 2023, Oji mentioned that he was not bullish on the chat interface as the primary way to interact with Large Language Models and other forms of transformer-based intelligence. However, what initially seemed like a limited interaction model, a simple text box that could never replace the rich, specialized interfaces of modern applications, is proving surprisingly durable and versatile.
We think Oji’s initial skepticism was reasonable. How could a single chat interface satisfy the vast variety of customer needs and use cases? The graphical interface evolved beyond our initial terminal metaphor (remember MS-DOS?) because the mouse and graphics gave great control to customers. Moreover, text/voice is a cumbersome procedural way to give commands, while the combination of screen, mouse, and touch provides faster and tactile input to cover a wider range of application use cases. For example, desktop publishing needs precise control over typography and layout. Data analysis requires interactive charts and filtering capabilities. Project management demands visual workflows and drag-and-drop functionality. Could typing into a chat box really replace all of this?
The Resilient Reality of Conversation
But something unexpected has emerged from the widespread use of AI chat platforms up till this point. The chat interface is thriving. Users are building complex workflows, creating detailed content, analyzing data, and solving problems that traditionally required specialized software, all within the confines of a conversational interface.
This resilience stems from a fundamental shift in how AI is intermediating human-computer interaction, which is fundamentally different from the old terminal. The first terminal had no intelligence; they could not ‘speak’ back except in the most stilted ways, and most of the time they told you what you did wrong (wrong syntax, type #command# -help). Compare that to now where users can express their intentions in universal natural language and have an intelligent agent translate those intentions into action or even drive specialized secondary interfaces.
The new intelligence behind the new AI chat interfaces has become what one observer called “all-encompassing utility inside this text input.” You can request data analysis and get interactive charts. You can describe a design and receive working code. You can outline a complex document and have it structured and written. The interface stays constant while the functionality expands infinitely.
Conversational computing makes people feel like experts
In addition to the fact that the response you get back from a generative chat interface is 1000x better than a command line, the appeal of chat interfaces for generative AI goes deeper than mere convenience. They tap into something psychologically powerful: they make users feel like experts. By removing technical barriers and jargon, conversation allows anyone to access sophisticated capabilities without feeling intimidated by complex toolchains or learning curves.
This forgiving nature creates a fundamentally different relationship between human and machine. Users don’t need to know the “right” way to ask for something; they can start with natural language and work toward precision through dialogue.
The interfaces are also becoming increasingly augmentable. New features like prompt assistance, suggested follow-ups, and contextual guidance can potentially help customers discover capabilities they might never have imagined. The platform can become a teacher, as well as a tool.
The Hidden Trade-offs
However, extended use reveals limitations that aren’t immediately apparent to casual users. The very features that make chat interfaces accessible can become constraints for power users and complex workflows.
The Precision Problem
Conversational interfaces excel at handling ambiguity, but they struggle with precision. Getting exact specifications often requires verbose, tedious explanations. A graphic designer who can instantly adjust spacing by dragging elements now must describe their vision in paragraphs, iterate through versions, and hope the AI interprets their intent correctly. Have you ever tried to actually edit a photo in Google’s Nano Banana? It can be excruciating without a prompt dictionary.
This imprecision leads to a cycle of mistakes and redos. In traditional interfaces, errors are often immediately visible and correctable through “undo.” In conversation, errors may not surface until the end of a long exchange, requiring backtracking through the entire dialogue thread.
Cognitive Load and Interface Fatigue
The cognitive overhead of processing long text responses can be surprisingly taxing. While a dashboard might display ten data points instantly scannable at a glance, the conversational equivalent requires reading and mentally parsing paragraphs of explanation. Users report feeling overwhelmed by verbose AI responses, even when the information is helpful.
There’s also a psychological fatigue that comes from constant precision requirements. When every request needs to be carefully articulated in natural language, users can find the interface exhausting, especially for tasks that would be quick and intuitive in a specialized tool.
Limitations of linearity
Chat interfaces are inherently linear and sequential. Unlike traditional interfaces where information can be accessed randomly, conversation flows chronologically. Users must tune into the signal to fully decode the message following the thread from beginning to end to understand context.
This linearity creates challenges for managing multiple outputs or comparing different approaches. There’s no easy way to organize response collections, track versions across iterations, or quickly reference earlier work without scrolling through conversation history.
The Imagination Gap
Perhaps counterintuitively, the accessibility of chat interfaces can limit user imagination. Without menus, toolbars, or visible capabilities, users often underestimate what’s possible. They tend to ask for familiar things rather than exploring the full potential of the underlying AI capabilities. The interface that removes barriers to entry can paradoxically create barriers to discovery. Prompt assistance has not shown itself to be a full counterbalance for this, although its evolution and full adoption may tilt the scales eventually. There are certainly many cons to the conversational interface, but customers seem to be taking to it in droves. We think it’s because it superficially mimics the way we have evolved to communicate over thousands of years and that familiarity overcomes a lot of the cons above.
The Super App Question
For us, these trade-offs raise important questions about the ‘super app’ potential of conversational platforms. Are we witnessing the emergence of comprehensive platforms built directly on top of ChatGPT and Claude? The concept of super apps has been successful in markets like China with WeChat, but Western tech has struggled to replicate this model.
Chat-based AI super platforms (ChatGPT, Claude, Perplexity, etc) may have found the key. By using conversation as the universal interface, they can integrate countless specialized functions without the complexity of traditional super apps. Need to schedule a meeting, analyze sales data, write marketing copy, and plan a vacation? All of these can happen within the same conversational thread, with context flowing seamlessly between tasks. These apps could become the black hole of creativity that Office and Windows was in the 90s and 2000s before the advent of the deeply programmable web or Web 2.0.
The recent release of the OpenAI ChatGPT SDK strongly supports this.
Rethinking the Interface Paradigm
We’ve assumed that efficiency requires purpose-built tools. But conversation offers something different: universal accessibility combined with contextual intelligence. The interface adapts to the user rather than requiring the user to adapt to it. This represents a positive fundamental inversion of the traditional relationship between human and machine.
The question isn’t whether chat interfaces are perfect—they clearly aren’t. The question is whether their benefits outweigh their costs for the majority of computing tasks, for most people. And whether the limitations can be addressed without sacrificing the core advantages.
The missing piece for AI Chat Interface hegemony
What if you could get the benefits of a conversational human interface with the precision of a graphical interface? Enter the idea of a dynamic UI. You chat with an AI and it paints a user interface that is unique and perfect for this particular session you have with it. If the topic requires a spreadsheet, then the interface is chat and an excel-like spreadsheet interface. If the topic is an image then the interface is chat and a precise line and pixel driven interface. The idea of ‘dynamic UI’ will give life and longevity to AI chat interface as a universal mediator of an infinite set of business and consumer workflows.The interface skeptics may have been wrong not about the limitations of chat, but about the adaptability of conversation itself. In a world where intelligence can be embedded in any interaction, the oldest human interface may indeed be evolving into the newest computing platform. But like all platforms, it will need to grow beyond its initial form to reach its full potential.
Want to dive deeper?
Check out our book BUILDING ROCKETSHIPS 🚀 and continue this and other conversations in our 💬 ProductMind Slack community and our LinkedIn community.
🎥 Check out our latest podcast on the ProductMind YouTube channel. In this episode we challenge the hype and discuss:
⚙️ AI standards are forming-but who defines them?
💸 Real costs like rework + maintenance are ignored.
👥 Job displacement is more complex than any spreadsheet can capture.
🎧 Tune in now.
🎥 YouTube → https://lnkd.in/gkmBamxC
🎵 Spotify → https://lnkd.in/gKYT-jxF
🎵 We are excited to announce we have expanded our podcast 🎙️to Spotify. Please give us a listen and if you like what you hear share with a friend, follow us, and (or) rate us 5 stars. ⭐️⭐️⭐️⭐️⭐️