How key interface decisions are shaping the next era of human-computer interactionMade with Midjourney.Interface designers are navigating uncharted territory.For the first time in over a decade, were facing a truly greenfield space in user experience design. Theres no playbook, no established patterns to fall back on. Even the frontier AI labs are learning through experimentation, watching to see what resonates as they introduce new ways to interact.This moment reminds me of the dawn of touch-based mobile interfaces, when designers were actively inventing the interaction patterns we now take for granted. Just as those early iOS and Android design choices shaped an era of mobile computing, todays breakthroughs are defining how well collaborate with AI for years tocome.Its fascinating to watch these design choices ripple across the ecosystem in real-time. When something works, competitors rush to adopt itnot out of laziness, but because were all collectively discovering what makes sense in this new paradigm.In this wild-west moment, new dominant patterns are emerging. Today, I want to highlight the breakthroughs that have captured my imagination the mostthe design choices shaping our collective understanding of AI interaction. Some of these are obvious now, but each represented a crucial moment of discovery, a successful experiment that helped us better understand how humans and AI might work together.By studying these influential patterns, we can move beyond copying what works to shaping where AI interfaces gonext.The BreakthroughsEarly ChatGPT on top. The dev playground that preceded itbelow.1. The Conversational Paradigm (ChatGPT)Key Insight: Humans already know how to express complex ideas through conversationwhy make them learn something else?Impact: Established conversation as the fundamental paradigm for human-AI interactionThe chat interface is so ubiquitous now we barely think about it, but its the breakthrough that launched us into our current era. GPT had already been available in OpenAIs developer console, but that interface didnt resonate with a wide audience. It looked and felt more like any other dev tool. I remember playing around with it and being impressed, but it didnt capture my imagination.The decision to shift that underlying technology into a conversational format made all the difference. Whats interesting is how little the company itself probably thought of this change. I mean, they named it ChatGPT for crying out loudnot exactly the brand name youd pick if you thought you were making a revolutionary consumer product. But it proved to be the single most important design choice of this generation. The chat interface has since been copied far and wide, influencing virtually every consumer AI tool that followed.I used to think that the chat interface would eventually fade, but I dont anymore. This entire wave of generative AI tools is built around natural language at the core and conversation is the central mechanic for sharing ideas with language. Clunky chatbots will evolve, but conversation will persist as a foundational paradigm.While minimal in terms of UI, citations were a bigstep.2. Source Transparency (Perplexity)Key Insight: Without seeing sources, users cant validate AI responses forresearchImpact: Set new expectations for verifiable AI outputs in search and researchtoolsOnce people started using ChatGPT frequently, common complaints emerged around the lack of sources. While GPT could generate responses based on its massive training data, there was no way to understand where that information came from, making it difficult to use for legitimate research.Perplexity changed the game by introducing real-time citations for its AI responses, making its answers traceable and verifiable. This feature has since been heavily copied, including by OpenAI with its web search integration in ChatGPT. It addressed a fundamental trust issue: users wanted not just answers, but confidence in where those answers camefrom.This breakthrough was essential for addressing peoples concerns about using AI as a new form of search engine, but the reality is that AI does much more. LLMs can enhance question-answer style tools like Perplexity, but they also open the door to entirely new creative workflows.Conversation drives creative outputs with Artifacts3. Creative Integration (Claude Artifacts)Key Insight: Conversation can do more than generate textit can drive the creation of structured, reusableassetsImpact: Enabled new creative workflows where dialogue produces tangibleoutputsUsing artifacts was the first time I felt like I was actively creating something with AI rather than just having a conversation. My previous chats with ChatGPT and Claude had been valuable for ideation, and Perplexity had been useful for research, but artifacts gave me that a-ha momentI could start my creative workflow with a conversation and translate the best parts into tangible outputs I could export and reuse later. We still have a ways to go to make it easy to continue your workflow after creating an asset with this dialogue-based interaction loop, but were moving in that direction.For me, Artifacts proved AI collaboration would be the core of a new creative workflow, shifting my expected interaction model: instead of AI being a supporting tool, the dialogue with Claude became the core mechanic, generating creative output we refined together. AI wasnt just my assistant or copilotit was increasingly in the driversseat.Dictation input on ChatGPT iOSapp4. Natural Interaction (VoiceInput)Key Insight: Speaking allows for richer, more natural expression compared totypingImpact: Reduced friction for providing detailed context and exploring ideas withAIA lot of people are still overlooking voice as an input method. I think we have collective disbelief that it can work, thanks to a generation of mostly incompetent voice assistants (Im looking at you, Siri). But the reality is that AI transcription is very goodnow.Voice input is crucial because it allows you to actually use natural language. We forget, but as soon as we go to write anything down, we start to edit ourselves. Speaking out loud allows your brain to tap into its full improvisational creativity. This output provides much richer context to the LLMwhich is exactly what it thrives on. I think people get self-conscious or worry about seeing the messiness of real spoken language in text (all the umms and ahhs, for instance). But I can tell you from experience that current LLMs dont care about that. They see past it and even filter out a lot ofit.What youre left with is a much more natural creative ideation flow that gets captured and interpreted quickly and thoroughly by the AI. Im very bullish on dictation as a central creative skill for the next generation. Start practicing it today because it does take some time to get used to if youre new to it like Iwas.Deep integration into existing coding workflows ispowerful5. Workflow Integration (CursorIDE)Key Insight: Deeply embedding AI can supercharge where people alreadyworkImpact: Transformed code editors into AI-powered creative environmentsCursor brought the AI-led creative workflow I first experienced with Claude artifacts directly into the context of my existing codebases. Some of its features felt like no-brainersof course an IDE should do this kind of moments (like its powerful tab-to-complete feature).While I was a professional UI developer earlier in my career, I hadnt written code regularly for years until picking up Cursor. Getting back into it was always challenging because Id get stuck on new syntax or unfamiliar framework features. Tools like Cursor help sidestep many of those blockers. For example, it can be overwhelming when you first open an existing codebase because you dont know whats available or where to find it. With Cursor, I can ask detailed questions about whats going on and any code Im unsure about and get answersquickly.Working with Cursor also reinforced for me how powerful it is to have AI reading and writing directly to your file system. My work with Claude is great, but always requires an additional step to get the output out of Claude and into whatever platform I want to pick it up with later. With tools like Cursor, the output is immediately available in its final destination which makes the workflow muchtighter.The Grok button gives me instantcontext.6. Ambient Assistance (Grok Button onX)Key Insight: Users need AI help most at the moment they encounter something they dont understandImpact: Made contextual AI assistance instantly accessible alongside contentThe usefulness of the Grok button took me by surprise. Theres so much content flowing through the X feed that I regularly feel like I dont have the right context to fully understand a given post. The direct integration of the Grok AI button at the content level gives me one-click context for real-time interpretation of the information Im being bombarded with online. Whether its a meme, an article headline, or anything else, its very useful to be able to call upon the AI assistant to help me interpret what Imseeing.I think this kind of assistance will become more important as the content we encounter online is increasingly up for interpretation (is this AI generated? who published it? what are their biases? how are they trying to influence me?).This is still new and, like many things in the X platform, the design execution leaves something to be desired. But I quickly found myself wishing for this kind of ambient give me more context button on other sites I use around the web. Eventually, it feels like OS level assistants (Gemini, Siri, etc) will deliver this functionality, but the Grok button is a good example of how valuable ambient assistance can be when integrated well.DeepSeek shows the thinking that leads to itsresponse7. Process Transparency (Deepseek)Key Insight: Showing how AI reaches its conclusions builds user confidence and understandingImpact: Humanized AI responses by making machine reasoning visible and relatableThe most recent entrant to this list is Deepseek, which blew up the internet recently with the release of its R1 reasoning model. While it wasnt the first reasoning model to market, it made a critical design choice that fundamentally changed the experience for many people: it exposed the models thinking.This caught peoples attention because it shows how the machine arrives at its answer, and the language it uses in its thoughts looks an awful lot like what a person would say or feel. This visibility helps build trust in the output as users can verify whether the thought process makes sense. Another side effect is that there might be useful ideas in the reasoning itselflike maybe an idea that came up in the middle was interesting and worthy of further exploration on itsown.It reminds me of the importance of progress bars in the last generation of web apps. If an interaction happens instantly, it can feel jarring. But if it happens slowly without any indication, people will wonder if its working or broken. The progress bar helps smooth that out by helping users understand that the machine is working. Showing the AIs reasoning feels similarit reinforces that the model is indeed working. Going forward, I dont think exposing model reasoning upfront will be necessary, but it should at least be clearly accessible so users can follow along if theychoose.Leveraging Discords UI meant Midjourney could postpone creating theirown.8. Interface Deferral (Midjourney)Key Insight: Getting the core technology right matters more than having a polished interfaceImpact: Demonstrated how focusing on capability first leads to better informed interface decisionsSo much of the design conversation focuses on visual interfaces that it makes Midjourney all the more interesting. The companys choice to avoid building a custom UI in its early days and instead leverage Discord is fascinating and strategic. Even though Midjourney is a tool for visual creators, the companys core product is the tech that makes the visuals possible. That is the engine for everything else. If it wasnt excellent, people wouldnt care whether or not they had a web interface.While Midjourney now has a web UI, choosing to avoid custom UI initially allowed them to focus on the core capability of the model over the interface. Starting in Discord controlled the demand for the product by putting it in an environment where many people who werent early adopters simply wouldnt go (myself included). It also provided super-powered community-based feedback loops that enabled highly informed product decision-making.So, depending on the kind of AI youre creating, Midjourney serves as a reminder that choosing not to build a custom UI can itself be a strategic designchoice.Final ThoughtsThese eight breakthroughs arent just clever UI decisionstheyre the first chapters in a new story about how humans and machines work together. Each represents a moment when someone dared to experiment, to try something unproven, and found a pattern that resonated.From ChatGPT making AI feel conversational, to Claude turning dialogue into creation, to Deepseek showing us how machines thinkwere watching the rapid evolution of a new creative medium. Even Midjourneys choice to avoid building a custom UI reminds us that everything we thought we knew about software design is up for reinterpretation.The pace of innovation isnt slowing down. If anything, its accelerating. But thats what makes this moment so exciting: were not just observers, were participants. Every designer, developer, and creator working with AI today has the chance to contribute to this emerging language of human-AI interaction.The initial building blocks are on the table. The question isnt just What will you build with them? but What new blocks and patterns will you discover?Id love to hear which breakthroughs have captured your imagination or what patterns youre seeing emerge. Your insights might just shape the next chapter of thisstory.If you enjoyed this post, consider sharing it, subscribing to my newsletter, or following me on social media: X, LinkedIn.8 design breakthroughs defining AIs future was originally published in UX Collective on Medium, where people are continuing the conversation by highlighting and responding to this story.