When ChatGPT was released to the public in late 2022, I was conducting user research on conversational AI, giving me a front-row seat to one of the most rapid technological pivots in recent history. After the initial novelty of using large language models (LLMs) wore off, my colleagues and I started asking a new question: what do people want AI interfaces to become next? To answer it, we went to the users who spend the most time using them.

To help participants imagine what AI interfaces could look like in five years, we built an interface toolkit designed to bridge the present and the future. We analyzed the attributes of today’s interfaces, from standard chatbots to agentic platforms, and combined them with inspiration from science fiction to create a library of interface elements.

We organized these elements into 47 distinct “building blocks” across four areas: Triggers (how tasks start), Outputs (how results are delivered), Customization (personalization settings), and Task Oversight (monitoring progress). We asked experienced AI users to share what they like about current tools, design ideal interfaces for five real-world scenarios, and share their most “blue sky” ideas. Every input was mapped back to the same set of building blocks.

By triangulating those three data streams, we uncovered four themes that point to the next era of AI interface design.

1. The Future is Seamless and Ambient

Right now, most people experience AI in one place: a single app or chat window. In the next five years, users expect it to follow them across devices and contexts. When we asked participants to describe their ideal AI interaction they described an ambient, multimodal presence that moves with them.

83% of participants described a future where their conversation with an AI continues seamlessly across devices. While text still matters, participants said they want their experiences to be centered around voice, mentioning voice commands and spoken responses far more often than text.

Participants also wanted this “invisible partner” to respond to context, not just commands. Many expressed interest in passive triggers like location-based cues and biometric sensors so the AI would understand when and where it’s needed without being asked.

As one participant described: “I’d like it to be integrated and seamless. When I get in my car, [it] just would transition through the car, and then to my smart home, to really have that constant connection.”

2. The Experience Must Be Deeply Personal

AI users in our study largely think of AI less like a software tool and more like a human relationship. When we tell a close friend something, we expect them to remember it. We don’t want to explain that we are vegetarian every time we order dinner, we expect them to know.

Participants prioritized deep customization because they want AI to honor this same social contract. They want to front-load an AI with information and they want it to retain and apply that knowledge proactively.

One user noted they wanted an AI that “just gets me”, while another noted: “In 2030, it [should] actually be able to learn my routines … what I should be eating … know the ins and outs of my work … or even bring up a song. It’ll know my music taste.””

Marketing Technology News: MarTech Interview With Fredrik Skantze, CEO and Co-founder of Funnel

3. The Dominant Model is Delegation (With Transparency)

One of the biggest shifts we saw was a move away from collaboration and toward delegation. For years, the industry has focused on “human-in-the-loop” systems, where a user and AI constantly pass the baton back and forth. However, our data suggests that  users are tired of micromanaging; they want to stay in control, but not do every step themselves.

Many participants said they were ready to delegate tasks to AI like meal planning and travel booking. To borrow a Star Trek reference, they want to be the Captain on the bridge, saying “Make it so,” rather than the Engineer tweaking the engine.

However, autonomy comes with a condition: transparency. Participants said they want AI to provide “glass box” interfaces that offer them final control. For example, if an AI is booking a flight, they want to see the plan and approve it before anything is confirmed. As one participant put it, “My ideal AI would be one that is competent and can complete tasks on its own, but never without my oversight and direction.”

4. AI Interaction Needs are a Spectrum

To create a true “one-size-fits-all” partner, AI interfaces will need to be radically adaptive. Our data showed that users do not wish to interact with AI in a single way; they shift between mindsets depending on the task.

Our network analysis identified five primary user mindsets:

• The Delegator: The “human-at-the-lever” who wants fast, autonomous execution (with clear visibility and final control).
• The Pragmatist: The “human-in-the-loop” who prefers real-time co-creation to build trust.
• The Confider: A companion-focused user who values constant presence.
• The Deliberator: A cautious user who wants a sounding board for complex decisions.
• The Vibe-setter: A user looking for immersion and atmosphere.

The winning interfaces of 2030 will be able to detect these shifts in real-time. The same user might be a Delegator when ordering groceries (“Just get the usuals”) but a Deliberator when planning a vacation (“Where should we go?”). The interface needs to change its tone, pacing, and autonomy to match the moment.

The Road Ahead

AI users are no longer asking, “What can this cool technology do?”, but more practical questions like, “How can this partner help me live my life?”

Building for this future requires a pivot in design thinking. We need to move beyond the chat box and build interfaces that are personal enough to know us, transparent enough to earn trust, adaptive enough to match our changing mindsets, and capable enough to let us step back, as long as we can keep one hand firmly on the control lever.