The interface of the future is no longer just a chat box. As artificial intelligence moves from simple question-and-answer interactions to autonomous agents capable of performing complex knowledge work, the way we design software is undergoing a fundamental shift. At the forefront of this transition is design engineering, a hybrid discipline that merges systems thinking, interaction design, and technical implementation.

We sat down with Flora Guo, founding Design Engineer at Paradigm AI to discuss why the chat interface breaks down at scale, how AI is reshaping how we think about software, and why agentic workflows are emerging as a new frontier in productivity.

Breaking Traditional Product Boundaries

Tech Bullion: Let’s start with your role. You describe yourself as a design engineer. How is this different from a product designer or software engineer, and why is it becoming so critical in the AI era?

Flora Guo: Traditionally, product teams are built around handoffs. Designers explore user behaviour and visual interfaces, while engineers implement technical systems. Product quality depends on the accuracy of that translation.

But translation is also where things break down. Subtle decisions get lost, constraints surface too late, and teams end up optimizing for their slice of the scope instead of the whole. By holding context across interaction and implementation, design engineers can craft better end-to-end decisions. 

If you understand design but not the system, you can propose things that don’t scale or hold up in practice. If you understand the technical details but lack design thinking, you can ship something that’s correct but brittle. Design engineering lives in that overlap.

TB: Why has that overlap become more important in the AI era?

FG: Working on AI systems raises the cost of working in silos. In traditional software, outputs are well-understood, meaning you can afford some separation between design and engineering. With AI, behavior depends on models, latency, and uncertainty. Failure modes aren’t always obvious. Design engineering compresses the product development loop. You’re designing based on how the system actually behaves, not how you hope it behaves.

TB: Why does that matter more now, with AI systems? What’s the risk if teams don’t take that approach?

FG: The risk is that you build systems that look powerful, but aren’t useful for real workflows. AI agents can accomplish a lot, but aren’t deterministic. They succeed in some cases, and fail in others. That uncertainty compounds as you run them hundreds or thousands of times. If the interface treats agent behavior as predictable or self-explanatory, users won’t know what to trust or how to recover when something goes wrong.

What people actually need are handles: ways to see what worked, what didn’t, and why, and when to intervene or rerun part of a workflow. That’s why design and engineering can’t be cleanly separated here. The usefulness of the system depends on how clearly users can understand the right mental model.

Beyond the Chatbot

TB: That brings us to AI interfaces. At Paradigm, you are moving away from the chatbot model that people are used to with ChatGPT or Claude. What’s the alternative you’re betting on?

FG: Chatbots are great for isolated questions, but they’re terribly ineffective for complicated reasoning workflows repeated en masse. If you need to research hundreds of companies or analyze a thousand documents, you’re not going to sit there and chat back and forth a thousand times. It’s unscalable.

At Paradigm, we’re shifting from single-turn chat to parallel, structured research. You define the logic once, and a fleet of agents executes it across a large dataset in parallel. We’re turning the spreadsheet into a canvas for reasoning.

TB: You used the term agentic workflows. This is a buzzword we are hearing a lot in Silicon Valley. What does it mean in practice for a user?

FG: It means moving from talking to AI to managing AI.  When agents are browsing the web, enriching data, and cross-referencing sources on your behalf, the interface has to change. It starts to look less like a single-threaded conversation and more like a control surface.

My job is to make these workflows traceable. How do you trust the output of hundreds or thousands of agents? You need to see where the data came from, how logic was applied, and where confidence is high or low. If users can’t audit the reasoning, they won’t trust the results for real work.

Shaping the Future

TB:  The ideas of trust and clarity come up a lot when people talk about AI. What do you think most teams still get wrong about it?

FG: A big misconception is that the model does all the work. Some people treat AI as a black box you prompt and then wait for an answer. That works for small tasks, but it breaks down as soon as the work gets more complicated. The challenge isn’t generating output, but helping people reason about what’s happening and how to intervene when things go off track. You move from designing for prompting to designing for orchestration. The interface becomes less about asking questions and more about steering systems as they operate.

TB: How does this change the role of design engineering going forward?

FG: I see this integrated way of working becoming standard. As AI absorbs more rote execution, human effort moves upstream toward shaping systems. As we shape our tools, our tools shape us. 

That puts the emphasis on communication: making it clear what a system is doing, where it’s uncertain, and how a human should intervene. As agentic technologies evolve, so do these interfaces. Designing new ways of working, not just new tools, is precisely what makes this moment so compelling.