Photo Credit: Courtesy of Mountain Blue Vision Photography

Most mainstream AI interfaces are built for conversation: ask a question, get an answer, refine, repeat. But what happens when you need AI to research not one company, but ten thousand? Not one policy scenario, but every possible variation across a dataset?

Flora Guo designs systems for that scale. As founding design engineer at Paradigm AI in San Francisco, she’s pioneering what are called agentic workflows: interfaces where a single question spawns hundreds or thousands of coordinated investigations, each with a transparent reasoning trail. Instead of treating AI as a simple chat partner, her work at Paradigm enables users to orchestrate fleets of agents within the structure of a spreadsheet, making complex, large-scale reasoning visible and controllable.

“Design engineering is about shaping how people actually think and work with these systems,” Guo explains. In her practice, interface and infrastructure are treated as a single system, designed to influence how people reason with AI. “It goes beyond asking a model for an answer; we’re creating systems where reasoning is structured and traceable at every step.”

This approach reflects a broader shift in how modern knowledge work is being organized, audited, and scaled—and Guo is working at the leading edge of that transformation.

Design engineering as a new discipline

Guo’s path into design engineering didn’t follow a traditional route. Initially a software product designer, she learned how to code with v0, Vercel’s AI-powered interface builder. “I started building software with v0 before I had deep programming expertise,” she says. “AI gave me a way in. I could iterate in public, ship experiments, and learn the underlying systems by improving what the model generated.” Those experiments, shared online, eventually reached Vercel’s CEO, who brought her on to work on v0 itself.

At Vercel, Guo worked on v0 itself, contributing design and implementation of interface patterns that have since spread widely through the frontend ecosystem. “You cannot design good AI interfaces from the sidelines,” she says. “You need to be inside the stack, close to the models and the data, if you want to make these systems legible to real people.” That conviction defines her current role as Paradigm’s first founding design engineer, where she collaborates directly with the leadership team on product architecture.

The emergence of design engineering as a recognized discipline parallels broader changes in software development. Analysts tracking the sector estimate that AI could automate or significantly transform more than 40 percent of tasks in knowledge-intensive jobs by 2030, reshaping demand for tools that combine usability, transparency, and automation in a single layer of interface. This forecast has intensified interest in practitioners who can bridge visual design, systems thinking, and AI infrastructure, a niche that Guo has come to occupy.

From single prompts to parallel research

Guo’s work centers on a question that sounds simple but proves difficult at scale: how should people ask AI to perform structured research across thousands of entities without losing control or context? Typical chat interfaces weren’t built for this. Their one-to-one loop: prompt, response, refinement, buckles under the weight of mass-scale research. Paradigm reimagines the interaction entirely. Prompts become parallelizable templates, applied across datasets with thousands of rows. Each cell triggers its own agent, workflow, and chain of thought. The result is coordinated investigations at a scale chat interfaces can’t approach. Enterprise interest in these agentic systems has surged, with adoption in pilot programs nearly doubling from 37 percent to 65 percent in a single quarter between late 2024 and early 2025.

“In Paradigm, you are not chatting with one model,” Guo says. “You are orchestrating a whole cohort of agents, each running its own workflow, all inside a spreadsheet interface. The design problem is making that complexity feel understandable.”

Her team has built specific patterns to manage this: cell-level reasoning that exposes how each value was derived, multi-step enrichment where one agent’s output feeds another’s input, and column-level configurations to tailor the minutiae of user’s prompts and outputs. Guo argues that this structure is not just a convenience but an answer to a deeper trust problem. “If AI is going to support high-stakes work in finance, consulting, or public policy, people need to see how an answer was produced,” she says. “Agentic workflows let you turn a black box response into something closer to a traceable investigation.” She’s refined this argument through talks at HackMIT and Columbia University, framing parallelized agentic workflows as a critical approach to reliable and auditable large-scale reasoning.

Not everyone is convinced that agentic interfaces are the right approach. “There is a real risk that agentic workflows simply bury hallucinations under more interface,” says one critic, a researcher who studies AI governance and requested anonymity to speak candidly. “Spreadsheets full of model-generated fields can look authoritative while still harboring subtle errors. Unless teams invest heavily in verification and domain expertise, structured automation might give a false sense of security.” The concern highlights a tension that runs through much of the current wave of AI tooling: whether better design can compensate for underlying model limitations.

An emerging voice in agentic interfaces

The criticisms haven’t slowed interest in Guo’s work. She’s become a convening point for founders and engineers experimenting with multi-agent systems, hosting thought-provoking conversations in San Francisco with speakers from OpenAI and Cursor. The focus stays practical: how to make agent behavior comprehensible, what patterns scale in production.

“People are hungry for concrete patterns, not just demos,” she says. “Design engineering gives us a language to discuss patterns at the intersection of interaction and infrastructure: how the interface shapes what the system can do, and how the system’s capabilities shape what the interface should be.”

Her public work also documents something larger: what it looks like to build influence through AI-accelerated learning. Guo is candid about how v0 shaped her trajectory. “AI is changing who gets to participate in building,” she says.  “I contributed to flagship products before fitting the conventional profile of a software engineer. My story doesn’t exist in isolation; it’s a preview of where the industry is going.” For designers watching from the sidelines, her path offers proof that public experimentation and open contribution can translate into institutional influence.

When asked how she squares her enthusiasm with the field’s criticisms, Guo anchors her response in legibility. “Skepticism is healthy,” she says. “The answer isn’t retreating to chat interfaces—it’s designing workflows where errors surface, get corrected, and inform the next iteration. If we treat AI as a collaborator within well-defined structures, we improve our odds of aligning it with real-world constraints.”

Her optimism is measured, grounded in iteration rather than inevitability. In this way, Guo’s story tracks the broader trajectory of AI in software: excitement tempered by questions about reliability, governance, and lasting impact. Design engineering, as she practices it, is less a settled profession than a live response to those questions.

“What matters is whether the tools we build help people reason more clearly, not just move faster,” she says. “Agentic workflows are one attempt to do that, and they’re still very much a work in progress.” Guo’s remark is an invitation rather than a conclusion, signaling that the field she’s helping define is only beginning to take shape.