Show all articles
Toyo logo Aidan Hornsby

How AI is changing design engineering, with Aidan Hornsby from Toyo

The way designers contribute and think about design has changed. For Aidan, building with AI means making design decisions in the browser, then working out how to carry them into the next page, prompt, and engineering handoff.

By Grum

Aidan had planned to keep Toyo's brand color orange. Then, near the end of building the new homepage, he changed his mind.

"I just felt like, I don't really like how this looks, so I'm just gonna change it to blue."

But he didn't go back to the proverbial drawing board. He just made the change ad hoc with his agent directly. He could make the change with his agent and judge the result in the browser, without first producing a new set of mockups. What followed was harder to account for. Older colors still appeared in the code. Landing pages from earlier experiments remained. A design choice made in a moment now needed to become a rule that other pages, agents, and contributors could follow.

In a conversation with Grum, Aidan Hornsby, a design engineer at Toyo, described a workflow in which building and deciding happen together. Coding agents let him move quickly from an idea to something he can judge in the browser. Keeping track of what he decided, and making those decisions survive the next change, takes more work.

"I know it in my head," he said.

Designing as the thing takes shape

Hornsby contrasted his current process with a more familiar sequence: work through static mockups, define the brand kit and design tokens, discuss the choices with developers, then build.

That sequence left behind a shared record. People had seen the alternatives and participated in the decisions. A button's appearance came with some understanding of where and why to use it.

Working directly with agents changes the order. Hornsby might settle on a primary call to action, adjust a secondary style, or refine a header's behavior while looking at the implementation. The choices emerge through iteration.

He described it as "shaping it from clay as you go."

The flexibility is valuable. He can try a direction, see it working, and change his mind without first completing a separate design exercise. But the reasoning can remain in his memory while the implementation accumulates elsewhere.

"Well, I asked it to record it after the fact, but I never sat and looked at all of the CSS and did that."

For him, that leaves a gap between getting a site built quickly and understanding how consistently its design decisions have been applied. Asking an agent to document the work is another step that needs attention and verification.

A website he could no longer see all at once

The problem became concrete while reviewing Toyo's marketing site in Grum. Hornsby recognized URLs from landing pages created months earlier. Some belonged to old experiments. He was unsure which had received the newer styles.

Before deciding what to fix, he needed to understand what was there.

The main navigation represented one part of the site. Footer pages, blog posts, and less visible landing pages extended it further. Earlier advertising experiments had left pages that were easy to forget once the team moved on.

Deleting them all would have been premature. Hornsby wanted to check whether any still had search value or were being cited by an LLM. Those were possibilities to investigate, rather than established reasons to keep every page.

His first request was visual: a sitemap with small screenshots, showing which pages appeared consistent and which deserved attention. He wanted to distinguish the main site from the experiments around its edges, then choose what to update or retire.

He wanted a control panel that could "surface the signal from the noise in your own thing."

A list of colors and components could expose individual differences. The overview he wanted would help him decide where those differences mattered. A mismatch on a prominent page and a mismatch in a forgotten experiment could call for different responses.

That distinction also shaped his feedback on prioritization. Knowing that a value was outside the contract was useful. He wanted more help understanding why one finding deserved his attention before another.

The pressure of the last 20%

Hornsby cares about the details: typography, button styles, animation, and how an interface feels. There is usually another improvement available.

"The other tension is like, this is good enough to ship, go and do the other things."

His question was how to spend a limited amount of polishing time well. A growing inventory of discrepancies still leaves the designer with the work of judging them, choosing a response, and explaining the fix to an agent.

He imagined an onboarding experience that would show the biggest issues first, let him choose which to address, and turn those choices into a cleanup prompt. A prompt might ask an agent to apply the approved design contract to a selected set of pages.

The conversation also explored prompting someone to add a missing component guideline while that omission was in front of them. It would bring documentation closer to the moment when a person could explain the intended use.

These were ideas for how the product could help. Together, they describe the assistance Hornsby was looking for: make the state of the site understandable, help him choose the next useful change, and reduce the work of asking an agent to carry it out.

When the design has to leave his hands

The marketing site was largely Hornsby's own space to experiment. Toyo's product introduced a different requirement: someone else needed to carry his design work forward.

He described working on a prototype branch, refining an interface that had initially received the previous UI kit as a stopgap. He was consolidating button styles and cleaning up animation before handing the branch to engineering through a pull request.

The design system work, he said, was something he hadn't "really documented properly yet."

At that point, consistency became a handoff problem as well as a personal workflow problem. The next contributor needed to know which details represented deliberate choices. Hornsby needed a way to see whether those choices survived implementation on staging.

He planned to set up Grum on the product branch while the new design was fresh. Capturing that state could give the team a shared reference before subsequent work introduced differences.

The discussion connected that plan to change notifications: making a departure from an agreed component visible to the person responsible for its design. For a small team, a change to a carefully styled detail can otherwise become something the designer discovers later, while browsing the product.

This was the next use case Hornsby wanted to try. The conversation did not yet establish how the product rollout would go or what it would save the team.

Finding a process for what comes next

Hornsby expects the tools to keep improving. During the conversation, he considered the possibility that an agent could eventually produce an excellent design from a single prompt, while remaining uncertain about how soon that might happen.

His immediate concerns were closer to the work already in front of him: understanding the site he had built, recording choices made during iteration, and handing a coherent design to another contributor.

For Grum, his experience points toward a useful role alongside fast-moving agents. Help a team recover a shared understanding of what it has built, make its intended patterns explicit, and notice when later work departs from them. His requests for maps and cleanup guidance extend that ambition toward helping people decide what deserves their attention next.

Hornsby sees the process itself as something still taking shape.

"So we're stretching our brains, we're stretching our language, we're breaking the tools we had before."

X Discord LinkedIn