Coding agents: The new era of product design, where “taste” becomes startups’ key asset

In an era of rapid artificial intelligence development, coding agents are fundamentally transforming the product design landscape. They not only reduce development costs, but also bring human “taste” to the forefront — the ability to create a unique user experience that becomes a key competitive advantage for startups. Yasha Grozyan, lead designer at Lumos AI, shares his experience and vision for the future.

From “firefighting” to strategic advantage: transforming the development process

Building modern, complex interfaces, especially for projects that work with large volumes of data, has always been a challenge. It requires high precision, attention to detail, and careful consideration of every possible user scenario. At Lumos AI, as in many young startups, this process often turned into “firefighting”: limited time, a lack of resources, and the need to ship quickly led to compromises in design quality.

“Our first year was like a race. Runway was limited, and as the only designer, I had to sketch interfaces in a hurry. That meant many nuances were left underworked, and the final product that reached production often differed from the mockups. Details were lost, implementation became simplified, and there was no talk of ‘polishing’ at all. And in our field, where data quality directly affects payment, every poorly thought-out step had serious consequences. Experts got stuck, the team wasted time on repeat questions, and data increasingly failed quality checks,” says Grozyan.

The classic workflow, where a designer creates mockups, engineers bring them to life, and then a revision cycle follows, often proved inefficient. Design shortcomings discovered during development remained unresolved due to lack of time, and improvements affecting data quality were pushed into the backlog. This created a vicious cycle in which product quality suffered and, consequently, business metrics did too.

The AI revolution: lowering development costs and raising the role of design

The arrival of artificial-intelligence-based coding agents was a real breakthrough. They significantly reduced the cost of “building” products, turning development from a resource-intensive process into a much more accessible one. Now, for thousands of dollars a month, it is possible to create a full-fledged product that previously would have required far greater investment. Tools like Figma, which allow designers to make changes directly in the working product, have sped this process up even further.

“These changes meant that the question ‘can we build this’ stopped being the main one. There are almost no constraints left. Instead, the main challenge became ‘will people use this and come back’. And that part — product and design — remains expensive, because AI is not omnipotent here. It works brilliantly with existing data, but true design often requires synthesizing new ideas, which is still a weak point for models. This is exactly where human ‘taste’ becomes decisive,” Grozyan emphasizes.

Designing in code: a superpower for a designer

A turning point for Yasha Grozyan came while working on a project for a major client, when he had to take on development himself without prior experience with coding agents. Starting with the traditional approach, he quickly realized the advantages of designing directly in code.

“By connecting a coding agent to our codebase, I gained the ability to build a product based on existing components, patterns, and data structure. I no longer needed to explain the basics. I wrote the plan, thought through scenarios, built, tested, and polished. It felt like gaining a superpower. The speed it gave me allowed me to create a design system, improve key parts of the platform that we had previously never gotten around to, and optimize the data collection process itself,” the designer shares.

This approach made it possible to significantly accelerate the launch of new features and projects, as well as improve the quality of the result. According to him, data from projects created with coding agents passed quality checks about twice as often as those developed in the traditional way.

Quality control: the key to successful AI-driven development

However, the experience also showed that automation does not eliminate the need for control. The first failure, when technical bugs in code generated by an agent led to lost work for experts and financial losses, taught Grozyan an important lesson.

“I realized that the quality of generated code can easily spiral out of control if it isn’t managed. So I rebuilt my process around multi-layered quality checks. This includes visual testing during development, checking alignment with the design system, automated code review by several AI agents, and, of course, a final review of critical points by an experienced engineer. Over the last three months, thanks to this system, there have been no issues with the products I shipped,” he notes.

“Taste” as the new competitive advantage (moat)

In a world where technology makes development accessible to everyone, the key advantage becomes “taste” — the ability to create an exceptional user experience. This is not just visual aesthetics, but a deep understanding of user needs, anticipation of their actions, and respect for their time and attention.

“Taste is not only about judgment, but also about speed. When a product can be assembled in minutes, the bottleneck becomes the speed of making the right product decisions. AI has become my second brain, but the vision of how a product should work and feel remains mine. That vision grows out of empathy for the user and close observation of them, which models still lack,” Grozyan explains.

In an era when the App Store receives thousands of new apps every hour, many of them raw and lacking value, it is “taste” that makes a product something people choose and return to.

A new role: designer-builder

Changes in technology bring changes in hiring. The traditional roles of designer and engineer are gradually merging into a new one: the “builder.” This is a person capable of taking a feature from idea to release independently, with strong product taste, an understanding of engineering, and the ability to make decisions under uncertainty.

“If the hardest part for you is the product side, hire a builder who can’t stop thinking about the person on the other side of the screen. If the hardest part is under the hood, in infrastructure, hire an engineer. I recently wrote a job description for a design engineer myself — someone who takes a vague user and business problem and carries it through to a polished product. They have strong taste, can get into the backend if needed, and calmly make decisions with incomplete information,” Grozyan says.

5 steps for designers to start building with vibe coding

1. Understand development: Learn the basics of development, repositories, branches, and PRs so you can work effectively in a team.
2. Choose a coding agent: Select a tool (Cursor, Claude Code, Codex) and connect it to your codebase.
3. Build a small feature: Try creating a small feature, going through all stages from description to testing and polishing.
4. Build a design system: Develop a design system based on tokens, components, and patterns, and add UX and copywriting rules for the agent.
5. Make the transition gradually: Over time, move most design work into code and take on full implementation of product decisions, leaving complex engineering to professionals.

Ultimately, when technology becomes accessible to everyone, the person who understands what to build, how it should feel, and what experience to create becomes the main asset. And this “taste” — more than ever before — determines a product’s success.