The developer has an AI agent and is still waiting for the design. What Shopify changed

More reconciling than coding

In October 2025, a frontend developer posting on Hacker News as linchien asked others how they handle the gaps between design, product and engineering. They described something most teams will recognise. Designs change after delivery, specs don’t cover everything, important context slips through. They had tried AI coding tools and admitted they speed things up. But without full context, the generated UI rarely landed cleanly and needed a lot of patching.

“The actual waste isn’t writing code, it’s trying to keep intent and changes aligned.”

Den Delimarsky from GitHub explains where this comes from in a post about spec-driven development. Language models are great at completing patterns, but they don’t read minds. Given a vague prompt, a model will make reasonable assumptions, and some of them will be wrong, often discovered deep into implementation. Delimarsky adds that design system constraints belong in the plan from day one instead of being bolted on later.

Shopify: the prototype that becomes the product

The AI in Design Report 2026 by Designer Fund and Foundation Capital describes how Shopify’s design team closes that gap. When the company announced in April 2025 that using AI was no longer optional, designers got broad access to tools and time to experiment. Today, according to the report, roughly 60% of them push code to production, and the team has built more than 5,500 prototypes on Quick, Shopify’s internal hosting platform. A prototype a designer builds together with the company’s AI coding agent is no longer a static reference for engineers. Quite often it becomes the final product.

What convinces us most is a smaller story from the same case study. Taylor Cottrell, a design apprentice, was about to design a new card for the app’s feed. Before opening Figma, he dropped an old engineering spec into an AI assistant and asked it to map out the technical constraints. The tool surfaced latency, lazy loading and deduplication before he drew a single screen. His conversations with engineers then started from a shared understanding of the problem rather than a list of questions.

Another designer, Luke Dupont, built a working store prototype in an afternoon, using real merchant data and the app’s existing components. Instead of making a deck, he shared a link where anyone could watch the prototype change as he worked. The link travelled up the chain, and by Monday Luke had a team and a real project.

Sierra cut the loop from days to a morning

The same report describes Sierra, where the design team is nearly 20 times smaller than engineering. Static mockups often fell apart when exposed to dense, real-world data, so the company replaced its multi-day handoff loop with continuous prototyping in the product environment. We cover this case in more detail in our post on a design team that can’t keep up.

What a design for developers should include

Not every company has designers who ship code, and not every company needs them. Both cases point to a simple rule, though: developers and their agents should get answers before they have to ask. A complete design handoff usually covers:

• every screen state, including empty, loading, error and no access,
• behaviour on phones, tablets and large monitors,
• design system components named the same way as in code,
• colours, spacing and typography as tokens rather than one-off values,
• final copy, with language versions and error messages,
• a description of interactions and animations,
• criteria that tell you the implementation matches the design.

AI can now prepare or check a good part of this list on the design side. It can catch missing states, describe components in language a developer and their agent understand, and compare the result with the design after release. In our service this is the designs ready to build area: developers get a complete, documented design and come back with questions less often. We hand over designs prepared this way in our UX/UI design projects for websites and apps too. How a well-prepared design system helps here is covered in our post on design systems for AI agents.