Product Design Audit: designers shipping code to production
October 31, 2026
The real unlock with designers shipping code isn't "generate more code." It's learning to build the systems and workflows where you can focus on impact vs. on the work around the work. Coding agents have improved dramatically, but I've also learned to be more intentional about how and when I use them in my design work.
This May at @DoubleDiamondHQ, I watched the Anthropic team walk through how they ship daily, and one idea stuck with me: be more ambitious with models. Watch the full talk here.
Frontier models can now handle most of the testing, review, pipelines, splitting of code and iteration. As a designer that loves details, it's easy to get caught up in shipping only polish, but leaning on models for the rest lets me keep shipping while still making time for other important work like strategy, UX, and above all, collaboration. I've shared a few reflections below and hope they help anyone exploring similar workflows.
The chart below shows a realistic view of what I've pushed to production this year. Drafts are preview links, and Merged means visible to customers. To keep the chart clear, I've omitted other contributions.
If I had to compress my learnings, I'd recommend getting the reps first, and then spending code time where it counts, building systems designed for shipping.
Repeat, Repeat, improve your craft
My first recommendation is to first focus on repetition and volume. Without this you won't be able to determine what is worth doing, volume and consistency are key. Close the gap between seeing a problem and fixing it yourself. As you learn make sure to spend focused time with engineers, learn how they work and ask for help on small changes, build trust.
What the reps are actually for:
- Learn what a "small" change costs once it hits review, edge cases, and deployment.
- Understand the pain points around the code, how to get yourself unstuck.
- Think like an engineer, understand how to reduce complexity and split work into smaller atomic units.
Optimize for leverage, not lines of code
Just because you can ship a change doesn't mean you should. It's easy to focus on small UI fixes and spend time on polishing vs. shaping an experience. What I've found has the biggest impact is work that scales beyond a single fix. For example, this year I worked on helping the design team share prototypes with URLs, which has made a big impact, and on shipping markdown prose changes first in our design system. Both are good examples of durable work to ship.
Other examples of high leverage work:
- Problems where taste matters: Interaction, responsive behavior, motion, and complex states. Places where the outcome is actually better outside of a design canvas.
- Shared primitives: Design-system improvements that raise the quality of multiple surfaces at once.
- Tools for the team: Workflows and utilities that help other designers and engineers move faster.
- Bold prototypes: Exploring bigger product ideas in code before committing to a direction.
The goal was never to ship the most code. The goal is to use code to build a better product, to learn and help others.
Learn to build systems that let you ship safely
Writing the code is often the easy part. Shipping software includes review comments, tests, rebasing, merge conflicts, regressions, changing dependencies, CI, etc.
So the real unlock with agents isn't "generate more code." It's building the systems and workflows where you can focus on impact vs. repetitive work around the work.
What you'll build here:
- Reusable skills and instructions for recurring work.
- Tests and visual checks.
- Agents that can inspect their own output.
- Clear constraints from the design system.
- A harness that catches issues before review.
- Self-learning tools that analyze sessions daily to identify new skills.
Everyone's workflow is different, I have some notes on the tools I use in my AI Design Stack.
Conclusion
Design roles continue to blur. I'm bullish on where that takes us. The best design teams will work across the stack from canvas to code. They'll build systems and custom tools that allow them to focus on impact. Let's go.