Context
Front-end developer capacity on my product dropped from six developers to two. You would expect product delivery to fall significantly, because we had fewer front-end developers. Thanks to Claude Code and the workflows I designed around generative AI, we still released 25% more features per quarter than the same quarter the year before, before I had joined the team. That number comes from my PM, and he says the quality held.
The strategy
Every feature still starts the same way. I go through a discovery phase with my PM. I ask: the problem, why now, how success is measured, who the users are, the constraints, the timeline, and the risks.
I work in the product locally, in the real product code: pushing to branches, taking code review, matching the same standards as our senior developers. Developers review the code after, of course, but they do not get a stack of screens and have to figure out how to implement them. I handle the front end.
I built skills and agents to generate spec files, handover documents, and other outputs like a snapshot file, which captures the UI in every state. That is useful for curating the product.
As a result of the quality of those outputs, handover questions dropped by roughly 50% per feature. I measured that through refinement sessions. We used to spend about an hour going through a feature, every Tuesday and Thursday. Now it takes about 30 minutes. We can cover roughly twice as many features in the same time, thanks to the handover documentation I provide.
Example feature
One of the features I released was realistic lighting in P4 DAM. I replaced the flat grey background in the asset viewer with switchable image-based environments, so 3D assets actually pop, and users and reviewers can judge shape, materials, and quality in realistic lighting.
Claude is not perfect. It includes the wrong components sometimes. Because I have worked in front-end development before, with Tailwind, CSS, and React, I could rewrite the CSS by hand and give the developer a clean handover. Claude missed some of the component details. I still fixed the problem, and I took responsibility for the output.
Sharing the success
I turned the system I created into reusable Claude skills and agents that generate specs and handover documents, so other people could run the same pipeline. One of those systems won Best Designed at the company’s annual AI hackathon, and I got a prize for it. Other teams across the business now use it.
The outcome
When the team shrank, output went up because I had built the foundation: the spec files and the workflows we needed to keep working at scale and at speed, despite losing front-end capacity. The team delivered the features on time. I took responsibility for the front end, and I drove it further than I could before.
