Figma is becoming a graveyard for small AI-driven teams
The core friction is that as we rely more on AI-generated frontend code, we're seeing a spike in minor UI bugs and visual inconsistencies. These are the kind of trivial fixes that an LLM agent could handle in a heartbeat, but most agentic coding tools are built for developers. My designer co-founder can't easily spin up a complex dev environment just to tell an agent to move a padding value by 4px or fix a hover state. This creates a loop where the designer sees a bug, tells me, and I have to manually prompt the AI to fix it, which is a waste of everyone's time.
Even worse, our "source of truth" has completely evaporated. We started with a clean Figma library, but now that we ship so fast, I usually just implement changes directly in the code. Once the designer approves the live version, we ship it. The problem is that Figma never gets updated to match production. The design files are now basically useless artifacts, and our actual production component system is becoming a fragmented mess because we're iterating in the browser rather than the design tool.
For those of you running lean teams, I'm curious how your AI workflow looks. I'm trying to figure out if there's a way to implement a real-world AI workflow that doesn't involve the developer acting as a human proxy between the designer and the codebase.
If you've moved toward a "code-first" design process, how do you prevent your UI from becoming a sprawling disaster of inconsistent components? I'm looking for a practical tutorial or a set of habits that allow designers to interact with the implementation directly without needing a full CS degree to configure their IDE. Maybe we should be leaning harder into a strict design system that the AI is forced to follow, but getting that system locked down feels like the opposite of the speed we're chasing. It feels like we need a bridge between the "vibe coding" phase and a sustainable production codebase.