Tools the Vercel Product Design Team Actually Uses | Hannah Hearth
Briefly

Tools the Vercel Product Design Team Actually Uses | Hannah Hearth
"Even within our small team, everyone is using completely different tools right now. We have a wide range of AI usage, from early adopters who haven't gone a day without using AI in 6 months to skeptics who still find it mostly useless in their creative endeavors. AI tools for design workflows are clunky and lagging behind relative to AI tooling for engineering workflows."
"Timo has been using Codex as his primary coding agent but created a skill for Claude to review Codex's work. The skill file instructs Codex to share its plan with Claude and ask for feedback (i.e. Is this more complex than it needs to be? Is there duplication? Is the API actually well designed?). The file also instructs Codex not to blindly comply, but defend what it believes in, only updating the plan when both models agree."
"You can try this without any custom tooling. Just ask your agent to share its plan with a second model and hash it out before coming back to you. If you want to use Timo's skill, check it out here or use npx skills add timolins/claude-review to install."
"It was easy to get into a flow state in traditional design tools. We spent hours of focus time going broad and deep with creative juices. But now, we send a prompt and sit there waiting for the agent to complete a series of tasks. Before we know it, we're doomscrolling Slack. A few people on the team have landed on ways to use parallel threads to get into a flow state."
Product designers use a wide variety of tools even within a small team. AI usage ranges from frequent daily use to skepticism about usefulness for creative work. AI tooling for design workflows lags behind AI tooling for engineering workflows and can feel clunky and slow. One approach pairs coding agents by having Codex generate a plan and then having Claude review it for complexity, duplication, and API design quality, with updates only when both models agree. Another approach targets flow state by using concurrent threads, since waiting for sequential agent tasks can lead to distraction and Slack doomscrolling.
Read at Hannahhearth
Unable to calculate read time
[
|
]