How SpaceXAI designers use Grok Bot and Figma MCP to ship faster

This episode of "How I AI" features Claire Vo interviewing John and Pong from the Grok Bot design team at SpaceX AI. The discussion centers on how AI agents can fundamentally change workflow efficiency and creative practice for designers.

Pong begins by sharing a personal project that evolved into a sophisticated check-in system. After photographing a produce store in San Francisco's Chinatown, he decided to rebuild his personal website with a twist: using Grok Bot to create an automated pipeline that turns location photos into 3D miniature-style renderings. The workflow now allows him to simply send a photo, screenshot, or place name, and the bot handles everything—looking up coordinates via Google Places API, processing images in light and dark versions, removing people from the frame, and generating 3D clay-style visuals. He later expanded this to include companion information by sending screenshots of friends' social profiles, and the bot automatically structures everything for his website. This entire system was built without a detailed spec or Figma mockup; instead, Pong worked iteratively with the bot, using a "trust the vibe" approach that revealed possibilities as they built.

John demonstrates how he uses Grok Bot to accelerate professional design work at SpaceX AI. His first example shows Figma Bro, a specialized bot template he created for production design tasks. While at the gym, he was asked to match icons across Figma artboards—a normally tedious task. Using voice recording, he described the problem to the bot, which took screenshots of the icon library and existing work, then applied consistent scaling and positioning across all instances. This allowed work to continue asynchronously while John finished his workout. His second example involved creating marketing materials for a new bot marketplace. Again using a template and linked Figma files, John prompted the bot to apply consistent branding, backgrounds, and styling across multiple marketing assets—work that would normally require manually updating each design instance.

John also describes Devbot and Experiments, bots he uses to prototype interaction ideas that come to him on the go. When he had an idea for a screenshot effect on the share button of their Slack search agent, he recorded a voice memo describing the desired interaction, and the bot created a working prototype with two different approaches—all without John touching the computer. Though they ultimately didn't ship the feature, the low-cost nature of prototyping meant it was worth exploring without organizational friction.

The conversation highlights a broader philosophical shift in design practice. Claire notes that AI removes the gatekeeping of specialized skills—she describes how she always wanted 3D rendering capabilities but previously would have needed to ask someone with that expertise. Both designers echo that AI unlocks previously inaccessible skill sets, allowing them to be more well-rounded creators. John mentions using AI for motion design and Pong for 3D visualization, capabilities they describe as previously out of reach.

Claire articulates the "trash can method" of software development—the ability to build, test, and discard ideas cheaply without organizational overhead. Previously, pitching a feature idea meant navigating product managers asking for ROI justification and engineers concerned about tech debt priorities. Now, designers can explore ideas freely, knowing that abandoning them carries no real cost.

In the lightning round, Pong reveals his extensive bot ecosystem organized into "life" and "work" buckets. He describes bots serving as a chief of staff (handling shopping and inventory management), email manager (surfacing action items), calendar manager (he rarely opens the calendar app anymore), and a personalized news bot that delivers AI-relevant articles each morning. He even uses a bot for visa and green card waitlist checking, eliminating manual monthly status checks. John describes learning Japanese with a bot that provides hiragana, katakana, and kanji forms simultaneously, and an insurance bot that analyzes receipts and claims to ensure he's maximizing his coverage benefits.

When asked how they handle frustration with their bots, both acknowledge the sensitivity of bot interactions being logged (relevant in a SpaceX AI environment), so they've become gentler over time. They share the practical tip of asking bots to memorize successful workflows so they improve with repeated use—essentially having bots learn from past interactions.