Building Banani: How a Canvas-First AI Designer Is Raising the Floor on Product Design

This episode of Just Now Possible features the founders of Banani, a canvas-first AI product designer aimed at giving solo founders and small teams access to tasteful UX. The conversation covers how the company grew from a Figma plugin proof-of-concept into a tool producing hundreds of thousands of designs weekly. It digs into agent architecture choices such as surgical screen edits rather than full regeneration, per-screen context management across large canvases, and the "gulf of specification" between visual designers and text-based agents. The discussion matters because it offers a concrete look at designing AI-native creative tools where the human stays in control while the agent handles production work.

01Key takeaways

• Build AI design tools around a canvas so designers keep spatial control over outputs.
• Make surgical, per-screen edits instead of regenerating entire designs to preserve what already works.
• Maintain per-screen history alongside shared project context to handle canvases with hundreds of frames.
• Recognize the gulf of specification between visual intent and text prompts, and design tooling to close it.
• Evaluate design quality by side-by-side comparison of multiple generated outputs when formal evals fall short.
• Decide deliberately which model limitations to engineer around now and which to wait for as models improve.

02Key sections

Origins and early distribution

The team started with a Figma plugin prototype and learned from organic early adoption. Their path shows how a small tool can find users before a full product exists.

Canvas-first design over chat

Banani is built around a design canvas rather than a chat interface, letting designers see and steer work spatially. This choice shapes how the agent edits and presents output.

Agent architecture and surgical edits

Prompts are split so the agent can make targeted changes to individual screens instead of regenerating whole designs. Parallel edits and per-screen context keep large canvases manageable.

The gulf of specification

Designers think visually while agents understand text, creating a gap in how intent is expressed. The team is building tools to bridge that gap and help both sides share a visual language.

Evaluating taste without traditional evals

Rather than formal benchmarks, the team compares models by generating ten screens from one prompt and judging the outputs. They also decide which model limitations to work around versus wait out.

03From the post

“Listen to this episode on: Spotify | Apple Podcasts What if the future of product design isn't about replacing designers — it's about giving every team access to one? For solo founders, stretched design teams, and early-stage startups, great UX has always been out of reach. Banani is trying”

04Frameworks mentioned

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.