Skip to content
How I AIHow I AI

Gamma’s head of design on using AI to synthesize feedback and generate on-brand imagery | Zach Leach

Zach Leach, head of design at Gamma, reveals how his small team uses AI to analyze global feedback, create on-brand imagery, and maintain design quality while serving users in more than 60 countries. *What you’ll learn:* 1. How Gamma analyzes feedback from their 60% international user base using ChatGPT’s deep research capabilities 2. How to transform hundreds of multilingual feedback items into actionable design insights 3. A simple workflow for creating on-brand imagery using Midjourney-style references 4. How to use AI to maintain brand consistency across a globally distributed product 5. The secret to removing image backgrounds instantly using Replicate 6. How to create consistent, high-quality job descriptions in minutes using AI templates *Brought to you by:* WorkOS—Make your app enterprise-ready today: https://workos.com?utm_source=lennys_howiai&utm_medium=podcast&utm_campaign=q22025 Retool—AI that’s designed for developers and built for the enterprise: https://retool.com/howiai *Where to find Zach Leach:* LinkedIn: https://www.linkedin.com/in/zleach X: https://x.com/thisiszach *Where to find Claire Vo:* ChatPRD: https://www.chatprd.ai/ Website: https://clairevo.com/ LinkedIn: https://www.linkedin.com/in/clairevo/ X: https://x.com/clairevo *In this episode, we cover:* (00:00) Intro (02:42) Building the Gamma AI image editing feature (05:25) Using ChatGPT’s deep research for feedback analysis (09:10) How feedback was analyzed before AI tools (10:10) Benefits of deep research vs. basic scripting (12:40) Insights from ChatGPT's deep research (16:41) Demo of Midjourney workflow for creating on-brand art (23:54) Using Replicate for background removal (25:40) Style references (SREF) and brand consistency in Midjourney (29:19) An AI workflow for creating consistent job descriptions (32:27) Conclusion and final thoughts *ChatGPT feedback prompt:* “This is some feedback we’ve received about our AI image editing feature. I want you to analyze the feedback and find where we are doing poorly and where we are doing well. Break down for our product team what kinds of things we are doing well and why, and what kinds of things we are doing poorly and why. What do people love? What do people hate? Where can we improve?” *Tools referenced:* • Gamma: https://gamma.app/ • ChatGPT: https://chat.openai.com/ • Midjourney: https://www.midjourney.com/ • Midjourney Style Reference (SREF): https://docs.midjourney.com/hc/en-us/articles/32180011136653-Style-Reference • Replicate: https://replicate.com/ • Figma: https://www.figma.com/ • Claude Projects: https://claude.ai/projects • GPT 4o image model https://openai.com/index/introducing-4o-image-generation/ *Other reference:* • LaunchDarkly: https://launchdarkly.com/ _Production and marketing by https://penname.co/._ _For inquiries about sponsoring the podcast, email jordan@penname.co._

Claire VohostZach Leachguest
Jun 8, 202536mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Gamma design leader uses AI for feedback synthesis and branding

  1. Zach Leach (Head of Design at Gamma) shows how he uses AI as a research assistant to digest large volumes of messy, multilingual product feedback—about 550 responses in a week—without manually translating or sampling only a small subset.
  2. He demonstrates using ChatGPT “deep research” to translate, summarize, and classify feedback by themes (what’s working, what isn’t, prompt patterns), then exporting those classifications for charts and team-ready reporting.
  3. On the brand side, he walks through a Midjourney-based workflow that uses style references/personalization to rapidly iterate toward consistent, art-directed imagery for UI “empty states,” plus Replicate models for fast background removal before dropping assets into Figma.
  4. Finally, he shares a lightweight Claude “project” workflow to standardize job descriptions across hiring managers, and reflects on what human designers should keep owning: fun, personality, and craft details.

IDEAS WORTH REMEMBERING

5 ideas

Deep research turns unstructured feedback into usable product signals.

Instead of hand-reviewing a tiny sample, Zach uploads a feedback file and lets ChatGPT deep research translate, summarize, and identify what’s working/not working, including prompt-level patterns across many languages.

Quality of analysis improves when moving beyond keyword scripts.

Zach contrasts deep research with a prior approach: Python keyword matching (or row-by-row prompting). Deep research provided more nuanced classification and insight than basic text queries.

Feedback classification becomes a pipeline, not a one-off summary.

After deep research, he asks for per-row categorization and exports it into a spreadsheet to graph outcomes (e.g., which operations perform best/worst), enabling ongoing product tracking and prioritization.

Model tier differences show up in measurable satisfaction gaps.

Gamma compared paid vs free user feedback because different model access can change outcomes; they observed ~5% rating difference, reinforcing that model choice materially affects UX.

User complaints can directly inform UX “guardrails” for prompting.

A key issue was multi-step edits failing (the model completes only part of a compound request). Zach suggests UX responses like follow-up questions, automatic prompt splitting, or nudges toward step-by-step edits.

WORDS WORTH SAVING

5 quotes

Over the course of a week, we got about 550 individual responses.

— Zach Leach

If I’m being totally honest, I probably would’ve hand-looked at maybe 20.

— Zach Leach

It sort of went through all the feedback, understood what’s working, what’s not working, what prompts work, what don’t work.

— Zach Leach

It’s almost like I can follow those rabbit holes of creativity.

— Zach Leach

I would hope that it’s never gonna be as good as making things fun.

— Zach Leach

Multilingual feedback at scaleChatGPT deep research for synthesis/classificationFrom insights to internal presentations (Gamma)AI image editing UX learnings (multi-step failures, artifacts)Midjourney style references for brand consistencyReplicate models for background removalClaude Projects for reusable job-description templates

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome