How I AI: Gamma's 3-Step AI Workflow for Global Feedback, Art Direction, and Hiring

For this episode, I was really excited to sit down with Zach Leach, the Head of Design at Gamma. Gamma is an AI-native product for creating presentations, but what’s especially interesting is how global their user base is. They’re a small team of around 30 people serving a huge number of users—60% of whom are international and don’t speak English.

That setup presents a really interesting challenge: how does a small team stay connected to such a diverse, global user base and still maintain an incredibly high bar for brand and product quality? As you'll see, Zach and his team have come up with some clever AI workflows to solve this. He's effectively using AI as his own personal data researcher, art department, and hiring coordinator. That frees him up to focus on the craft, care, and fun that make Gamma's user experience feel so special.

I’ve been in this industry for a while, and I’ve never met a design team that felt they had enough research capacity. It's often the first thing that gets cut or under-resourced. What Zach is doing shows that it’s now possible to get large-scale, deep user insights in minutes, not months. We're going to walk through three of his specific workflows: how he analyzes hundreds of pieces of multilingual feedback with ChatGPT, how he generates on-brand art with Midjourney, and how he standardizes hiring with Claude. Let's get into it.

Workflow 1: Analyzing Global User Feedback with ChatGPT's Deep Research

One of the first things Zach showed me was how Gamma handles user feedback for a new feature—their AI image editor. This tool lets users chat with an AI to modify images directly within their Gamma presentations. To improve it, they collect user feedback, but with 60% of their users being international, that feedback comes in a dozen different languages. Over just one week, they collected 550 individual responses. For a small team, manually translating and categorizing that much feedback just isn't feasible.

Before AI, Zach admitted he probably would have just cherry-picked 20 or so English responses and tried to spot trends. Now, his approach is completely different.

Step 1: Upload Data to ChatGPT

The workflow starts by simply exporting all 550 pieces of feedback into a single file. Zach then opens ChatGPT and uploads the file directly. This is where he taps into a really useful, but often overlooked, feature in ChatGPT: its ability to do deep research on files.

Step 2: The Deep Research Prompt

He doesn't just ask for a summary. He gives it a detailed prompt designed to get a product-focused breakdown, explicitly asking ChatGPT to act as an analyst for the product team.

“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?”

ChatGPT then gets to work. Zach said it took about 19 minutes, but the result was a comprehensive analysis that went far beyond basic keyword matching. It translated feedback from Turkish and other languages, identified common themes, and categorized the sentiment of each response.

Pro Tip: Zach first tried this without specifying 'deep research' and ChatGPT just wrote a simple Python script for keyword matching. By using a more direct prompt and letting the model take its time, he got a much deeper, more insightful analysis that truly understood the nuances of the feedback.

Step 3: From Analysis to Actionable Presentation

The analysis ChatGPT returned was incredibly detailed. It highlighted what users loved (like image upscaling) and what they hated (like generating extra arms or failing on multi-step edits). It pulled out direct quotes with translations and provided a high-level summary.

To make this analysis immediately useful for the team, here's where the workflow comes full circle: he copied the entire text output from ChatGPT and pasted it directly into Gamma. With a simple instruction to "use charts and graphs," Gamma automatically generated a polished, shareable presentation summarizing the research findings. This deck became the basis for a conversation with product and engineering, leading to real roadmap decisions, like designing a better UX to handle complex, multi-step prompts.

---

Workflow 2: Scaling High-Craft Art Direction with Midjourney and Replicate

Gamma recently went through a beautiful rebrand. Their visual style is imaginative, airy, and surreal—the kind of art direction that would traditionally require a dedicated art department to maintain consistency and quality. For a lean startup, that's a huge expense. Zach showed me how he uses Midjourney as his on-demand art department, allowing him to create great-looking, on-brand assets in minutes.

Step 1: Establishing a Brand 'Kit' in Midjourney

The key to consistency is having a well-defined starting point. Working with their creative director, the Gamma team developed a set of styles within Midjourney. They use the Style Reference (`--sref`) feature to create a kind of brand 'kit.' By including these style references in their prompts, anyone at the company can generate images that feel instantly on-brand.

Step 2: Rapid Creative Iteration

Zach walked me through creating an illustration for an 'empty state' in the new AI image editing feature. This is where things get really fun. Instead of briefing an illustrator and waiting days for a first draft, Zach could explore creative rabbit holes in real-time.

1. He started with an idea of a painting. It was weird, but not quite right.
2. He pivoted to a person chatting. Still not there.
3. He thought about transformation, prompting for an apple, half green, half red.
4. A happy accident occurred—Midjourney randomly included a bird in one of the apple images. This sparked a new idea.

He then dove deep into the bird concept, rapidly iterating and refining his prompt. We scrolled through probably a hundred variations as he honed in on the final concept: a vertically split image, half bird, one half is this, one half is that. It was amazing to watch him follow his creative intuition without any friction. It’s a process that would have taken weeks and been miserable for both client and agency in the 'before times.'

Step 3: The Final Polish with Replicate

Once he landed on the perfect image, there was one final step. The image had a background, but he needed a transparent version to use in the UI design in Figma. In the old days, I would have painstakingly traced the object with the pen tool. Zach has a much slicker way of doing it.

He uses Replicate, a platform for running machine learning models. He found a model specifically trained for high-quality background removal. He simply uploaded his bird image, and in seconds, he had a perfect, clean PNG with a transparent background ready to be dropped into his final design in Figma. This final step allows him to add that extra layer of craft—like having the bird pop out of its container—that makes the user experience feel so polished and delightful.

---

Workflow 3: Standardizing Job Descriptions with Claude Projects

Finally, Zach shared a workflow that every hiring manager needs to see. Writing job descriptions is a slog, but getting them right is critical for attracting top talent and communicating your company culture. To ensure consistency and quality across the company, the Gamma team built a simple but really effective system using Claude Projects.

Step 1: Setting up the Claude Project

This is a classic 'few-shot' prompting technique. Someone on the team created a new Project in Claude and uploaded a few of Gamma's best-written, existing job descriptions. They then added a simple instruction: use these examples as the template for any new job descriptions.

Step 2: Generating a New Role

Now, any hiring manager at Gamma can go into that Claude Project and ask for a new job description with a simple prompt. To show me how well it works, Zach asked it to create a posting for a fictional role:

make a job for head of popcorn

Claude instantly generated a complete job description that perfectly matched Gamma's tone, structure, and values. It included sections on hybrid work, company mission, and even had a little fun with the role's responsibilities and qualifications. My personal favorite was this requirement: "Five years of experience in professional popcorn production with a strong emphasis on kernel-driven solutions."

Step 3: Publishing with Gamma

While the generated text is about 80% of the way there, it's a massive head start. The hiring manager can make a few tweaks and then, just like with the research findings, copy the text and paste it into a pre-made Gamma template to instantly create a beautiful, on-brand careers page for the new role. This workflow is a triple win: it saves a ton of time, reinforces brand and quality consistency, and is reusable for the entire team.

---

From Efficiency to Craft and Fun

What I love most about Zach's workflows is that they're about more than just efficiency. He's using AI as a creative partner, which lets him focus on what truly matters: the craft, the details, and the fun of the user experience. By offloading the tedious work of data analysis, image creation, and administrative writing, he and his small team can deliver a world-class product that feels thoughtful and cared for.

These workflows show how AI can help lean teams operate at a global scale without sacrificing quality. Whether it's staying closer to your customers, scaling your brand, or building your team, these tools can help you focus on the human element of design. I hope this inspires you to try some of these workflows in your own work!

Thanks to our Sponsors

• WorkOS—Make your app Enterprise Ready today
• Retool—AI that's designed for developers, and built for the enterprise

Episode Links

• Watch on YouTube
• Listen on Spotify
• Listen on Apple Podcasts