How I AI: Gumroad CEO Sahil Lavingia’s Workflow for Building Features 40x Faster

Today, I got to sit down with Sahil Lavingia, the founder and CEO of Gumroad. If you're not familiar with Gumroad, it's the platform that has helped creators sell over a billion dollars' worth of products directly to their audiences.

Sahil is doing some of the most interesting work I've seen when it comes to integrating AI into the software development process. At Gumroad, the AI engineering agent Devin is already writing 41% of their pull requests, and they're aiming for 80% by the end of the year. This is a huge shift in how they operate. Sahil describes it as a 40x productivity increase—turning tasks that used to take two weeks into something that can be done in just two hours.

In our conversation, Sahil showed me his exact process for how this works. He walked me through identifying a user experience issue, prototyping a solution with AI, handing it off to an AI agent to build, and getting it merged into production. We also looked at a second, more common workflow: using AI for quick maintenance tasks and code improvements.

This episode is packed with practical advice for any product leader, engineer, or founder who wants to understand the concrete steps it takes to build faster and smarter with AI. We get into the specific tools, the exact prompts, and the kind of team culture you need to make this happen.

Workflow 1: From Idea to Shipped Feature in Under an Hour

One of the coolest things about these new AI tools is how quickly you can go from spotting a problem to shipping a fix. Sahil demonstrated this for me live by tackling a small but annoying UX issue in one of their internal products, Flexile: a clunky, native browser date picker on the contractor invitation page.

"Can you do something that used to take two weeks in two hours? And that's like a 40 times speed increase. So that's kinda like the number that I have in my head."

Instead of writing a detailed spec, creating a ticket, and waiting for a designer and engineer, Sahil showed me how he can solve the problem himself in one sitting. Here’s how he does it.

Step 1: Prototyping and Iterating with v0

He starts in v0, a generative UI tool that creates interactive prototypes from text prompts. Sahil’s goal was to replace the native date picker with a better component, inspired by the library shadcn/ui.

He didn't try to get it perfect on the first go. Instead, he used it to iterate and refine his ideas in real time, almost like he was brainstorming with the AI to figure out exactly what he wanted.

1. Initial Idea: He noticed the date picker in Flexile was basic and wanted something more human-friendly, like the examples from shadcn/ui.
2. Iterative Prompting: Sahil started building out the form in v0, adding more features as he went. He realized he didn't just want a better calendar; he wanted natural language input.
3. The "Dope" Prompt: After a few iterations, he landed on a much more ambitious and descriptive prompt that captured the full user experience he was aiming for.

build a really dope natural language day picker for an HR product onboarding form

The result was a great-looking, functional React component that included smart suggestions like "Next Monday" and "In two weeks"—features perfectly suited for its HR context. This prototyping phase, which might have taken days of back-and-forth with a designer, was done in minutes.

Step 2: Implementation with Devin, the AI Engineer

Once the prototype looked good, it was time to get it into the actual codebase. That's where Devin, the AI software engineer from Cognition, comes in. Instead of handing off a Figma file, Sahil hands off the prompt.

Passing the Prompt: He takes the final, refined prompt from v0 and gives it directly to Devin, specifying the repository and the page to work on.

"Normally I would take like the final prompt and I would just paste that into Devin... build this form."

Autonomous Execution: Devin gets to work. It sets up the development environment, identifies the relevant files, installs any necessary dependencies (like the shadcn/ui component), writes the new code, and replaces the old date picker.

Full Transparency: The entire process is recorded, so Sahil can watch a time-lapse of every command Devin runs, every file it edits, and every thought process it logs. This is really important for debugging and understanding how it arrived at the solution.

Step 3: Review and Merge

Within minutes, Devin creates a pull request on GitHub. The final step is a human review. In this case, the AI-generated code was clean, well-structured, and included a clever parseNaturalLanguage function that handled inputs like "next Monday" or "tomorrow."

Sahil noted that if any fixes were needed, he would typically open the code in Cursor, an AI-native code editor, to make final tweaks. However, with Devin's new pairing mode, he can now jump directly into the agent's environment to collaborate.

During our recording, the PR for the initial, simpler shadcn/ui date picker was created and merged. The entire workflow, from identifying a UX flaw to shipping a fix, happened in real-time. This is what that 40x speed increase actually looks like.

Workflow 2: Quick Maintenance with Devin on Slack Recaps

Not every task is a full-feature build. Sahil also showed how he uses Devin for smaller, day-to-day maintenance tasks that improve internal processes. This is a great way to fix those small annoyances without distracting the engineering team from larger priorities.

The Problem: Gumroad has an AI-generated weekly recap that gets posted to Slack. Sahil noticed a couple of minor issues:

• It was listing project categories that had no updates, creating unnecessary noise.
• It was including backend-only changes that weren't user-facing "shipments."

Here’s how he used Devin to fix it:

The Prompt: Sahil assigned the task to Devin with a clear, simple instruction.

"Hey, at Devin, like, could you like, you know, only show the projects that actually have shipments and like hide the other ones... and also, like some of these aren't really shipments... make sure you know the pro update, the AI prompt that we're using for this."

Devin's Work: Devin located the script responsible for the Slack recap. It correctly identified two things that needed to be changed:

• The code that filtered the projects.
• The system prompt used by the LLM that generates the summary text.

The Result: Devin submitted a pull request that added a .filter to the code to only include projects with more than one item and updated the LLM prompt to focus on user-facing features. A human engineer could then quickly review and merge the change.

This workflow is a perfect example of using AI for continuous improvement. It allows Sahil to fix small annoyances as he sees them, keeping internal tools sharp and efficient without creating overhead for the team.

How Gumroad Drives AI Adoption: Bounties and Culture

Having the right tools is one thing; getting an entire team to use them is another. Sahil has been intentional about creating a culture that embraces AI, using a mix of leadership by example and financial incentives.

He shared the details of a recent competition at Gumroad:

• The Challenge: A $33,000 prize pool was split amongst any engineer who opened and merged more pull requests using Devin than Sahil himself during the month of May.
• The Outcome: Three engineers beat him, and he came in fourth with 27 PRs. The competition successfully gamified the learning process and motivated the team to integrate Devin into their daily work.

This approach, along with Sahil sharing his own recordings of his AI-driven workflows, helps take the mystery out of these tools. He's showing his team what's possible and making it fun and rewarding to learn, instead of just telling them to use AI.

Conclusion: The Future of Building Products

Talking with Sahil felt like getting a peek into the near future of product development. The workflows he demonstrated aren't science fiction; they are practical, repeatable processes that are delivering real results today. The big takeaway for me is that AI is changing who can build and what they can focus on. It's about more than just speed.

When implementation becomes this fast and cheap, the bottleneck shifts from engineering capacity to the quality of our ideas and the depth of our user understanding. We can afford to spend more time in v0 or Figma, obsessing over the user experience, because we know an AI partner like Devin can handle the execution. This frees up engineers to focus on more complex challenges: architecture, system design, and removing the tech debt that prevents AI from working effectively.

Sahil is showing what a high-velocity product organization can look like. I encourage you to watch the full demo and start experimenting with these tools. The 40x speed increase he talks about isn't a myth—it's waiting for teams who are willing to embrace a new way of building.

Thanks to Our Sponsors

A huge thank you to our sponsors who make this show possible:

• Enterpret — Customer SuperIntelligence Platform for Product and CX teams
• Vanta — Automate compliance and simplify security with Vanta.

Episode Links

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