Check out the conversation on Apple, Spotify and YouTube.
Brought to you by:
Maven: Improve your PM skills with awesome courses. Discount with my link
Pendo: the #1 Software Experience Management Platform
Vanta: Automate compliance across 35+ frameworks
NayaOne: Airgapped cloud-agnostic sandbox
Kameleoon: AI experimentation platform
Today’s Episode
Laura Burkhauser started as an IC PM at Descript, the $550M AI video editing platform.
Three years later, she’s CEO.
She shipped AI features that worked (and ofc was an awesome leader).
In today’s episode, she walks through the exact features the team built, the eval frameworks used, and her complete IC → CEO path.
The coolest part? You can get a free 12 months of Descript’s paid plan by becoming an annual subscriber. (You get access to Descript + 8 other premium AI product tools I recommend.)
P.S. I have renamed the podcast to The Growth Podcast with it’s own fiery new branding.
Your Newsletter Subscriber Bonus
For subscribers, each episode I also write up a newsletter version of the podcast. Thank you for having me in your inbox.
Today’s guide covers:
Building AI Tools That Users Keep Using
A deep dive into Descript’s AI Agent
Her IC to CEO Career Path
1. Building AI Tools That Users Keep Using
1a. Pick Features Based on Real Pain Points
Descript understood their customers from years of watching people edit videos.
When LLMs became available, they asked: What just became possible?
For script readers who looked away from camera, they integrated an “eye contact” model.
For people who recorded the same line 5 times, they built Remove Retakes.
These features instantly worked.
The takeaway? Map the user journey. Find the pain points. Check what LLMs newly enable. Build that intersection. This is good AI product strategy.
1b. Build Buttons, Not Chat Boxes
Descript then put these features into prepackaged buttons: Edit for Clarity, Remove Filler Words, Remove Retakes, Add Chapters.
Not a blank chat box. Specific buttons for specific jobs. Behind each button: a prompt that delivers reliable results every time.
But the art of product building is in the details:
Early on, the context windows were too small for full transcripts. So, they devised a solution to sent content in chunks.
They caught the nuance that if you have more than 1 jump cut per ten seconds, you look insane on camera. .
That’s good AI product building in action. And this comes from deeply understanding your customer.
1c. Roll Out With Human Evals First
Before all the thought pieces on evals, Descript did it simply.
They tested on production data. They asked themselves:
“Would I use this as a customer?”
If yes, it shipped. If no, it didn’t. For straightforward features, they even skipped private beta and went straight to public.
They measured adoption and retention. The baseline was Remove Filler Words, their first AI feature and very popular.
The ultimate metric: Did users export their video WITH the AI edits? If they removed the edits before export, it didn’t meet their quality bar.
2. A deep dive into Descript’s AI Agent
Eventually, Descript has rolled out an all-purpose agent. But it works in a very specific way that’s worth understanding.
Their agent, Underlord, works across all Descript features, not just one area.
This required context (understanding user needs), tool coverage (access to all tools), and an eval system.
The rollout started with regression tests. Then private alpha with real customers they needed people using natural language, not “Descript employee language.”
They brought in people with different skill levels. Some sophisticated at video and AI, some not. This helped them understand prompts like “edit this video and make it better.”
They built regression tests from alpha queries, organized by categories. Then tested: Is Underlord better at activating new users than the current experience? Testing AI features like this requires both technical rigor and customer empathy.
The result was yes, so they gave it to all new users and let existing users opt in.
3. Her IC to CEO Career Path
3a. Earn Founder Trust Through Command
Why should a founder trust you more than themselves?
They built the first 7 versions by trusting their instincts. Andrew Mason, the founder of Descript, also created Groupon.
Laura’s answer was to get command. Use the product extensively. Talk to customers. Understand the business.
When you speak, people should think: “Smart.”
Laura was hired as an IC PM. Andrew liked having her in strategic conversations. She got invited to more rooms, and, eventually, she belonged in THE room.
What kills this is when the founder gives you latitude → You ship something misaligned → Founder micromanages → Death spiral
3b. Ship Features Before Doing Strategy
One trap many aspiring CEOs fall into is they get invited to product strategy meetings. They love it. They do more strategic work.
And then… their IC work falls apart.
Laura’s solution:
“Be deep in IC work and excel at that before you’re invited to plan next year.”
You need both: ship features AND show strategic thinking. You can’t skip the first to focus on the second.
This applies at all levels - even head of product!
3c. Cold email for the win
Laura’s path wasn’t linear:
German Literature major → Management consulting → Built learning platform → Amazon (too big) → Series C startup (addicted to this feeling) → Twitter (learned org leadership) → Started podcast → Found Descript → Cold emailed them.
So how did she end up eventually making?
She cold emailed twice in her career: once for a learning platform, once for Descript.
Both times:
“I love your product. Give me a job.”
If you’re wondering how to get your dream job, start with an email.
Don’t miss the clips channel for shorter versions of the key moments.
Where to Find Laura Online
I’m coaching 100 PMs to their dream job from Feb-Apr 2026. Grab a spot early for a discount.
Related Content
Podcasts:
Newsletters:
PS. Please subscribe on YouTube and follow on Apple & Spotify. It helps!
