From Idea to Product in 30 Min Using AI Agents (Full Tutorial) | Claire Vo
A live demo of how Claire uses AI agents to build a feature end to end and how she's advising PMs to adapt to AI to avoid getting left behind in the next 18 months
Dear subscribers,
Today, I want to share a new episode with Claire Vo.
Claire is the Chief Product Officer of LaunchDarkly, founder of ChatPRD, and host of the How I AI podcast. No product leader uses AI as much as Claire so I was thrilled to get her to demo how she builds a new feature with AI agents and how she’s advising PMs to adapt to AI in the next 18 months.
Watch now on YouTube, Apple, and Spotify.
This episode is brought to you by Vanta — Join 9K+ companies like Atlassian who use Vanta to manage risk and prove security in real-time. Get $1000 off at vanta.com/peter
Claire and I talked about:
(00:00) Why traditional PM is dying (and what's coming next)
(01:46) Live demo: Building a new feature using ChatPRD and Devin
(05:34) The prompting trick that makes AI work like a real team member
(08:08) Why this tool has replaced Cursor for 70% of Claire's coding needs
(21:05) How Claire uses AI to build product strategy
(28:30) Playing with the new feature live
(34:22) Balancing CPO duties, side projects, and family
(42:43) 3 skills that will future-proof your PM career in the AI era
(50:05) Why we’re entering the year of multiplayer AI agents
Building a new feature with AI agents
Let's get right into it. Can you show us how you build a new feature with AI?
Sure, let me show you how I build a new feature using ChatPRD and Devin.
When users chat with ChatPRD, they can rate responses with a thumbs-up or thumbs-down. For thumbs down, we ask what could be improved with options like "too long," "too short," or "inaccurate." But for thumbs up, we just say "thank you" without learning what they liked. Let’s build a feature to category this positive feedback.
First, I can ask ChatPRD directly in Slack: "Help me build a PRD for receiving positive feedback when people like a ChatPRD response. We do this for negative feedback, but I want it for positive feedback too."
One prompting trick I use is telling AI what kind of team member will receive this document: "Write me a PRD I can send to my engineering team." The AI tunes the content for what that persona might need.
Since the modal already exists in my code base, I'll skip prototyping and send this PRD directly to Devin to write the code.
So you use Devin for coding rather than Cursor or Windsor?
Yes, since Devin improved, I've dropped my Cursor usage by 70%.
It's less about code quality and more about user experience – I can't always be synchronously working in an IDE. Yesterday, on my morning walk, I got a support ping about a bug. I couldn't open an IDE on my phone, but I could tag Devin to fix it. This asynchronous workflow is what makes the difference.
Now this is what I’m going to tell Devin: “I want to capture positive feedback the same way we capture negative feedback. We already have a shared feedback modal component you can use. Make up the positive feedback categories and review the attached PRD.”
You see that Devin acknowledged the message quickly and created a 14-step plan. Next, it’s going to submit a PR for me to review.
Do you review the pull request (PR) manually?
Absolutely. I’m reviewing it now. You see that Devin even updated the analytics tracking without me asking, which is great.
Now I’m going to test it locally, and it works! Again:
What's great about Devin is the async workflow – I can be doing something else while Devin works, and it will Slack me when it’s ready.
Balancing a CPO job with a 6-figure side hustle and a podcast
Ok I have to ask you this, how do you balance everything that you do?