Pricey subscribers,
In the present day, I need to share a brand new episode with Claire Vo.
Claire is the Chief Product Officer of LaunchDarkly, founding father of ChatPRD, and host of the How I AI podcast. No product chief makes use of AI as a lot as Claire so I used to be thrilled to get her to demo how she builds a brand new function with AI brokers and the way she’s advising PMs to adapt to AI within the subsequent 18 months.
Watch now on YouTube, Apple, and Spotify.
This episode is delivered to you by Vanta — Be part of 9K+ firms like Atlassian who use Vanta to handle threat and show safety in real-time. Get $1000 off at vanta.com/peter
Claire and I talked about:
-
(00:00) Why conventional PM is dying (and what’s coming subsequent)
-
(01:46) Dwell demo: Constructing a brand new function utilizing ChatPRD and Devin
-
(05:34) The prompting trick that makes AI work like an actual crew member
-
(08:08) Why this instrument has changed Cursor for 70% of Claire’s coding wants
-
(21:05) How Claire makes use of AI to construct product technique
-
(28:30) Enjoying with the brand new function dwell
-
(34:22) Balancing CPO duties, facet initiatives, and household
-
(42:43) 3 abilities that may future-proof your PM profession within the AI period
-
(50:05) Why we’re getting into the yr of multiplayer AI brokers
Let’s get proper into it. Are you able to present us the way you construct a brand new function with AI?
Positive, let me present you the way I construct a brand new function utilizing ChatPRD and Devin.
When customers chat with ChatPRD, they will price responses with a thumbs-up or thumbs-down. For thumbs down, we ask what may very well be improved with choices like “too lengthy,” “too quick,” or “inaccurate.” However for thumbs up, we simply say “thanks” with out studying what they favored. Let’s construct a function to class this optimistic suggestions.
First, I can ask ChatPRD instantly in Slack: “Assist me construct a PRD for receiving optimistic suggestions when individuals like a ChatPRD response. We do that for unfavourable suggestions, however I need it for optimistic suggestions too.”
One prompting trick I take advantage of is telling AI what sort of crew member will obtain this doc: “Write me a PRD I can ship to my engineering crew.” The AI tunes the content material for what that persona would possibly want.
Because the modal already exists in my code base, I will skip prototyping and ship this PRD on to Devin to jot down the code.
So you utilize Devin for coding slightly than Cursor or Windsor?
Sure, since Devin improved, I’ve dropped my Cursor utilization by 70%.
It is much less about code high quality and extra about person expertise – I am unable to all the time be synchronously working in an IDE. Yesterday, on my morning stroll, I obtained a help ping a couple of bug. I could not open an IDE on my telephone, however I might tag Devin to repair it. This asynchronous workflow is what makes the distinction.
Now that is what I’m going to inform Devin: “I need to seize optimistic suggestions the identical manner we seize unfavourable suggestions. We have already got a shared suggestions modal element you should use. Make up the optimistic suggestions classes and evaluate the connected PRD.”
You see that Devin acknowledged the message rapidly and created a 14-step plan. Subsequent, it’s going to submit a PR for me to evaluate.
Do you evaluate the pull request (PR) manually?
Completely. I’m reviewing it now. You see that Devin even up to date the analytics monitoring with out me asking, which is nice.
Now I’m going to check it domestically, and it really works! Once more:
What’s nice about Devin is the async workflow – I may be doing one thing else whereas Devin works, and it’ll Slack me when it’s prepared.
Okay I’ve to ask you this, how do you steadiness all the pieces that you just do?
