Welinder says founders see the whole company as their responsibility, while product leaders often focus more narrowly on product strategy and managing PMs. She argues that product work should be cross-functional in practice: PMs should do user research, design, and as much coding as they can.
A Founder's Playbook for Shipping 10x Faster with AI
Speaker
Yana Welinder, head of AI at Amplitude and former founder, talks with Peter Yang on his podcast about shipping faster with AI, learning from users, and putting customer feedback alongside product analytics.
Welinder says she would rather execute on several options than spend too long debating which one is best. At Amplitude, CEO Spencer Skates banned decisions by committee to help teams take ownership and move faster. She recommends demonstrating a faster approach through real launches, so colleagues can see what works.
Welinder says a rough first release can build trust when teams respond quickly to customer feedback. She describes fixing reported issues within 15 minutes and keeping users engaged with prompt updates and follow-up releases. Teams can start with a small group, improve the product, and expand access as it becomes more polished.
Welinder iterates on prototypes between customer calls, incorporating one person's feedback before showing the next version. She also uses Sora videos in marketing and says some held viewers through to the end at rates of 50% or more. With ChatGPT Atlas, she triages email, drafts replies, and summarizes emotionally charged messages; she says AI-generated prompts and evals still need careful human review.
Welinder demos Amplitude Feedback, which brings support tickets and public reviews together with quantitative product analytics. Teams can find frequent requests, inspect what users said, create cohorts to study their behavior, and generate a draft PRD from feedback. She says accuracy remains essential for analytics, where even small data errors can undermine trust.
Takeaways
- Give teams clear ownership so they can test ideas without waiting for committee decisions.
- Ship to a small group, listen closely, and follow up quickly when users report problems.
- Use AI to revise prototypes between customer conversations and test each iteration with users.
- Treat AI-generated prompts as drafts: run evals and have a person check whether changes improve results.
- Pair qualitative feedback with behavioral data to understand both what users say and what they do.
Mentioned in this talk
Search inside this talk
What was claimed 12
- ceo ofSpencer Skates CEO of
Amplitude“CEO and co-founder of Amplitude”
0:37inHow Amplitude Went From Skeptics to “All In” on AIYouTube Verified2 sources· Talk
- role atYana Welinder is Head of AI at
Amplitude“now head of AI at Amplitude”
YouTube Unverified1 source· Talk
- foundedYana Welinder founded Grapple
“before starting Grapple”
YouTube Unverified1 source· Talk
- acquired
“we were also acquired for the team”
YouTube Unverified1 source· Talk
- featureSpencer Skates: banned to enable PMs and engineers to own decisions and ship fasterbanned to enable PMs and engineers to own decisions and ship faster
“has recently banned decisions by committee”
YouTube Unverified1 source· Talk
- usesYana Welinder uses
Soramarketing videos“using Sora a lot in marketing collateral”
YouTube Unverified1 source· Talk
- useswriting, summarizing, and interacting with her mailbox
“I use it for writing everything, summarizing, interacting with my mailbox”
YouTube Unverified1 source· Talk
- metric
GPT-5: Prompt it generated worked as well as a human-written prompt for GPT-4.1, but only when used with GPT-5“only with GPT-5”
YouTube Unverified1 source· Talk
- feature
GPT-4.1: Non-reasoning model used in an analysis pipeline for speed
“we use GPT-4.1, a non-reasoning model, for speed”
YouTube Unverified1 source· Talk
- featureAmplitude Feedback: Connects feedback sources, prioritizes requests, and links feedback to cohorts, session replays, and surveys
“show both quantitative and qualitative data in one place”
YouTube Unverified1 source· Talk
- integrates withAmplitude Feedback integrates with
Amplitudecombines quantitative and qualitative data in one place“we've been integrating Craft into Amplitude”
YouTube Unverified1 source· Talk
- usesAmplitude Feedback uses MCPmakes product feedback data available in Amplitude, Claude, or ChatGPT
“What I just demoed is in our MCP”
YouTube Unverified1 source· Talk










