Tobi Lütke: AI Agents, Better Decisions, and the Future of Work
Episode
65 min
Read time
3 min
Topics
Remote Work, Relationships, Startups
AI-Generated Summary
Key Takeaways
- ✓AI Workflow Integration: At Shopify, approximately 50% of production pull requests are now generated through conversations with River, an AI agent embedded in Slack across 10,000 channels and 7,000 employees. Engineers who still write code directly run 10–50 simultaneous agent instances. River has a distinct personality, can be sarcastic, pushes back on poor requests, and maintains per-person memory files updated nightly through a self-reflection process called "dreaming."
- ✓Multi-Agent Decision Framework: Lütke uses a structured AI council for high-stakes decisions: five to six specialized sub-agents (data, research, business, engineering perspectives) each analyze a problem independently, then a randomized synthesis agent consolidates outputs, and a final top-tier model delivers the conclusion. The entire process costs roughly $15–20 in tokens and completes in under 30 minutes — compressing what would otherwise take weeks of human research.
- ✓Slop Grenades — The Hidden Cost of AI: AI creates a new failure mode Lütke calls "slop grenades" — where low-effort, AI-generated outputs (bloated emails, unreviewed pull requests) get lobbed at colleagues, wasting review time and eroding quality standards. The solution is using AI to synthesize concisely at the point of creation rather than generating verbose outputs that others must then re-compress. Output volume is no longer a signal of effort or quality.
- ✓Optimal Paths Lack Feedback Loops: Lütke argues the correct strategic choice is frequently the one without an attached feedback loop. When multiple valid options exist, the one producing measurable short-term results pulls decision-makers toward it — not because it is best, but because it generates quarterly "attaboys." Founders with deeper company relationships can resist this pull; hired executives structurally cannot, which explains persistent corporate short-termism more accurately than intent or intelligence.
- ✓Affirmations as Behavioral Reprogramming: Lütke used written affirmations — specifically writing "I love public speaking about things that are interesting to me" repeatedly for roughly one week, five minutes daily — to overcome a genuine fear of public speaking. He applies the same principle at Shopify, banning the phrase "I'm not good at this" without appending "yet." The mechanism: repeated written statements reshape default mental grooves before conscious reasoning catches up.
What It Covers
Shopify CEO Tobi Lütke details how AI agents have restructured work at Shopify — where roughly 50% of pull requests now originate from AI conversations in Slack — while exploring decision-making frameworks, the future value of taste and judgment over technical skills, and why pruning systems matters more than adding to them.
Key Questions Answered
- •AI Workflow Integration: At Shopify, approximately 50% of production pull requests are now generated through conversations with River, an AI agent embedded in Slack across 10,000 channels and 7,000 employees. Engineers who still write code directly run 10–50 simultaneous agent instances. River has a distinct personality, can be sarcastic, pushes back on poor requests, and maintains per-person memory files updated nightly through a self-reflection process called "dreaming."
- •Multi-Agent Decision Framework: Lütke uses a structured AI council for high-stakes decisions: five to six specialized sub-agents (data, research, business, engineering perspectives) each analyze a problem independently, then a randomized synthesis agent consolidates outputs, and a final top-tier model delivers the conclusion. The entire process costs roughly $15–20 in tokens and completes in under 30 minutes — compressing what would otherwise take weeks of human research.
- •Slop Grenades — The Hidden Cost of AI: AI creates a new failure mode Lütke calls "slop grenades" — where low-effort, AI-generated outputs (bloated emails, unreviewed pull requests) get lobbed at colleagues, wasting review time and eroding quality standards. The solution is using AI to synthesize concisely at the point of creation rather than generating verbose outputs that others must then re-compress. Output volume is no longer a signal of effort or quality.
- •Optimal Paths Lack Feedback Loops: Lütke argues the correct strategic choice is frequently the one without an attached feedback loop. When multiple valid options exist, the one producing measurable short-term results pulls decision-makers toward it — not because it is best, but because it generates quarterly "attaboys." Founders with deeper company relationships can resist this pull; hired executives structurally cannot, which explains persistent corporate short-termism more accurately than intent or intelligence.
- •Affirmations as Behavioral Reprogramming: Lütke used written affirmations — specifically writing "I love public speaking about things that are interesting to me" repeatedly for roughly one week, five minutes daily — to overcome a genuine fear of public speaking. He applies the same principle at Shopify, banning the phrase "I'm not good at this" without appending "yet." The mechanism: repeated written statements reshape default mental grooves before conscious reasoning catches up.
- •Pruning Over Addition: Using SpaceX's Raptor engine iterations as a model, Lütke argues organizations cannot improve by layering additions onto existing systems. Departments and products need periodic "refounding events" — deliberate version-two restarts that discard accumulated sediment. Companies displaced in the early 2000s tech wave failed precisely because competitive pressure was absent, so problems were solved by addition rather than reconstruction, burying original intent under compounding layers.
Notable Moment
Lütke reframes superintelligence as already present: a city like Toronto functions as a superintelligence that individuals access daily through specialization and shared systems. He argues synthetic AI superintelligence will arrive the same way — gradually and undramatically — pointing out that the Turing test was solved around 2020 with no ticker-tape parades and no societal rupture.
Episode Transcript
Things need to be pruned. You cannot make things better and better by adding stuff. You can't. You must prune. You must rebuild. You must create an end for things. Toby, welcome back. Shane, it's so good to be back. I'm glad you're doing this again. How are you using AI internally at Shopify? We find some ways for it to be supportive now. It's actually look. When have you recorded last time? Oh, we recorded like two, three years ago. Yeah. So a hundred years of Internet. Yeah. Yeah. Like, look, I'm I'm 10 out of 10 nerd. I cannot bear the idea of, like, somehow not being at the forefront of a technology shift. I live for these things. Anyone growing up reading sci fi books wanted to live or my my take from sci fi books I read was, like, I wanted to live in that world. Like, how can I accelerate us there? Right? Like, so, you know, even even in whatever minor steps, we can get there. So inside of Shopify, the amount of people I know who really write code is, like, vanishingly small, and now it's it it still exists in the at the limits of complexity for sure, and, obviously, in the reviews and so on, and then state management of all things, it seems to be remains to be the thing that's really the hardest to get right, which people do by hand, and then sort of wipe the rest around it. Inside of Shopify, this is what things look like. Very, very, very few people are writing code directly. Everyone who does does it deeply assisted by many agents. They often ten, twenty, thirty, forty, 50 instances of them all through, you know, either sub agents or just different windows coordinating. They're pushing all sort of engineering infrastructure towards absolute limits. I'm a I'm a student of computing history, really, because I think it's actually mainline history as it will be taught a thousand years from now looking backwards. But, like, the main accomplishments of these years are going to be clearly the emergence of AI and the technological breakthroughs and also the interconnectiveness of the Internet and all this kind of infrastructure we created. Those are the great works of our time. But where we started, as a young industry, we tend to not be seeped in tradition, or we mistrust the great lessons that have been found by the people by by by the greats of our industry. Right? In fact, we are the only industry in computing that doesn't even under know its heroes. Imagine people in physics not knowing who Richard Feynman or Richard Feynman is actually like like, he might even be too obscured, but, like, I mean, Isaac Newton, Albert Einstein. But you go into computer science, it's like, who's who's who's you, Newton? And no one knows Allen Key and Dennis Ritchie and Ken Thompson. This matters, I think, because we discard great …
Get the full transcript (11,206 words) + summary by email — free
One-time email with the complete transcript and AI summary of this episode. No account needed.
One email, no spam. We’ll also show you what SignalCast does.
You just read a 3-minute summary of a 62-minute episode.
Get The Knowledge Project summarized like this every Monday — plus up to 2 more podcasts, free.
Pick Your Podcasts — FreeKeep Reading
More from The Knowledge Project
Steve Jobs in Exile
Sep 1 · 63 min
The AI Breakdown
OpenClaw 2.0 Shows Where AI Agents Are Going Next
Sep 1
More from The Knowledge Project
Roblox CEO: How to Make Better Decisions by Fixing Yourself First
Aug 18 · 46 min
Lex Fridman Podcast
#501 – DHH: Future of Programming, AI, Agentic Engineering, Vibe Coding & Linux
Aug 26
More from The Knowledge Project
We summarize every new episode. Want them in your inbox?
Steve Jobs in Exile
Roblox CEO: How to Make Better Decisions by Fixing Yourself First
The Mindset Behind Building Billion-Dollar Companies | Brad Jacobs
John D. Rockefeller: The Principles Behind The Greatest Fortune in History
Truth Over Feelings: Inside Opendoor’s Massive Turnaround
Similar Episodes
Related episodes from other podcasts
The AI Breakdown
Sep 1
OpenClaw 2.0 Shows Where AI Agents Are Going Next
Lex Fridman Podcast
Aug 26
#501 – DHH: Future of Programming, AI, Agentic Engineering, Vibe Coding & Linux
David Senra
Jul 26
David Heinemeier Hansson (DHH), Co-founder of 37signals
How I Built This
Jun 8
Shopify: Tobias Lütke. How a snowboarder built a $150 billion business (2019)
How I AI
May 11
Spec-driven development: The AI engineering workflow at Notion | Ryan Nystrom
Explore Related Topics
This podcast is featured in Best Business Podcasts (2026) — ranked and reviewed with AI summaries.
Read this week's Startups & Product Podcast Insights — cross-podcast analysis updated weekly.
You're clearly into The Knowledge Project.
Every Monday, we deliver AI summaries of the latest episodes from The Knowledge Project and 192+ other podcasts. Free for one show.
Start My Monday DigestNo credit card · Unsubscribe anytime