TL;DR
The short version
AI makes execution cheap. Karri Saarinen's warning: if you automate building without protecting learning, your product org hollows out — context, not code throughput, becomes the product.
From his Lenny & Friends Summit talk: automate known work, spend savings on customers, build shared context with watcher agents, and run quality rituals so judgment compounds.
Karri Saarinen (CEO, Linear), Lenny & Friends Summit talk published on Lenny's Podcast 29 Sep 2026.
Factories scale output; they can starve judgment
After a summer offline from AI news, Saarinen found new models and agents — and the same hard problem of shipping something people want. The industry loves software factories. His cut is cultural: you can build factory machinery without becoming one. Customers buy outcomes, not experiment count.
Making products produces two things. The product and the learning.
Karri Saarinen
Automate execution carelessly and you sever the learning loop. Compound understanding — why strong product companies stay strong — stops compounding.
Automate the known; spend savings on customers
Linear uses agent loops to investigate bugs; humans verify. Repeatable, low-learning work is a good machine job. Freed time should go into the customer space. AI also helps as a context engine: collect feedback, send watcher agents for daily briefings on themes leaders care about.
Rituals against solo-agent silos
Quality Wednesday: everyone finds one defect and looks together — the trained eye matters more than the fixes. Feature roasts: optional company-wide raw critique before a feature hardens. Put customer and product thinking somewhere agents and humans both read.
