Key Takeaways
- Making software yields two distinct outputs: the finished product and the team learning gained while solving hard problems.
- Handing all engineering execution over to AI agents separates builders from customer friction and technical nuance.
- Linear protects product judgment with internal rituals like Quality Wednesdays, Feature Roasts, and AI-driven feedback summaries.
- The correct approach automates repeatable, zero-learning chores while redirecting saved engineer hours directly into customer contact and craft.
The Two Outputs of Building
Most software leaders treat development purely as an output problem. They measure tickets closed, story points burned, and pull requests merged. When generative tools arrive, their immediate instinct is simple: generate ten times more code with half the people.
Karri Saarinen sees this as a dangerous misunderstanding of how great products actually get built. Speaking at the Lenny and Friends Summit, Saarinen pointed out that the code itself is only half the prize.
“When I think about making things I think about two things,” Saarinen explained. “Making products produces two things. The product and the learning. So the actual effort you put into building things, it also teaches you something about what the problem is, what the customers want or what is their view.”
When engineers wrestle with edge cases, strange database constraints, and confusing user flows, they build deep intuition. That intuition informs every subsequent product decision. If you outsource that wrestling match to an automated system, the code still ships, but the team ends the week just as blind as they started it.
The Danger of the Low-Learning Software Factory
If speed of code generation becomes your only metric, your startup turns into a low-learning software factory. You produce features faster, yet your strategic understanding of the customer stalls.
Saarinen warned about this exact separation: “The danger really now with the product organization is that the more we automate stuff to AI to do and more we automate it or tell teams to use AI, you kind of create this separation from execution and from the learning. Eventually if you don't learn from the work you do I think eventually you would lose the advantage of what you have currently in your space or in your company.”
A company that stops learning cannot defend its market position. Competitors using the same AI models can copy features in days. The only moat that lasts is compounding institutional context: knowing exactly why a feature failed three years ago, how power users actually navigate your settings, and where your architecture creaks under load.
How Linear Separates Chores From Craft
Saarinen is not anti-AI. Linear uses AI tools extensively, but they draw a clear line between what gets automated and what stays human.
Linear automates the noisy, low-context work. They use AI to summarize and cluster raw customer support tickets into structured briefings. But humans still read the briefings, debate the trade-offs, and design the solutions.
To keep technical craft sharp, Linear runs deliberate rituals. During Feature Roasts, team members scrutinize unreleased work to expose clumsy interactions. On Quality Wednesdays, engineers drop their roadmaps to polish rough edges, fix tiny UI bugs, and obsess over performance. They use AI to eliminate boilerplate code, then invest the saved time straight back into human critique and direct user conversations.
What to Do With This
Audit your engineering backlog this week and label every recurring task as either high-learning or low-learning. Automate the low-learning chores (like writing test fixtures, drafting boilerplate endpoints, or triaging error logs). Take the hours you save and mandate that your engineers spend them shadowing two live customer support calls and running one internal feature review.