Linear’s Secret to Building Powerful AI Products | Nan Yu, Head of Product (Linear)
From redefining product with AI harnesses to rethinking accountability through agent design, Nan Yu shares how Linear is shaping the future of AI-powered product development.
Today, we’re joined again by Nan Yu, Head of Product at Linear.
In this episode, Nan shares:
Why you’re wrong to think of AI as a way to reduce costs — and why the real value is in enabling new work that was previously impossible
The 3 must-haves he’s deduced from across the most successful AI products to produce consistently great outcomes for users
The behind-the-scenes details of Linear’s new Agent Interaction Guidelines
1. AI harnesses and the future of product design (1:36)
Nan describes every great AI tool as a harness, a structured system that wraps intelligence in workflow, context, and constraints.
“The chat interface, the code editor, the context window — those are all harnesses. The magic isn’t the model; it’s how you build the frame around it so the median outcome is consistently great.”
At Linear, these harnesses ensure AI doesn’t just perform one-off miracles for power users but reliably delivers value to every engineer and designer on the team.
2. From cost wavings to new value creation (6:27)
According to Nan, cost reduction is the least interesting part of AI.
“The value you were getting from it was zero. And by applying AI in a well-harnessed manner, you end up making the cost negligible, Like free-ish. And the result isn’t, you’ve saved some money.
The result is the thing you weren’t doing at all before. Now you’re doing it a huge amount. You were getting zero value from this process that theoretically could have existed, but now you’re getting almost unbounded value because you can just keep running it on repeat.”
3. Measuring AI quality through “tweak time” (13:30)
Building AI features isn’t just about shipping UI; it’s about validating whether the model’s output is actually good.
Nan shares how Linear evaluates success using a mix of behavioral metrics and qualitative observation — something Oji Udezue calls “tweak time.”
By studying how people edit AI output, Linear gets closer to understanding when its features truly deliver value.
4. Inside Linear’s Agent Interaction Guidelines (18:30)
Linear recently published a set of design rules for developers building agents on its platform.
One of the rules?
“An agent can’t be held accountable.”
Nan explains that this principle reshapes how teams design responsibility into AI systems. Agents can assist and delegate, but the human remains the accountable party. Linear even reworked its issue-assignment model to reflect this shift, introducing delegation instead of ownership to clarify accountability.
5. How AI is redefining engineering roles (22:15)
At Linear, designers can request code changes directly from agents, freeing engineers to focus on higher-impact problems. The result is faster iteration, fewer interruptions, and a culture where even 10% ideas can quickly become prototypes worth testing.
“Being a good code reviewer is now a core skill. You’re no longer just writing code — you’re evaluating generated code for correctness, readability, and intent.”
Chapters
00:00: Intro
03:24: Using AI as a harness, the systems and context layers that make AI tools repeatable and useful at scale
7:54: Scaling AI tools to ensure consistency and quality
14:14: Analyzing user interaction, tweak time, and feedback
18:29: Linear’s new Agent Interaction Guidelines
22:14: How AI is changing the role of engineers
27:42: AI’s impact on junior roles
30:10: Conclusion
Links
Resources
Nan’s previous episode of LaunchPod:
What does LogRocket do?
LogRocket’s Galileo AI watches user sessions for you and surfaces the technical and usability issues holding back your web and mobile apps. Understand where your users are struggling by trying it for free at LogRocket.com.