Hello everyone, I’m Captain Jack Smith. In the fast-moving world of tech, it’s easy to get swept away by hype or paralyzed by doomsday prophecies. Today, I’m diving into a refreshing perspective from Karri Saarinen, the founder and CEO of Linear. He recently shared some candid reflections on the current state of AI that move past the "AI will save/destroy the world" binary and focus on what’s actually happening on the ground.
The Six-Month Reality Check
Karri points out a significant milestone: it has been nearly six months since the last major leap in model coding capabilities. Six months is typically the length of a "honeymoon period." Once it ends, the rose-colored glasses come off, and reality sets in.
His stance is one of cautious optimism. While AI’s capabilities are undeniable, its limitations are equally real. Karri argues that the public discourse has become too polarized. We are missing the "middle ground"—the space where we ask: What is truly changing? What is actually useful? What is pure hype? Navigating this space requires the rare ability to remain calm between the extremes of excitement and fear.
Is Planning Obsolete in the AI Era?
There’s a growing sentiment that because things move so fast now, planning is a waste of time. Karri disagrees. He notes that the value of planning was never about the document itself; it’s about the "forcing function." Planning forces an organization to sit down, debate priorities, and align on a direction.
In the AI era, building things has become cheaper and faster, which paradoxically makes "choosing what to build" more critical than ever. When execution cost drops, it becomes much easier to build the wrong thing. At Linear, they maintain a six-month directional plan but adjust priorities weekly. Without a compass, you’ll likely find yourself being led by your tools rather than leading them.
The Expertise Paradox: AI Looks Like Magic Only to Novices
One of Karri’s most astute observations is that AI is most impressive in fields where you are least knowledgeable. When you lack judgment in a domain, you can’t spot the hallucinations or the mediocrity, so it feels like magic. However, in your area of expertise, you see the missing context, the made-up details, and the lack of nuance.
He likens this to a combination of "Gell-Mann Amnesia" and the "Dunning-Kruger Effect." The paradox here is that expertise actually makes AI harder to use because you become more critical of its output. Yet, expertise also makes AI more valuable, as you are the only one who knows how to guide, constrain, and evaluate the results. Professional skills aren't devalued; they are refocused toward judgment and taste.
The Reality of AI Coding: Useful, Not Autonomous
Despite the narrative of fully autonomous "AI Agents," the reality inside top engineering teams is different. Almost no one is running independent agent swarms. Instead, engineers remain deeply involved, managing two or three agents at a time to handle boilerplate, bug fixes, and tests.
Linear’s own data shows that while usage of coding agents has grown 5x in months, they are used for "scaffolding" rather than core architecture. AI increases bandwidth for the small, tedious tasks that weren't worth doing before, but the "hard problems"—trade-offs, system understanding, and deciding what should exist—still require human intelligence.
Design in the AI Era: A Need for Semantic Tools
As a design-led CEO, Karri is skeptical of current AI design tools. Image generation is powerful but miserable to iterate on—changing one detail often ruins the whole image. He also argues against tools that force design to happen directly in production code. Design is about exploration and messiness; it shouldn't be constrained by the rigidity or token costs of production environments. He envisions "semantic design tools" where AI helps explore variations of components (like a "pop-up" instead of just a "rectangle") rather than just spitting out code.
Conclusion: Living Within Present Capabilities
Karri responds to the common refrain that "this is the worst AI will ever be" with a dose of pragmatism. While true, he chooses to live within the capabilities of today. Predicting the end of the world or a utopia is easy; building a product that works right now is hard.
He concludes with a sharp adaptation of a nursery rhyme: “If 'potential' were 'revenue,' these capital expenditures would have been 'profits' long ago.” In the gap between potential and reality, only those who remain clear-eyed will go the distance.
Thanks for joining me on this deep dive. I'm Captain Jack Smith, and I believe that in the age of AI, our human judgment remains our most valuable anchor. What do you think? Is your AI honeymoon over, or are you just getting started? Let's discuss in the comments.
Original Source: https://x.com/karrisaarinen/status/2048267794924650791
Top comments (0)