A team from Apple, Google, and Nike is building something the AI industry has never seen: models that don’t just generate, they judge.
I’ve been covering AI startups for years, and most pitches sound the same. Faster models. Better prompts. More parameters. Everyone’s racing to generate more…more text, more images, more code, more everything.
Then I got a breakdown from Osyle, and realized we’ve all been solving the wrong problem.
The company, built by a team drawn from Apple, Google, Nike, Spotify, and top research institutions, isn’t trying to make AI generate faster or cheaper. They’re trying to teach it something far more valuable: taste.
Not style. Not aesthetics. Actual judgment, the invisible layer of expertise that separates a senior designer’s work from a junior’s, a principal engineer’s code from a bootcamp grad’s, a world-class scientist’s reasoning from surface-level analysis.
And if they’re right, they’re not just building a better tool. They’re creating an entirely new category of AI infrastructure.
The problem hiding in plain sight
Here’s what struck me during our conversation: We’ve gotten so used to mediocre AI output that we’ve stopped questioning it.
Ask ChatGPT to design a dashboard. It’ll give you something. Buttons, menus, and layouts which are all technically functional. But will it feel intuitive? Will users understand it immediately? Will it communicate the right hierarchy?
Probably not. Because the model has no sense of what actually belongs, it’s pattern-matching from millions of examples scraped from the internet—the brilliant and the terrible, all blended together.
“AI can generate anything now,” explained one of Osyle’s founders, who spent years leading product at a major tech company. “But it still doesn’t know what’s good. That’s not a data problem. It’s a judgment problem.”
The more I thought about it, the more I realized they’re right. Every designer I know spends hours fixing AI generated work. Every engineer I talk to says AI code is “okay” but needs significant refactoring. Every creative professional has the same complaint: AI gives you a starting point, not a final product.
Because AI doesn’t think like experts think. It generates based on patterns, not principles. It produces output, not decisions.
What Osyle actually built
This is where it gets interesting. Osyle isn’t trying to train bigger models or engineer better prompts. They’ve created what they call “Taste & Judgment Models,” an entirely new layer that wraps around existing LLMs and teaches them how experts actually reason.
Think of it like this: Most AI fine-tuning is done by engineers who’ve never worked in the domain they’re optimizing for. So the model learns a general way of solving problems, not an expert one.
Osyle’s approach is different. Their system extracts the actual decision-making patterns from top designers, engineers, and scientists. Not their style or templates but their reasoning. The structure behind their choices. The judgment that makes their work unmistakable.
Then it converts that expertise into a reusable cognitive layer that can wrap any LLM.
The result? AI that doesn’t just generate, but it actually decides with the same clarity, structure, and intuition as someone who’s spent years and years mastering their craft.
Why this matters beyond the tech
We’ve seen plenty of startups promise to make AI “smarter” or “more human.” Most are just clever prompt engineering.
But Osyle’s approach solves three problems that have been bothering me for a while:
The ownership problem. Right now, AI training scrapes the open internet. Your work, your style, your expertise gets blended into generic outputs with no credit or control. Osyle’s Taste Models are fully owned by the experts who create them. It’s a new way to license and protect expertise instead of having it stolen by training data.
The mediocrity problem. Most startups don’t have access to world-class mentors. They’re building with AI that doesn’t understand quality because it was never taught by people who do. Osyle gives them access to expert-level thinking they’ve never had before.
The craft problem. There’s a quiet erosion happening across the internet. Everything’s starting to look the same with generic design, cookie-cutter copy, and soulless interfaces. People can easily sniff it out. Osyle is betting that we’re ready for AI that restores quality instead of diluting it.
The bigger bet
What Osyle is really arguing is that the future of AI isn’t about generation, it’s about judgment.
We’ve solved the “produce output” problem. What we haven’t solved is the “know what’s good” problem. And that’s increasingly what matters.
For founders building products, it’s the difference between shipping something that works and shipping something that feels right. For designers, it’s the gap between technically correct and genuinely excellent. For anyone trying to create something meaningful, it’s the ability to work at a level that was previously locked inside the top 0.1%.
The team is starting to roll out early access to 100 people per week over the next month. Their target audience is exactly who you’d expect: AI-forward founders, design leaders, engineers, investors, and domain experts who understand the difference between “okay” and outstanding.
What I’m watching for
Here’s what will determine if Osyle succeeds: Can they actually capture expert judgment in a way that scales? And will people pay for better thinking when they can get free generation?
My gut says yes on both counts. Because the more AI-generated content floods the internet, the more valuable genuine expertise becomes. And the more we use AI tools, the more frustrated we get with outputs that are technically correct but creatively flat.
If Osyle can deliver on their promise with AI that thinks like the top 0.1% designers and founders then they’re not just building a product. They’re creating infrastructure for a future where AI amplifies expertise and creativity, versus getting rid of it.
This is something to be ecstatic about, especially if you use AI heavily.
Disclaimer: GeekWire newsroom and editorial staff were not involved in the creation of this content..
