I’ve been watching Anthropic’s shipping pace for the past six months, and honestly, it pisses me off a little. Not because they’re good—they’re obviously good—but because it feels like this should be our gig. You know? The whole “move fast, break things” DNA, the 996 grind, the obsession with product iteration cycles that makes Western PMs cry. If you asked anyone five years ago which country’s tech teams would dominate the AI product race, they’d’ve pointed east without hesitation.
But here we are. Anthropic is dropping new features like clockwork: Claude 3.5 Sonnet, Computer Use, long context windows that actually work, and now they’re apparently shipping some new reasoning model under the codename “Diamond.” Meanwhile, the Chinese AI camps—and I watch them closely—are still mostly playing catch-up on baseline model performance, let alone product innovation. Why?
First thought: tools. Anthropic’s got the best damn AI programming tools and models in the world right now. Sonnet 3.5 alone is a productivity multiplier that makes every engineer there run faster. But that’s a weak excuse for us, because Chinese teams can use those tools too. API access might be restricted, but open-source alternatives exist, and the gap in raw capability isn’t that wide anymore. So tools alone don’t explain the velocity gap.
I think the real answer is messier, and way more uncomfortable: organizational inertia and incentive design.
Anthropic isn’t just a company—it’s a weird, almost radical experiment in how you structure a team when the product itself is rewriting the rules of labor. They’ve got a “responsible scaling” policy that effectively gives a safety team veto power over product launches. They’ve got no stock options for most employees (instead, they use a “benefit corporation” structure). And they’ve kept their engineering teams unusually small and flat.
Now compare that to a typical Chinese AI lab inside a major internet company. You’ve got layers of management that came up in the mobile internet era, when success was defined by capturing user attention, optimizing ad revenue, and executing rapid A/B tests on a massive installed base. Those leaders—many of them “老登” (old uncles) who built their careers on legacy paradigms—are now trying to direct AI research using playbooks from 2015. They push for “quick wins” like chatbot marketing stunts, or they demand alignment with existing business KPIs that have nothing to do with AGI capabilities.
The result? Teams that are simultaneously over-resourced and under-performing. Too many people, too many layers, too many meetings about “strategy alignment.” And the real builders—the cracked engineers and researchers who could ship at Anthropic speed—are buried under process, bureaucracy, and the expectation to deliver quarterly metrics that a PM dreamed up.
I’m not saying Chinese talent is worse. I’ve seen demos from ByteDance, Tencent, Alibaba that are technically breathtaking. But they rarely make it to production as a polished product. They get stuck in internal review, killed by competing teams, or diluted by “productization” requirements that strip out the magic.
Here’s the thing: I think this is a temporary state. Once an organization—any organization—figures out the right structure and incentives, the flywheel starts spinning, and it spins fast. Chinese teams have the raw material: brilliant engineers, high willingness to work hard, and a market that’s hungry for real AI applications. The bottleneck is the old guard and their outdated org charts.
But evolution, once triggered, accelerates. I’ve seen it happen in other waves. The first team that breaks free from the legacy mindset will pull ahead so fast it’ll make everyone else look stupid. The question is which side of the divide your team ends up on.
So yeah, Anthropic’s pace is a wake-up call. Not because we need better models or more computing—those will come. But because we need a different kind of organization. Leaner. Flatter. More willing to trust engineers than PMs. More obsessed with capability than revenue. And led by people who actually understand what they’re building, not people who got promoted for surviving the last war.
我观察了一下，this transformation has already started inside a few small teams I know. They’re small, hungry, and shipping stuff that makes you rethink the whole hierarchy thing. It’s only a matter of time before one of them breaks out. When that happens, the dam breaks. And nobody’s going to care about Anthropic anymore.