Lately I've been thinking about a problem. I've thought about it for a long time, but haven't figured it out.
How to lead a team, how to divide work—these are things I used to think had established patterns. Now I'm finding those patterns don't work anymore. AI has shattered many assumptions that used to hold true. To be honest, I still haven't found a new approach.
Functional Division of Labor Is Failing
The premise of traditional division of labor is simple: you're good at frontend, I'm good at backend, he's good at design. Everyone does their own part, and when pieced together, it becomes a complete product.
That premise is collapsing. AI is expanding each person's capability boundaries, and fast. One person plus AI can accomplish more than a small team used to—and the processes and systems that big companies accumulated have become burdens instead. So does it still make sense to divide roles based on "who's good at what"?
What's more headache-inducing is that within the same team, some people have already used AI to 10x their productivity, while others haven't moved at all. This isn't a traditional skill gap—it's different acceptance levels of new tools. You simply can't evaluate and assign work using the old methods.
Building Things Is Faster Than Meetings
The bottleneck of collaboration used to be communication. Meetings, aligning on goals, discussing plans—these took up huge amounts of time. But back then, production itself was slow, so spending time talking was worth it.
Now it's reversed. In the time you're still sitting in meetings discussing, the thing has already been built, shipped, and is serving users. So why would you still sit down and talk with someone for half a day?
Here's a tricky thing: the speed of absorbing information can no longer keep up with the speed of creating and validating. People don't want to talk anymore; they want to go directly build. This isn't an attitude problem—the efficiency structure has changed.
Only Validated Ideas Are Valuable
When everyone has high productivity, what becomes scarce?
Not ideas, and not output. Both of those are depreciating. What's valuable is validated hypotheses. You propose an idea, test it, and find it works—it solves the problem. It's this process of "being proven right" that has meaning.
So what should the team do? Accelerate validation speed.
I have an immature hypothesis: do horse racing under a shared goal. Everyone explores and validates independently; no need to converge from the start. Good ideas that emerge get extracted and formed into shared Infra that everyone can use. Seek common ground while reserving differences, allowing personalization under one platform. For example, assign some AI Agents to specifically run analyses on which directions are more likely to produce good things.
Passion May Matter More Than Skill
Traditional division of labor looks at skills. But after skill differences are flattened by AI, what makes people build different things is what they care about.
Some people's passion lies in creating new things—they want to try everything. Others care more about whether the system is stable and secure. Some care about making the product look better and the experience more pleasant. And others want good ideas to be seen by more people—not just building, but also promoting. You can't just build, right? Spreading validated ideas is important in itself.
These concerns differ, but a good product precisely needs to pass muster across many dimensions. Being outstanding in just one area while rough everywhere else won't get you far.
Diversity Might Be the Direction
I have a vague feeling that diversity might be the answer.
Not the slogan kind of diversity. I mean teams no longer need a bunch of similar people, but rather very different ones. Different passions, different perspectives, different aesthetics, even using different AI models—these differences themselves create value.
You add a brick, I add a brick. Some people modify it to make it look better, some to make it run faster, some to keep it from crashing. Building an engine together, each person contributing a different dimension.
Conversely, putting the same type of people together actually makes it hard to form synergies. When everyone focuses on the same dimension, they end up competing with each other more than collaborating.
No Conclusion
Writing up to this point, I still have no conclusion.
AI is driving organizational-level change—I'm certain of that. But how to respond, I'm still feeling my way through. Sense of purpose has become more important; if the goal itself can't inspire people, no division of labor method will work. Passion is replacing skill as the new basis for division. Diversity has become crucial.
But how do these turn into a collaboration system that actually works? I don't know.
I think this is a question every team will have to confront going forward. Productivity is high all around, old collaboration methods don't work well anymore, and new ones haven't grown in yet. And during this transition period, the real bottleneck for AI startups isn't technology—it's passion, focus, process, and stamina.