After 18 years as an architect, leading teams of dozens and once carrying an entire project solo, I’ve observed something over the past year—
Teams that use AI well and teams that don’t are pulling ahead of each other by the month.
But this gap isn’t what you think it is.
The Gap You Think You See: How Well You Use AI
Most teams still see AI as just another tool.
Give designers Midjourney, give programmers Cursor, give operations ChatGPT—and expect efficiency to skyrocket.
Did it? A little.
But soon you realize: everyone is faster individually, yet the team as a whole is still slow. Designers produce images faster, but coordination with development still relies on verbal handoffs. Programmers write code faster, but communication costs around requirement changes haven’t dropped at all. Operations produces copy faster, but still can’t sync with product’s rhythm.
Individual efficiency went up, but collaboration friction didn’t decrease.
It’s like giving everyone a sports car, but the road is still dirt.
The Real Gap: Whether Collaboration Methods Changed
I’ve been thinking about something lately: why do some teams transform after using AI, while others just get slightly faster?
The answer isn’t in the AI itself—it’s in whether collaboration methods changed with it.
Let me share a personal example.
In the past, when I led teams on projects, the process was: Product writes PRD → Tech review → Development → Testing → Launch. Sequential progression, each stage waiting for the previous one to finish.
Now? My collaboration with AI is completely different—
I throw the requirements to AI first, letting it help with technical research and initial draft proposals; meanwhile, I’m thinking through architectural tradeoffs on my own. After AI produces a proposal, I evaluate which parts are solid and which aren’t, then have it iterate.
It’s not me finishing then handing off to AI, nor AI finishing then handing off to me—it’s both of us pushing forward in parallel on the same problem.
This collaboration style simply can’t be supported by traditional tools. ChatGPT is a chat window that disappears when you close it. Cursor is a code editor that only handles code, not context.
What you need is an environment where human judgment and AI execution can collaborate continuously.
Three Overlooked Problems
In the process of building an AI collaboration platform, three issues kept coming up—and almost no products on the market truly solve them.
1. AI Forgets Everything After Finishing a Task
You had AI analyze a competitor and came up with great insights. The next day you want to continue, and it’s all gone.
You have to re-describe the background, re-explain the context, re-establish consensus.
What’s the most expensive thing in team collaboration? Alignment costs. Every time you have to realign, you’re wasting people’s scarcest resource—attention.
A good collaboration platform needs traceable memory. Not AI remembering on its own—that’s a black box where you don’t know what it remembered or forgot. But white-box memory that humans can see, search, and reference.
Pits you fell into three months ago don’t need to be fallen into again; preferences you’ve corrected AI on don’t need to be retaught every time.
This information shouldn’t exist only in someone’s head, nor be re-fed to AI every single time.
2. Different Tasks Need Different AI, But Switching Costs Are Too High
Claude for coding, GPT for data analysis, DALL-E for image generation.
But in real work, one task often needs multiple AIs working in sequence. You copy一段代码 from Claude, paste it into GPT to write documentation, then feed that document to another AI to generate test cases…
Every switch loses context. Time spent moving things around might exceed the time AI saved you.
A good collaboration platform should have intelligent routing—automatically selecting the most suitable model based on task type, while maintaining context continuity. You just state the requirement; the platform handles the orchestration.
3. AI Only Works When Someone’s Watching
This is the most underestimated problem.
Today’s AI tools are essentially Q&A models—you ask, it answers. You don’t ask, it stops.
But real collaboration doesn’t work that way. Your colleagues don’t sit waiting for your messages before they work—they proactively push things forward, report progress, and align with others.
AI should work the same way.
When a requirement comes in, AI should automatically break it down into tasks, start executing, and ask for your decisions at key checkpoints—rather than waiting for you to feed it instructions one by one.
From humans driving AI to AI driving execution, with humans driving decisions—that’s the core shift in collaboration model.
Back to Basics: What Is Collaboration
At the end of the day, the word “collaboration” has been used so loosely that we’ve forgotten its essence.
Collaboration isn’t you finishing then handing off to me—that’s handoff.
Collaboration isn’t you helping me with something—that’s delegation.
Collaboration is: two people (or person + AI), each leveraging their strengths, continuously aligning, and jointly pushing forward toward the same goal.
This means—
- Both parties have context (knowing why we’re doing this, where we are)
- Both parties have proactivity (not passively waiting for instructions)
- Both parties have memory (conclusions from last discussion don’t need to be rehashed)
Today’s AI tools satisfy execution needs, but they don’t satisfy collaboration needs.
That’s why teams of the future need an AI collaboration platform—not a better tool, but an environment where human judgment and AI execution can truly merge.
Closing Thoughts
After 18 years as an architect, my biggest insight is: technology is never the bottleneck, collaboration is.
No matter how good the technology, if collaboration methods don’t change, you only speed up the old process a bit. It’s like putting a faster horse on a carriage—the road is still crooked.
AI has given us an opportunity to redesign the road.
Not to help you run faster, but to help you replan the route.
The window of opportunity will not stay open for long. During the last electrical revolution, enterprises spent four decades completing organizational adaptation. For the ongoing AI revolution, businesses may only have one or two years.
It is no longer a question of whether transformation should happen, but whether there is enough time to get it done.
← Back to Articles