AI Didn't Make Collaboration Easier. It Just Moved the Bottleneck.
Does AI make teamwork easier? Where’s the real bottleneck in your pipeline?
My take — AI didn’t reduce the friction. It relocated it.
We needed to put together a review deck. Engineers and PMs, about eight people.
The beginning was the same as always: a meeting to align on the storyline and structure. AI doesn’t help here. Alignment is a carbon-based job.
Then everyone went off to build their sections. I figured with AI, this would be fast.
It was. The problem came after.
Everyone Delivered a “Finished Product”
What came back was wild: polished charts, color-coordinated infographics, beautifully designed layouts. Each page looked great in isolation.
Together? They looked like eight different decks.
Inconsistent styles, varying information density, visual languages that didn’t speak to each other. Before AI, people submitted rough text boxes — the integrator had room to shape things. Now everyone submitted “finished” pages. They were harder to take apart than raw materials ever were.
The person stitching it all together was drowning in clutter, hunting for the actual content underneath the design.
It Wasn’t This Painful Before
There used to be a natural friction: making good-looking slides was hard. Most people didn’t bother learning advanced PowerPoint, so they defaulted to plain text. Content stayed with the individual, presentation stayed with the integrator. Clean separation.
AI killed that friction. Generating beautiful pages is now effortless. Everyone can’t resist “designing” their own slides. Content and presentation get shipped as one package — and the integrator receives not raw material, but a pile of incompatible finished products.
I Tried What I Could
- Extract design tokens and apply uniformly → Didn’t work. Each page had completely different component structures. No common layer to abstract.
- AI-powered page-by-page reformatting → Didn’t work. Font sizes and component types clashed. AI needs relatively structured input; faced with chaotic finished slides, it couldn’t produce cleaner output.
- Various AI presentation tools → Same problem. They’re great at generating a full deck from raw text, but can’t unify multiple incompatible finished pages into a coherent whole.
In the end? Manual labor. Page by page. Four to five hours just on stitching.
The Bottleneck Didn’t Disappear. It Moved.
This maps perfectly to the Theory of Constraints (TOC) — Goldratt’s 1984 framework: A system’s output is determined by its bottleneck. When you break through one bottleneck, it doesn’t vanish. It moves to the next stage.
AI accelerated “individual content production.” The bottleneck immediately shifted to “team integration.”
In any organization, everyone is plugged into a pipeline. Speed up one stage, and the adjacent stages become the new constraint. Friction and efficiency are a zero-sum game — at least for now. The effort AI saved you will be repaid by someone else downstream.
This Isn’t an AI Problem. It’s a Collaboration Reality.
Everyone’s discussing what “AI native organizations” should look like. Most of that conversation focuses on the OPC (one-person company) or “one human commanding multiple agents.”
Reality: in a real team, you work with real people. AI hasn’t changed the fundamentals of collaboration — alignment, integration, enforcing shared standards. These don’t disappear just because your tools got stronger.
Suggestion — Resist the Urge to Let AI Make It Pretty
If you’re collaborating with AI in a large team: step back. Think first. Start with plain text.
Get the content right. Leave presentation for a unified pass at the end. It feels like a step backward, but it avoids 60% of the rework downstream.
Visualization is dessert. In team collaboration, eating dessert too early makes the main course unbearably painful.
AI 在团队协作方面，有没有让我们每个人变得更轻松？在一个长管线中，瓶颈到底在哪里？
我的观点 - AI 没有让协作变轻松，它只是把瓶颈挪了个位置
最近要搞review，和大家一起合作出个PPT。 工程加上 PM，七八个人。
开局和以前一样：开会，对齐 storyline，讨论结构。这步没变——AI 帮不上忙，因为对齐是碳基生物要干的事。
各人领了任务回去做各自的部分。我以为有了 AI，这事会很快。
的确快了。但问题出在了后面。
每个人都交了一份”成品”
大家交上来的东西五花八门：精美的图表、配色讲究的信息图、设计感拉满的排版。每一页单独看都很漂亮。
问题是，它们放在一起完全不像同一个 deck。
风格不统一、信息密度不同、视觉语言各说各话。以前大家交的是粗糙的文字框，整合的人有空间统一处理再包装。现在每个人交的都是”成品”——反而拆不动了。去整合的那个人在零碎的信息中痛苦地翻找真正需要的内容。
以前可没这么痛
以前有一个天然的摩擦力：做漂亮 PPT 很难。大多数人不愿意花时间学高级排版，所以默认交纯文字。内容归个人，呈现归整合者——分工清晰。
AI 把这个摩擦力消灭了。生成漂亮页面太容易了，每个人都忍不住”设计”自己的那几页。内容和呈现混在一起交上来，整合者拿到的不是原材料，是一堆各自成型、互不兼容的半成品。
我试了我能想到的办法
- 抽象 design tokens，统一 apply → 不行。每页的组件结构完全不同，没有公共的底层可以抽象。
- 用 AI 逐页重排 → 不行。字体大小、组件类型不匹配。AI 需要相对规则的输入，面对混乱的成品，它整理不出更干净的结果。
- 各种 AI PPT 工具 → 同样的问题。它们擅长从 raw text 生成整套 deck，但不擅长把多个不兼容的成品统一成一个整体。
最后还是上了古法手工。一页一页重新调整。花了4，5个小时在 stitching 上。
瓶颈没有消失，它只是移动了
Theory of Constraints（TOC）——Goldratt 在 1984 年提出的：系统的产出由瓶颈决定。当你打通一个瓶颈，它不会消失，只会转移到下一个环节。
AI 加速了”个人生产内容”这个环节。瓶颈立刻转移到了”团队整合”环节。
在企业里，每个人都被插在不同的 pipeline 上。某个环节被加速后，前后环节就变成新的瓶颈。摩擦和效率此消彼长。AI 帮你省下的力气，会由别人（整合者）来偿还。
这不是 AI 的问题，是协作的现实
很多人在讨论 AI native 组织会是什么样子。但大部分讨论集中在 OPC（one person company）或者”一个人指挥多个 agent”的理想场景。
现实是：在真实的团队里，你要和真实的人合作。AI 没有改变协作的本质——对齐、整合、统一标准——这些工作不会因为工具变强就自动消失。
建议 - 克制住”让 AI 帮我做漂亮”的冲动
如果你在大公司的团队里用 AI 协作：退一步。理清思路，从纯文本入手。
先把内容说清楚，把呈现留给最后统一处理。这看起来像倒退，但它避免了下游 60% 的返工。
可视化是甜点。但在团队协作里，过早吃甜点会让正餐变得无比痛苦。