Don't Just Focus on "How Much Faster Individuals Are" – The Real Battlefield for AI Lies in "How Teams Stop Arguing"
Hello everyone, I'm your financial observer. Recently, there's been a hot topic in the AI community: Lark 8.0 has been updated, and many tech bloggers are excitedly analyzing its new features. In my view, the most valuable aspect of this update isn't what it can do with coding or drawing, but rather its attempt to address a universal pain point for all employees – "Why do everyone seem to be working faster, yet the company remains as busy as ever?"
Today, I'll break down this in-depth analysis of "team AI agents" into five key points, using simple language to help you understand what the next step of AI in the workplace means.
---
1. The Core Truth: Your Increased Efficiency May Be Being Offset by Other Factors
Let's face a harsh reality. Over the past three years, the biggest contribution of AI has been a significant boost in individual efficiency. It used to take two hours to write a proposal; now, AI can do it in 20 minutes. It used to take half a day to research; now, AI provides a framework in just one minute. On a personal level, that's certainly convenient, with efficiency doubling.
However, if you've led a team or participated in cross-departmental projects, you've probably experienced this:
- The proposal is done in 20 minutes, but you need to wait for another department to provide the necessary data.
- After completion, you have to check if the version has been updated, and some people might continue working with the old version despite new requirements.
- As a result, what used to take two hours now takes three days due to various communications, delays, and confirmations.
This is the core idea of the article:
> Team efficiency = The sum of individual efficiencies minus the cost of organizational collaboration
Until now, AI has mainly focused on improving the first half (individual efficiency), but no one has addressed the significant second half (collaboration costs). The trend now is that AI is starting to tackle this issue. It's no longer just doing the work for you; it's also helping to coordinate the work.
---
2. Role Transformation: From an "On-Demand Assistant" to a "Permanent Office Colleague"
Previously, we used AI like an external assistant. When you needed help, you opened ChatGPT, explained the context, and provided the information, and then you copied and pasted the results into the group or documents. It didn't know what had just happened unless you asked it.
The new "team agents" are more like a colleague with an excellent memory and a propensity to meddle in tasks. They are constantly integrated into your work流程 (group chats, meetings, documents, calendars). This means they:
- Understand the context: They remember what was discussed yesterday, who took on what tasks, and where the project is stuck.
- Proactively follow up: They provide needed information during discussions, make notes after meetings, automatically schedule tasks, and remind you if nothing is done after a few days.
The key difference is: Previous AI waited for you to ask questions; now, it acts in response to the tasks themselves. It's no longer a standalone tool; it has become part of the work process, helping to bridge the gaps that often cause delays.
---
3. Re-evaluating Value: The New KPI for AI is Not Just Accuracy, but Fewer Bottlenecks
For example, the workflow of a content team typically goes like this:
- Discover a hot topic → Discuss it in the group → Research → Organize into topics → Decision by the person in charge → Schedule → Notify the executor → Add to the calendar.
Each step isn't difficult, but each step can be interrupted:
- Do you forget to record the discussion?
- Does no one read the recorded information?
- Is the task scheduled but not notified?
A lot of office management is about reconnecting these broken links. The value of a team agent lies in its ability to automatically fix these issues, starting from group discussions and continuing with research, organization, scheduling, and notification.
So, in the future, the standard for evaluating whether an enterprise-level AI is useful will change:
- Old standard: Is its answer accurate? Is its writing style good?
- New standard: Can it reduce the number of bottlenecks in a task?
In a company, the real expense isn't the time spent on writing documents; it's the seemingly insignificant frictions. If AI can reduce these frictions, its value is much greater than just generating content.
---
4. Organizational Change: From "One Assistant per Person" to "One Agent for the Whole Team"
We used to imagine AI in the office as one assistant per person (one for sales, one for product, one for the boss). But the problem is this: Even if everyone becomes 30% faster, they still need to meet, synchronize, and confirm things. In fact, because everyone generates content faster, the amount of information the team needs to process increases, leading to higher communication costs.
The idea of a "shared agent" is to solve this problem. It serves the entire team, not just one person. New roles may emerge, such as:
- Project resident agents: They keep track of project progress without being specifically assigned to it.
- Shared sales agents: New employees can ask this agent for historical information directly.
- Business analysis agents: They automatically highlight unusual data for management to review weekly.
This raises a deeper question: Is the company's current work structure suitable for AI to join in? Perhaps the process needs to adapt to AI, rather than the other way around. If the company's processes are already chaotic, adding AI might only make things more complicated.
---
5. Potential Pitfalls: Money, Brainpower, Memory, and Responsibility
Despite the promising prospects, the article also points out some challenges:
1. Whose "brainpower" will AI use? (Model selection) A strong model may be needed for coding, but a cheaper one is sufficient for daily tasks. How can we ensure the agent's identity, memory, and permissions are properly managed while allowing it to switch tasks efficiently? This directly affects both cost and effectiveness.
2. How much will it cost? (Cost) Using AI occasionally is affordable, but for a team of dozens of people, 24/7 monitoring of group chats, meetings, and documents requires significant computational resources. Bosses will inevitably ask, "How much does it cost per month? What benefits does it bring?" If the benefits don't outweigh the cost, it's just an unnecessary expense.
3. Will too much information confuse AI? (Information management) Company groups contain a mix of important and trivial content. If AI remembers everything, it might become less efficient. Future companies will need to manage this new type of "information hygiene": What decisions are final, what are just discussions, and what is outdated? This is something that used to be managed by humans and now needs to be clearly defined with AI.
4. Who is responsible if something goes wrong? (Permissions and responsibility) It's fine if AI makes a mistake while summarizing, but if it can modify data, schedule tasks, or make decisions, where do the boundaries of responsibility lie? What it can see, what actions it can take, and when it must wait for confirmation—all these need to be clearly defined.
---
Summary: The Next Step for AI is Collaboration
In conclusion, the Lark 8.0 update, or the broader trend of team agents, represents more than just another office AI tool. In the past few years, we've focused on training AI to help individuals work better (by improving models, tools, and memory). Next, we need to train AI to work effectively with teams (by managing permissions, processes, responsibilities, and context). These two approaches differ significantly in their focus. While individual AI enhances capabilities, team AI focuses on improving efficiency and coordination.
If AI can solve the long-standing problem of reducing unnecessary friction among team members, its value will be much greater than simply making documents faster. This is the real direction we should pay attention to as AI moves deeper into corporate environments.