The cyber foreman is here. Actual test of CAO on duty: It took 4 minutes to assemble the team, but the whole day was a fiasco.

Since humans started focusing on the field of AI assistants, there has been quite a lot of commotion...

The year before last, everyone was frantically memorizing how to write prompt words, fearing that they would fall behind the times. Last year, people started searching everywhere for useful single Agents, hoping to find some free cyber workers for themselves. The competition became more and more intense.

After all the commotion, people found that relying on a single Agent alone, the experience seemed to have reached its limit. But if you want to prepare more Agents and Skills, you'll inexplicably become a foreman, and every morning when you open your eyes, you'll be assigning tasks to more than a dozen different AI tools.

This is not about liberating productivity; it's like taking on a new burden.

Is there a possibility that we can find a manager for the numerous Agents so that they can plan and complete their work on their own?

LobeHub thought so too.

(Image source: X)

On May 18th, the open - source AI project LobeHub announced its latest update on X - Chief Agent Operator, whose full name is Chief Intelligent Agent Operator. It claims to be able to help users recruit hundreds of Agents autonomously and form a professional team that operates non - stop for you.

Wow, has the era changed, and AI has suddenly become a supervisor?

The experience of being a cyber foreman is quite harsh

Let me briefly introduce the background of LobeHub to those who just got online.

This thing was originally a quite popular Agent integration and interaction project in the open - source community. On the GitHub platform full of balding tech experts, it has amassed nearly 80,000 stars. It's definitely a top - notch player in the open - source circle.

(Image source: Github)

Its main function is also very simple, which is to integrate various large models and Agents that users can deploy into one interface.

For example, Codex, Claude Code, OpenClaw, Hermes Agent, etc. As long as they are deployed, they can be connected to this unified entrance, and even the Skills and external service ports can be managed uniformly.

It's like fulfilling three wishes at once.

(Image source: LobeHub)

But these developers obviously thought it was too unchallenging to just create a nice integration UI, so they came up with this ultimate scheduling system called CAO.

According to the official promotion, you don't have to bother memorizing prompt words or looking for Agents anymore.

Now, you are the big boss who can lie back and command cyber workers.

For example, if you give CAO a task, like scraping all the reviews of a certain new energy vehicle on the whole network and making a table. After receiving the order, CAO will directly mobilize hundreds of thousands of skill templates from the library, instantly assemble a full team of Agent workers, and then arrange them to process tasks in parallel in the cloud.

You just need to lock your phone and have a good sleep. The next morning, you can check the daily briefing that CAO respectfully presents to you.

Doesn't it sound extremely cool? Anyway, I immediately tried it when I heard the news.

(Image source: Lei Technology)

As you can see, this time the requirement is relatively simple.

After receiving the requirement, LobeHub quickly called the Agent Management tool and asked me for a clear set of parameters.

(Image source: Lei Technology)

After I clarified the requirements, soon, a professional team was set up.

(Image source: Lei Technology)

The whole setup process took less than four minutes, and the division of labor and arrangement of the team were quite clear. Each member had its own name and responsibilities, rather than being a "jack - of - all - trades".

Next, it's time to let it generate and see the result.

Haha, just kidding. The result was a failure, and it was a continuous failure. You had to confirm and approve it countless times to keep trying, but still failed.

(Image source: Lei Technology)

Finally, there was something new... Why is there no quota left?

(Image source: Lei Technology)

To put it bluntly, this experience somewhat dampened my expectations. I'm not sure if I should be angry or laugh, but my face definitely turned green.

Although I really want to find a successful case for everyone, there are no successful users on overseas social media at present.

Instead of successful cases, many netizens complain that the threshold for using this thing is extremely high, the computing power consumption of automatic tasks is very confusing, and it's like throwing money at major model manufacturers. In the case of long - task chains, there is even a problem of a chain reaction of failures that makes people's blood pressure soar.

Leaving the task entirely to CAO is like giving a sports car without brakes to a novice who just got their driver's license. As long as the first Agent misinterprets the task and passes the wrong instructions to the second Agent, you may end up with a completely irrelevant result.

(Image source: Lei Technology)

If you use this thing for work, the time you spend cleaning up the mess the next day may be much more than doing the work from scratch by yourself.

Letting AI run wild? It's still a long way off

Seeing everyone crazy about complaining about LobeHub's CAO, I have a very strong sense of deja - vu.

Actually, this fully automated idea of letting AI manage AI is not unique to LobeHub. Old fans who often read Lei Technology's reviews may still remember that we tried a similar application called MasterAgent in July last year.

Our original intention at that time was very simple. We just wanted to have a cyber worker that could automatically write reports and make PPTs so that we could have more time to buy a few more cups of coffee downstairs.

The result... Although there were some bumps along the way, the final product easily outperformed LobeHub's CAO.

(Image source: Lei Technology)

Of course, MasterAgent also has its own problems. For example, the data it collects automatically may be from the wrong time, or several Agents may shift the blame and not do their work. It may even produce reports with perfunctory formats and short content.

But as long as you intervene in time, the final product it produces can be satisfactory.

Note the keyword here: intervene. From my current actual experience, it's still too sci - fi to completely let AI run wild.

However, whether it's last year's domestic pioneer MasterAgent or the currently popular LobeHub CAO, the blueprints they depict are trying their best to reduce human intervention. It seems that users really only need to give an instruction before going to bed and get a satisfactory result the next morning.

(Image source: LobeHub)

Dude, work isn't that easy.

Different from the question - and - answer dialogue - style Agents, applications like CAO are difficult to meet every demand. When you press the one - key start button, it's like pressing the countdown button for a rocket launch. You can only watch it roar into the sky and silently pray that it won't explode into a big firework in mid - air.

My attempt this time can be said to have crashed due to insufficient fuel.

Then comes the painful part. Since the work outline is assigned by CAO, the code in the middle is written by the underlying Agents, and the external data interface is called by another Agent. Facing an almost black - box execution process, you, me, and him will probably not be able to find out which Agent caused the problem in the end. The only solution is to start the whole task over again.

(Image source: Lei Technology)

I even encountered a bug where I couldn't see the skill calls, not to mention how to troubleshoot it.

Secondly, over - relying on such applications will also bring serious homogenization risks.

When the content operation, code development, and even market research reports of all companies in the market are produced in a pipeline - like way by a few open - source CAOs with similar Agent workers, the Internet may be filled with the same standard templates and soul - less nonsense.

Oh, I really can't imagine such a future.

In conclusion

However, despite the complaints, it can't stop the wheel of history from moving forward.

Just like when the fully automatic washing machine was invented, the washboard was quickly eliminated. No matter how many times there will be failures in the middle, no matter how much hair people will lose trying to adjust this application, handing over the cumbersome execution and distribution work to a fully automated management system is the future that Silicon Valley and domestic tech giants are all betting on.

Humans have never stopped exploring the path to becoming lazier.

As for those who are worried about losing their jobs to CAO... After all, CAO is not a real position, and it's not even a new concept. It can only be said that the speed of coining words in the tech circle is really amazing.

(Image source: LobeHub)

Moreover, this thing is different from CGO (Chief Growth Officer) and CTO (Chief Token Officer). Those two are at least real positions.

The former is essentially just a different way of writing CMO (Chief Marketing Officer). The latter is a bit far - fetched. I can't find many serious companies that have this position. It can only be said that at this point in time, most companies don't need someone to manage the