Reshare — content by Julien Mutzenhardt
A company 100% run by Artificial Intelligence: Myth or Reality? 
Researchers at Carnegie Mellon University conducted an unprecedented experiment: simulating an entire company and handing the reins to AI agents. The roles of financial analyst, project manager, and software engineer were assigned to well-known models such as Claude from Anthropic, GPT-4o from OpenAI, Google Gemini, Amazon Nova, Meta Llama, and Qwen from Alibaba.
The result? It should reassure all those who fear losing their jobs to the machine. 
A failure on more than 75% of tasks
The results are clear-cut. The best performer in the experiment, Claude 3.5 Sonnet, managed to complete only 24% of the assigned missions (34.4% counting partially completed tasks). In second place, Gemini 2.0 Flash peaks at only 11.4%. None of the other agents tested managed to cross the 10% threshold.
Operating costs
While Claude 3.5 Sonnet is the most performant, it is also the most expensive in this comparison at a cost of $6.34, versus very economical operation at $0.79 for Gemini 2.0 Flash.
Why do AIs fail so much when working autonomously?
The researchers identified several critical limitations in these virtual agents:
The inability to understand the implicit (for example, they do not infer that a file with the “.docx” extension corresponds to the Microsoft Word format).
A lack of social skills during interactions with simulated virtual colleagues, such as the human resources department.
Technical difficulties during web navigation, notably the inability to handle pop-up windows.
The tendency to take shortcuts when facing difficulty, omitting the complex part, while wrongly believing they have successfully completed the task!
In conclusion, these results demonstrate that even if AIs can deliver excellent results on very specific tasks, they are still far from being capable of operating autonomously today.
The summary of this article was generated by AI.
What do you think of these results: does full AI autonomy in business seem to you like a simple matter of time or a technological utopia?
#ArtificialIntelligence #Tech #FutureOfWork #Innovation #Management #AI
If you would like to discuss the topic with me or call on my skills, you can contact me by PM or by video call.
LinkedIn Sources
- Profile: Pierre Jean Duvivier known as Sage (LinkedIn)
- Shared author: Julien Mutzenhardt (LinkedIn)
The post 🚀 A company 100% run by Artificial Intelligence: Myth or Reality? 🤖 appeared first on les plumes qui mordent.
Frequently Asked Questions
FAQs A Company 100 Run by Artificial Intelligence Myth or Reality
1 What does a company 100 run by AI actually mean
It means AI makes all key decisions strategy hiring finances operations with no humans in charge In practice no company operates this way today
2 Is it a myth or a reality
Mostly a myth for now AI can run many tasks and even whole workflows but no real company is fully run by AI without human oversight
3 What is AI in simple terms
Software that learns from data and makes predictions or decisions like recommending products screening resumes or flagging fraud
4 Whats the difference between automation and AI
Automation follows fixed rules AI learns patterns from data and handles new situations Automation is a script AI is more like a trained assistant
5 Are there real examples of highly AIdriven companies
Yes partially Some AIrun hedge funds and crypto funds use algorithms for trading but humans still set risk limits and step in Many companies run AIheavy operations with small human teams
6 What can AI already do well in a business
Customer support chatbots data analysis fraud detection pricing scheduling content generation code writing and routine backoffice work
7 What cant AI do yet
Handle true accountability make ethical judgment calls navigate novel situations reliably build trust with people or take legal responsibility
8 Why cant a company be fully AIrun right now
Someone must be legally accountable AI makes confident mistakes it struggles with new situations and it lacks real judgment and empathy
9 What are the main benefits of AIrun operations
Speed 247 availability lower costs consistent decisions and the ability to process huge amounts of data
10 What are the common problems
Hallucinations bias in training data lack of accountability security risks high setup costs and difficulty handling rare situations
11 What does human in the loop mean
A person reviews or approves important AI decisions Its the standard safeguard in most businesses today
