Some AI agent customers say reality doesn’t match the hype

Hi and greet the eye Ai. In today’s publication … Companies experimenting with AI agents say technologies are waiting for expectations. NVIDIA announces its new chips and positions itself for post-baking landscape. Elon Musk and Nvdia join the Microsoft-Blackrock AI Foundation. Ai Spammers “Makes Makes” Internet. And Foxconn acts as a GLOBAL AI Race as a major player.

It is unlikely to pass a day without a technology company that announces a new AI “agent”, it is said that the work flows will be revolved and unprecedented efficiencies will be revealed. But before those agents are similar IonTo be in style ContricolativeTo be in style Fortune tellerAnd start-up tons, they admire, some of their customers are skeptical that these tools can ship, at least right now.

“Many customers report the gap between marketing and reality,” newly said. Report from CB idiomsthat analyze the main pain points of this product.

Throughout March, the Central Bank’s views examined more than 40 customers of AI agents and found themselves in reliability, integration and security. Recent recent events have highlighted some of the same problems. For example, there was an increase in the excitement of Manus, which was charged as the first autonomous “chief agent” and praised himself as China’s deep moment.

The idea of ​​a tool that can have independent and accurate orchestral and complete complex tasks, it makes sense to strive, and it can be achieved. But the current reality is that customers turn indefinite waters, and the use of the term “agent” causes confusion about what users can expect.

(UN) Reliability is mental

Deep Founder and General Director Demisses Hasabis Recently offered a insidious description of the reliability of the reliability of AI agents, comparing it with the complex interest.

“If your AI model has a 1% error rate and you plan more than 5,000 steps that 1% join the complex interest rate,” he said this week. Luck Event by: Computer weeklyA number of he continued to describe how the 5,000 steps left for work on time, the answer is “accidental”.

For companies that need to be provided with accurate information and serve their customers, randomly is not acceptable to accuracy. CB Insights reported reliability as anxiety among customers using AI agents, almost half, citing it as a problem. The customer described that partially recycled information and hallucinations are located from the AI ​​agent, for example.

Customers also have problems integrating AI agents in their existing systems. The lack of intergovernmentalism has long occurred in the world of headaches in the world, but integration with AI agents is one type. “It was a small gaming game, which we registered for a product where they did not have all the integration we wanted,” said the Central Bank’s ideas.

New SWATH SWATH

Security also leads the list of customers’ concerns and for a good reason. Having technology connect to different systems that contain sensitive information and take action autonomy, open risks. Garthes predict In 2028, 25% of the violations of the enterprise will monitor the abuse of AI agents both internal and external and harmful actors.

“Without proper management, AI agents can and carelessly present sensitive data, do not make unauthorized decisions, or create appropriate blind spots,” he said.

He said the best funds could be safe with AI agents, avoiding products that are not transparent as AI agent makes a decision. Companies must also be piloted in the pilot areas so that they can identify the risks and adapt to the need.

What is even ai “agent”?

The Market of AI is equipped with equipped, especially customer support and encryption. At the same time, “No one knows what the hell is the AI ​​agent” as Techcrunch In the history published last Friday, claiming that the term became “diluted to the degree of meaninglessness.”

Each company defines a slight difference “AI agent”. Some generally use the term to link full autonomous AI systems that can perform tasks on their own, while others use predetermined work flows. Some offer other definitions. And some, such as Openai, seem often changed and contradict their own initial definitions. Many instruments called “Ai Assistants” used are now called “Agents”.

For IT leaders, this border chaos creates messy and deployment headaches. Not only it is difficult to understand what products are doing and how they work, but it is impossible to compare comparison of benchmarks and performance measurements.

None of that means companies don’t start to see some benefits from AI agents. But a reminder that they are still very The early days of this technology, and the hips are going well before real.

And with that, here’s more AI news.

Sage lazzaro
sage.lazzaro@consultant.fortune.com
SageLazzaro.com:

This story was originally shown Fortune.com