Some AI agent customers say the reality does not match the hype

Welcome to Hello and Eye in AI. In today’s publication … Companies with AI agents say technology shortcomings from expectations; Nvidia announces new chips and positions for a post-Teep-Teep view; Join Elon Musk and Nvidia Microsoft-Blackrock AI Foundation; AI Spamers “compellers the internet”; Foxconn appears as the main player in the Global AI race.

One day a day, a new AI “Agent” continues without a technological company, will revolutionize the workflows and open the unprecedented effectiveness. However, as the manufacturers of these agent companies Righteousness, Amazon, OraceAnd the starting tones – it is hating them, some customers approach the doubt that these tools can handle at least.

“Many customers provide information about the space between marketing and reality,” he reads new Reporting CB ConceptsAnalyzing the main pain points that cover these products.

In March, more than 40 of the AI ​​agents were examined and found that there were validity, integration and security issues. Other recent headings stressed some of the same problems. For example, the first autonomous agent was calculated as a “overall agent” and some tests were evaluated as an excitement and an alarm by some until the invalid performance and suspicious speeches were detected.

The idea of ​​a means of a means of means, which can complete an expert and accurate orchestration and complex tasks, makes sense to make an effort, and it is possible to achieve. However, the current reality is the passage of clients and the term “agent” causes confusion about users who are truly expected.

(UN) reliability is superior to the mind

Depth Founder and CEO Demis Hassabis recently suggested a deep description of reliability problems by comparing the reliability issues, covering AI agents.

“If you have a 1% error rate in your AI model and you plan more than 5,000 steps, 1% combinations like complicated interest,” he said. Google Incident, according to Computer Weekly. With the correct response, the correct answer continued to describe how this 5,000 steps were processed.

For companies that should provide accurate information and serve their customers, a random accuracy opportunity is generally not acceptable. CB concepts are the most concerned between clients using reliability, AI agents, about half of them refer to him as an issue. A customer described the possibility of partially processed information and hallucinations from an EU agent.

Along with the integration of AI agents into existing systems, they work in matters. Lack of interaction has long been the headache in the world of enterprise program, but integration with AI agents is the entire point. “It was a game game that we have signed a product that we do not have all the integrations we want,” he said in concepts of CB.

A new marriage from security risks

Security, also a list of customer concerns and for a good reason. A technology contains sensitive information and connecting to a technology to take action, great risks are opened. Gartner predicts By 2028, 25% of enterprise violations will become the abuse of both domestic and foreign and harmful actors.

“Proper management, AI agents can accidentally monitor, make unauthorized decisions or comply with administrative decisions or compliance, CEO of Information and Compliance Company Great ID

According to him, the best road companies can test with AI agents, preventing the products that describe the decisions of the AI ​​agent. Companies should also open the risks, and if necessary, if necessary, the AI ​​agents must be able to adjust as necessary and regulated as necessary.

What is the AI ​​”Agent”?

For AI agents, the market is fed up in special niches such as customer support and coding. At the same time, “no one knows that hell is an EU agent Techcrunch Put it to a story that is broadcast last Friday, the term was “diligent to the point of meaning.”

Each company identifies the “AI agent” as a little different. Some generally use the term to apply for the full autonomous EU systems that can independently execute, while others use it to refer to systems following predefined work flows. Some suggestions are still other definitions. And like some, it is often changed and seems to be contrary to their previous definitions. In the past, many commonly called “AI Assistants” are also called “agents”.

For IT leaders, this praised chaos creates a headache of confusion and placement. It is not difficult to understand what the products do and how they work, but it is not possible to compare criteria and performance measurements.

None of this is that companies start to see somewhat Benefits from AI agents. However, these are still a reminder too The first days for this technology and the hype work better than reality.

And there is more news here.

SAGE LAZZARO
saage.lazzaro@consultant.fortune.com
sagelazzaro.com

This story was first displayed Fortune.com

Source link