Agency AI occupied the lead in the worlds of artificial intelligence, technology and business, as it dominates the discourse and enhances the pressure on companies to quickly integrate technology or fail their competitors. More than anyone else, it is the major artificial intelligence officers (Caios) who were directed to the dismantling of promises and facts in the latest word of artificial intelligence.
While overseeing experimentation with AI’s signs and directing other leaders on this trip, Caios also moves via noise, interests about security, trust, interconnection (or lack of) from these systems. Not to mention having to deal with the question: What even He is Artificial intelligence agent?
The matching of the noise causes companies to lose concentration
No one can agree on what the term “artificial intelligence agent” means, such as luck and I am free I mentioned. Companies define the term differently and often use it to describe various features and capabilities, including many of them previously described as other terms such as “artificial intelligence assistants”. to Tone The chief artificial intelligence officer because Joan, who led the construction of the solution of the artificial intelligence agent called the AI refinery for customers and also works directly with the publication of their artificial intelligence agent, and this has caused it to be dedicated to a great deal of time to help customers sort through contradictions.
“A year ago, everyone was saying,“ I needed Gen AI. ”Everyone says now,“ I need to work AI or AI Agency. ”She said that many of these things are still such things, at the end of the day, they are still the same.” Thus there is a lot of confusion in the market with our customers, “What is the agent of artificial intelligence? What did he publish?” And so we spend a lot of time in education. ”
The fugitive impact of this is that companies are rapidly revolving in the so -called artificial intelligence agents “only for the press statement.” Clicking on the answer to the moment AIC AI is to make some companies rename features or chase artificial intelligence agents to stay in the direction, and it is often just creating thin layers of factors at the top of the basic models.
“I see it completely. I see it every day.” “This is why this space is full of noise.”
Security, errors and confidence dominate risk analysis
Despite the noise and disorganized terminology, the basic idea of artificial intelligence factors – distinction systems designed to take independent measures to carry out specific tasks – still generate a lot of justified excitement. It is also a key to creating types of systems and science fiction technicians always dreamed of, able to carry out the sequence of complex tasks through multiple platforms on our behalf. But there are real road barriers.
URI Yerushalmi, founder and chief of Amnesty International in Fetcherr, who uses artificial intelligence for alarming pricing in the aviation industry, believes that the opportunities surrounding artificial intelligence employees are “enormous”, but opening this value depends on facing real challenges about confidence, integration and avoiding failure points. For example, agents must integrate with old systems and compatible with restrictions in the real world without disrupting the current workflow. While we give agents more autonomy, we need to build handrails, monitoring, transgression, and human alignment mechanisms.
He added: “The companies need to trust in the decisions of the agent.” “This requires transparency, consistency and return on investment.”
One of the most common failure points is to double errors. Google Deepmind CEO Dimis Hasabis Compare this problem to double interest In financial affairs, he explained that even if the agent model has a mistake of only 1 %, it will lead to the interaction of a series of errors that, after a few thousand steps, eventually the possibility of a completely correct “random” result. Bonat refers to this problem of doubling errors as a severe challenge in terms of confidence in artificial intelligence agents, saying that these capabilities are to include one mistake without being aware of it “can be created.”
This is particularly applied to the type of multi -agent systems that many companies think, which Joan said can cause blind spots and get a problem very quickly.
She said, “It may not work for you, and it may actually provide a lot of risks,” she said. She said, “Although her clients have a strong appetite to see the influence of artificial intelligence agents, they are also warned of high cloud bills and security risks.”
Security concerns are sure to lead the mind in the scene of the agent of artificial intelligence. By 2028, Gartner predicts25 % of institution violations will be returned to artificial intelligence agents, including ill -treatment of internal and external actors. The dominant factor that contributes to the security risks is a mixture of self -government and the interview with intended inter -agents, which would make them communicate and exchange data with them, and work independently through a wide range of platforms and systems. In other words, the exact nature of how these systems work and what it aims to do is what makes them very risk.
Dreams
Like all Caios, Ali Al -Ashaji, A senior Amnesty International and technology employee at the Omnicom Precision Marketing Group, is concerned about data leakage and other security risks. It is also concerned that many companies that market agents systems use security as a comfortable excuse to increase their customers within their ecological systems, and to face the cooperative and decentralized vision that many see as essential for the future of an agent: “not because it cannot be resolved, but because it is not in the commercial interest of the seller to solve it.”
He said: “Every seller builds” the framework of his agent, “but no one is permissible for inter -employment at the level of institutions. Without open work frameworks and semantic standards, we only build more intelligent silos,” adding that the agent’s cooperation protocols do not remain immature and that it is frustrated to see the main sellers and their double in the walls.
Deluette AI Jim Rowan president sees this play among his clients, noting that they are often adhering to the current service providers and using the capabilities of their agent when they are issued. It is another repetition of Platform feature This drives growth for service providers such as Openai and Google and Microsoft While they are on their current customers in pipelines and their new products.
Rowan said: “There is a specific tension in the market between those who want to own the agent’s registration system. Like, who will own the record, who will organize all the orders that occur around the agents.” “We see that with the presence of hyperactors and saas service providers and startups in the third tools in space as well. I think that the jury is still outside the one who owns it.”
Correction, September 3, 2025: I made a previous version of this story in the name AI AI AIC solution.
https://fortune.com/img-assets/wp-content/uploads/2025/08/FORTUNE-AI-PLAYBOOK-CAIO-Agents.jpg?resize=1200,600
Source link