CEOs from Zillow, Experian, OKTA and Salesforce says that artificial intelligence agents are more than just noise, but the big questions about security and trust remains

I am on my way back from Fortune Brainstorm Tech, where I spent the week in the Park City Mountains, Utah, with about 300 executives who focus on artificial intelligence-and still think about artificial intelligence agents. (maybe He should Think about the wind in greenery and the first hints of autumn color, but this is life on the rhythm of artificial intelligence.)

One of the most popular paintings to which you undergo artificial intelligence agents: independent software systems that can imagine their environment, decisions, and make action towards targets without small or non -human intervention. Full disclosure: The session was sponsored by the salesforce, which has a large space share with its AgentForce platform.

But the conversation exceeded one degree for one company. Leaders of Zillow, Experian and OKTA joined to discuss how Agency AI began to crystallize within large institutions. Their consensus: agents are not only Chatbots. Although it is still in the early days, it represents a shift from the tools that simply follow instructions to the systems that can work in the context – practically within the appointed handrail companies.

Shaybani Ahuja, the first vice president of the Information Technology Strategy in Salesforce, participated in what I consider to be a really useful “entitlement model” that helped me to understand the development of artificial intelligence agents within institutions companies:

• Level 0: Copilots that we all know – such as ChatGPT, CLAUDE or GIMINI that recalls information and responds to demands.
• Level 1: Agents who are not only surface information but also recommends the following procedures (for example, suggesting how to solve customer support tickets).
• Level 2: Agents who start dealing with routine tasks independently in silent data environments-renewal of meetings, sending emails to follow up, or withdraw from internal calendars and email systems.
• Level 3: The real transformation point – factors that can organize workflow across multiple areas. Think about an agent managing the sales pipeline by combining CRM data and supporting tickets and financial reports to provide a full customer image.
• Level 4: The ambition is still largely-a scientist of the vocabulary of the agent “from any person”, where the systems speak with each other through organizations. Imagine the agents from various platforms that cooperate to address requests, manage stocks and direct customer notes in the actual time.

Nicholas Stevens, Vice President of Home Loans and Home Loans in Zillow, gave a tangible example of the location of the company on this meter-which seemed to me, to the level of 1 of the 1st level.

Initially, Stevens explained that the tools are essential – the classification of calls or the formulation of a message that the human agent can edit. But over time, they start taking more independent tasks, such as sending a prior approval letter or reserving a tour. He said that progress is not only related to technology, but also for the user’s confidence. Many real estate agents prefer to review and allocate outputs early. However, the system learns its accent, population, and even emojis preferences, they become more comfortable in handing over responsibility.

Kathleen Peters, chief innovation at Experian, said that the company has already has conversation agents recommending procedures and moving them now – for example, walking a customer through the steps to enhance the degree of credit. On the B2B side, Experian agents help financial institutions in raising risk in loan and credit cards, and transfer decision -making strength from data science teams to product managers and even retail bankers. Peters said the next step is “agents talking to the agents, especially with a number of our partners like Zillow.”

One of the things that I feel particularly concerned, though, is the safety of the agent – especially as companies begin to implement protocols such as MCP (model context protocol), a frame that explains how LLMS connects to sources and tools of data, external services and interaction. When I asked how much in the room, they already implemented MCP in their organizations, many raised their hands. Bhawna Singh, the chief technology official on the cybersecurity platform OKTA, put my anxiety in its correct perspective: indicated that before MCP, there was no joint understanding of the start of artificial intelligence factors. While she was in the early days, she admitted that the development of Augleic AI opens the surface of a new attack that enhances the danger of malicious or consistent agents, getting a standard is a big step forward. Obviously, though, many institutions will not be able to reach level 3 or 4 level without more confidence in the safety and security of protocols such as MCP.

AHUJA from Salesforce agreed that the future was not related to the construction of “wild and free” agents. Instead, it is a matter of appointing inevitable parameters for things that must be consistent and reliable, while allowing adaptive thinking as the most richer and reliable answers are valuable.

I was a little surprised by the seriousness of these executives on the journey towards the implementation of artificial intelligence agents across industries. They seemed realistic but optimistic, although there are a lot of noise there – only six months ago, my former colleague Sage Lazarro books Taking into account the artificial intelligence that some agents of the artificial intelligence agent said that the reality does not match the noise. Perhaps it is like some small trees that were swinging outside the hotel window in Park City: growing quickly, not completely mature, but still worth watching.

However, here are the rest of the news of artificial intelligence.

Sharon Goldman
[email protected]
Sharongoldman

Our wealth

COHERE is looking to get rid of his weak position by renting the star AI and the new financial manager, and evaluating 7 billion dollars – with the “Return on Investing on AGI” Sharon Goldman

For Openai’s great revenue, research segments instead of “big marketing efforts” are the key to winning customers Sharon Goldman

AI to start Custpai Chemistry, $ 100 million in financing new investment capital Jeremy Khan

Amnesty International in News

Microsoft to buy artificial intelligence from anthropoor in a partial transformation from Openai. Microsoft is separated from its deep dependence on Openai by bringing Claud models in man to Office 365, according to Information, which determines its largest step so far to diversify artificial intelligence providers. In the coming weeks, the company will announce that the Office Copilot features in Word, Excel, PowerPoint and Outlook will be operated with a mixture of anthropologist and Openai. Executive officials say that the Antarbur models, which were hosted on AWS, outperformed Openai in the main areas such as generating financial jobs in data schedules and producing more polished PowerPoint floors. Although this shift can be considered a tactic of bargaining amid tense negotiations with Openai on the ambitions of restructuring and public subscription, the insiders emphasize that Microsoft is simply running with the best -performing models.

Oracle Spercket shares when announcing huge AI deals. the Wall Street Journal I mentioned Oracle shares increased nearly 40 % after the company revealed a series of contracts related to Amnesty International with millions of dollars, indicating much larger in the cloud infrastructure race than expected. The software giant recorded $ 455 billion in the revenue of the distinguished contracts for the quarter ending on August 31-more than four times last year-and the additional Mega tools could pay this number soon over $ 500 billion. SAFRA Catz described it as a “amazing quarter”, citing the request for OPENAI, Meta and Elon Musk’s Xai, among others. Analysts noted that the accumulation, which includes the 500 billion Supercux SuperCURCURCER project with SoftBANK and Openai, makes the long -term Oracle revenue targets to be largely achieved. Once it is seen as a cloud delay, Oracle now determines that he is a major player in providing the infrastructure of artificial intelligence, with the help of Larry Ellison’s close relationships with technology leaders such as Elon Musk and Nvidia’s Jensen Huang.

Artificial intelligence is coming to YouTube creators. according to Reporting from Atlantic OceanYouTube creators such as Jon Peters, who built an audience of more than a million subscribers by sharing wooden educational programs, are now facing the statute of limitations as their videos are quietly scraped to train artificial intelligence. The researchers have identified more than 15 million video clips on YouTube in the groups of artificial intelligence training data – often without the knowledge or permission of the creators – including educational content, news and educational. Technology companies such as Microsoft, Meta, Amazon and NVIDIA have used these data collections to develop generator video tools that can surpass custom clips, which raises legal questions that have not been resolved about copyright and fair use. While companies argue that this practice is legal, critics warn that they may undermine the livelihoods of creators, as videos created from artificial intelligence compete with the authentic work of man -making and even.

You have a calendar

6-10 October: world Amnesty International The week, Amsterdam

October 21-22: Teddy San Francisco. Apply to attendance here.

2-7 December: NeuPIPs, San Diego

December 8-9: Wealth storm Amnesty International San Francisco. Apply to attendance here.

Eye on artificial intelligence numbers

116 %

This is the amount of consumer spending on artificial intelligence tools that rose on an annual basis in the first half of 2025, according to what is mentioned a report From the consumer edge intelligence company.

Openai tops the package, as 40 % of the market calculated as subscriptions in Chatgpt Pro, more than 1000 % jumped over the past year. But some of the struggles acquire land: the search tool grows in the artificial intelligence of confusion, while Xai from Elon Musk has already reached two -thirds of its scale.

The biggest video shift, as it makes tools like Runway and Cistideo it is easy for anyone to create explanation clips or even cinematic scenes, which leads to more than 50 % growth in this category. Artificial intelligence also finds a house in the workplace: applications like OTER.AI and Fireflies.AI have become daily comrades, while sanitary ware and engineering expansion at Breakneck.

The report says that ready -made meals are clear – consumers no longer try only. Artificial intelligence has become an essential component of how people work, create, and spend.