The great artificial intelligence agent accelerated: Why does the institution adopt faster than anyone expected

Gossip around artificial general intelligence (AGI) may dominate the main headlines from Silicon Valley companies such as Openaiand Dead and xiBut for institution leaders on the ground, the focus is directly on practical applications and measurable results. In Venturebeat recently Transfer 2025 The event in San Francisco, a clear picture appeared: the true AI era that was published here, accelerates, and reshapes how companies already work.

Companies like Intuitand Capital is oneand LinkedInand Stanford university and HighMark Health Artificial intelligence agents quietly put in production, address concrete problems, and see concrete returns. Here are the four largest fast food from the technical decision makers event.

1. Artificial intelligence factors move to production, faster than anyone who realizes

Institutions are now publishing artificial intelligence agents in applications facing customers, and the trend is accelerated at a break. A recent survey of Venturebeat, which included 2000 professionals in the industry that was conducted before the VB converted, revealed that directly 68 % of institution companies (With more than 1,000 employees) They already adopted AIC – a number that looked high at that time. (In fact, I was concerned about being very credible, so when I announced the results of the event, I warned that high adoption might be a reflection of the specific Venturebeat readers.)

However, the new data checks this rapid shift. A KPMG survey It was released on June 26, one day after this event, this appears 33 % of organizations now publish artificial intelligence agentsThree times a sudden increase from 11 % in the previous two quarters. This market turns into the health of the Trend Venturebeat that was first identified for the first time in only weeks in pre -conversion survey.

This acceleration is fed with concrete results. Ins Williechief executive officer New residueAgainly note 30 % quarter of a quarter growth In monitoring artificial intelligence applications by its customers, this is mainly due to the movement of its customers to adopt agents. Companies publish artificial intelligence agents to help customers automate the workflow they need. IntuitFor example, the generation of bills and the reminder of the agents of its Quickbooks program. The result? Companies that use this feature Get salaries faster five days and more vulnerable to full payment by 10 %.

Even non -developers feel the transformation. Scott WhiteDriving the product from Antarbur Claud AI’s product described how, although it is not a professional programmer, It is now building ready -to -produce programs. “This was not possible six months ago,” he explained, with highlighting tools like Claude Code. Similarly, Openai’s The head of the product for its API platform, Olivier GodmentDetailed how customers love tape and fund We are Using SDK agents to build multiple agents.

2. Performance racing has no clear winner like a multi -style multi -model era

The days of betting on one large language model (LLM). There was a fixed topic throughout the Transform 2025 that was the move towards a multi -model and multi -party strategy. Companies want flexibility to choose the best tool for this job, whether it is a strong ownership or an open source alternative.

like Armand RuizVice -head of the artificial intelligence platform in IBM Make upThe company’s development of the model portal – which directs applications to use any LLM is the most efficient and performance of the specific condition – was a direct response to customer request. IBM started by providing its own institutions clients, then added open source support, and finally realized that she needed to support all models. XD Huang chanted this flexibility desire, CTO OF Zoom, whose company described Three -level model approach: Supporting the royal models, providing their own model to control and allow customers to create their own versions.

This trend creates a strong but restricted environmental system, where the graphics processing units and energy needed to generate symbols are limited in the width. like Dylan Patel to Half -solutions And the teammate of the committee Jonathan Ross to clear and Sean is a lie to Brains IndicateThis puts pressure on the profitability of many companies that simply buy more distinctive symbols when they are available, instead of locking in profits with a continued low cost of these symbols. Institutions have become more intelligent on how to use different models of different tasks to improve both cost and performance – this may often mean not to rely only on NVIDIA chips, but being more customized – echo also in a VB conversion session led by by Solidigm around The emergence of memory and storage solutions for Amnesty International.

3. The institutions focus on solving real problems, not AGI chase

While technology leaders such as Elon Musk and Mark Zuckerberg and the German Medal are talking about the dawn of cancellation, institution practitioners operate their sleeves and solve immediate business challenges. The conversations were actually based on the conversion.

Take HighMark Health, The third largest integrated health insurance in the country and a company. Senior data staff Richard Clark He said that Using llms for Practical applications such as multi -language communication to serve their best customer base, and simplify medical claims. In other words, benefit from technology to provide better services today. Similarly, Capital is one He is Building teams of agents who reflect the company’s jobsWith specific agents of tasks such as risk evaluation and audit review, including helping auto agents to connect customers to the correct loans.

The travel industry is also a practical shift. Ctos from Exxpia and Kayak boats Discuss how to adapt to the new search forms enabled by LLMS. Users can now search for a hotel with “Infinity” on Chatgpt, and travel platforms need to merge this level of discovering the natural language to maintain competition. The focus is on the customer, not technology for it.

4. The future of the artificial intelligence difference is small, smart, and enables

The age of artificial intelligence agents also transforms how to organize the difference. The consensus is that a small “difference” of three to four engineers is the most effective. Farun Mohanchief executive officer SurfingCustomer agent, fast -growing, The event was launched by controversy This small team structure allows a quick test of product hypotheses and avoid a slowdown that affects larger groups.

This transformation means that “everyone is creative”, increasingly, “everyone is the director of” artificial intelligence agents. like Gyrroup and Atlasian Note that the engineers are Learn now to manage the agents of agents. The required skills develop, with a more focus on clear communication and strategic thinking to direct these independent systems.

This acceptance is supported by increasing acceptance of sandy development. Andrew NangerA pioneering voice in artificial intelligence, advised the attendees to do so Leave safety, governance and observation until the end of the development course. Although this may seem intuitive for large institutions, the idea is to enhance rapid innovation within an controlled environment to quickly prove value. This feeling was reflected in our survey, which found that 10 % of the organizations adopting Amnesty International does not have the intended Amnesty International Salama teamThis indicates a willingness to set speed priorities in these early stages.

Together, these fast food draws a clear picture of the AI scene of the institution that ripens quickly, and is transmitted from a wide experience to concentrated implementation, which depends on value. The conversations in Transform 2025 showed that companies are publishing artificial intelligence agents today, even if they have to learn difficult lessons on the way. Many have already experienced one or two axes since they first tried to get out of artificial intelligence a year or two before – so it is good to start early.

To get more diving in these topics and more analysis from the event, you can listen to the full discussion you have with the independent artificial intelligence developer Sam Witteenen on the last podcast below. We have also downloaded the main stage conversations in the VB conversion here. And our complete coverage of articles from the event here.

Listen to PodCast Takenways Transform VB with Matt Marshall Wissal and Vivo here:

Editor’s note: As a thanks for our readers, we opened the registration of early birds for the VB Transform 2026-200 dollars only. This is the place where the ambition of artificial intelligence meets, and you will want to be in the room. Book your place now.

Daily visions about business use cases with VB daily

If you want to persuade your boss at work, you have covered VB Daily. We give you the internal journalistic precedence over what companies do with obstetric artificial intelligence, from organizational transformations to practical publishing operations, so that you can share visions of the maximum return on investment.

Read with us privacy policy

Thanks for subscribing. Check more VB news bulletins here.

An error occurred.