Open a chat of general artificial intelligence, ask a question, ask for a text or summary of an article. These are features that have become common for users, but the next step for companies, at least in the evaluation of Google Cloud (), is in another application: agents.
The theme is central to the company’s next steps in the development of AI and cloud computing solutions to companies, from chips through proprietary applications and platforms for third parties.
In the words of Google Cloud CEO, “agents are smart systems that demonstrate reasoning, planning, memory and the ability to use tools.” “They are able to think several steps ahead, use tools, including working with software and systems to perform tasks on their behalf and under their supervision,” he said in a presentation at the Google Cloud Next 25 ′ event last week.
Infrastructure
Among the main announcements of the event, dedicated to processing steps related to agents applications.
Ironwood is dedicated to inference, when language models already trained need to perform an action. In AI chats, for example, the model responds to user commands.
It turns out that with agents, results are proactively generated at all times. According to Morgan Stanley analysts, 75% of all AI date demand demands in the coming years should be centered on the inference process.
Continues after advertising
Multiagentes
In the not -so -distant future, Google Cloud believes hundreds of these agents will operate simultaneously with company operations and interact with each other. “Multiagent communication, multiagent interaction, multitapas workflows that may require 100 agents operating on behalf of various human functions. These are the things we are now focused on so that when the next evolution comes, the resources we have on the platform are ready,” said Google Cloud vice president and CTO, Will Grannis, in an interview with Infomoney.
For companies, however, the connection of internal data for consultation agents such as PDFs, images and emails is not simple. In addition to a shortage of AI experience professionals, many of their data are stored on their own servers rather than in the cloud.
Cloud has been trying to address the theme by carrying its agentSpace, a hub that connects AI agents of various models with corporate data, to the company’s private data environments. “Behind all this, we are making data more available,” says Grannis.
Continues after advertising
The company has also announced a protocol called A2A, which allows artificial intelligence agents to communicate and define the limits of their interactions with users through texts, forms, audios or videos. Companies such as PWC, Accenture, Deloitte, SAP and Oracle are already partners of the company for the implementation of the protocol.
