A growing concern about costs and governance in the use of artificial intelligence agents is leading large companies to a “more professional phase” based on structured projects and scale in the implementation of tools. The person carrying out the assessment is Eduardo Campos, vice-president of the Technological Solutions area at Microsoft Brazil ().
In recent years, the orbit of conversations about generative AI has migrated from chats to the use of agents: systems capable of carrying out tasks autonomously and that can be created only with text instructions from the user. It turns out that, in some corporations, the possibility of developing tools without using programming codes favored the proliferation of applications created by business areas that were not information technology.
“These low-code tools allow there to be no dependence on the technology area, but at the same time, they also do not have full support from the technology area for a super important part: the security and observability brakes”, comments Campos.
Study abroad
Upgrade your career!
Platforms developed by Microsoft notify malicious actions that may attempt to be executed via agents, for example. Furthermore, the tools notify those responsible for IT areas about potential risks in use. “We are now evolving into a more professional phase. More structured projects, with scale and governance.”
Interestingly, the proliferation of the use of AI in companies began precisely in the areas of software development, where major gains in productivity were noted. “The developers’ productivity skyrocketed, not necessarily the quality. People were writing a lot of code”, says the executive.
Until the cost element came into play. To perform their tasks, AI agents use the language models through which they are created — ChatGPT and Copilot, for example. The cost for it to perform an activity is represented in tokens, types of fragmentation of the demands made on the AI. For each token, there is a cost.
Continues after advertising
“As companies begin to scale to serve millions of customers using AI tools, token consumption becomes an evaluated element”, points out Campos. One of the ways to reduce costs is to provide options for companies to choose the most optimized for specific tasks: Microsoft offers, for example, over 11 thousand of them on its platform, such as ChatGPT, Claude (from Anthropic) or its own solutions.
Another solution on the cost side is to define limits for the use of AI in different areas of the companies’ business: a monthly number of orders that someone can make or agents that an area can create, for example. “There are several stories in the market of companies that had a certain token budget and consumed it all in a few months”, says Campos.
