AI in 2025 was defined by increased accessibility, security, and agents

After a few years of big hype and even bigger promises, the AI sector in 2025 settled into a more mature mode, with advances in agentic technologies and a more developed tech stack.

That’s partly because of advancements in generative AI, said Melissa Ruzzi, AppOmni’s director of artificial intelligence. She noted that a reduction in latency has led to quicker turnarounds, making GenAI less of a tool and more of a conversation partner. This has led to higher capabilities—though, Ruzzi warned, there’s still a lot of hype.

“We still see a lot of the companies out there that are joining AI right now, and they’re jumping to GenAI—that’s it, that’s all we’re going to do, that’s all we’re going to need,” Ruzzi said.

As IT Brew reported this summer, low-latency is a priority for companies looking at AI adoption. Network connectivity is likewise important, Expereo CIO Jean-Philippe Avelange said: “AI starts putting strain on the network because it’s much more data-intensive. Each affects each other, because the increase of load affects the latency.”

Increased accessibility to AI models has led to a rise in usage, Globant SVP of Digital Innovation Agustín Huerta told IT Brew. That has been behind an explosion in how the technology is deployed, from robotics to security. And in 2025 there was a further development in model accessibility as companies like Google, AWS, and Microsoft consolidated around model context protocols, tools that allow large language models to connect with external data sources.

“They are not competing with each other anymore, in that sense, in terms of creating protocols,” Huerta said.

Shifting over. Another big shift? The move to agentic AI. As IT Brew found earlier this year, AI agents were the talk of the RSAC conference in San Francisco. It seemed nearly everyone we talked to had an opinion on how AI agents were going to change the industry.

David Brauchler, NCC Group’s technical director, said recently that while it was exciting to watch AI evolve from chatbots to “multi-user integrated, complex operations,” there are still some concerns about the security underlying the technology.

It’s great that AI agents can act autonomously, Brauchler added, but that doesn’t mean systems should be left without proper oversight.

“This is the year where organizations are finally starting to realize that we can’t just rely on AI to self-govern or police its own behavior,” Brauchler said.

Secure process. AI is having an effect on all aspects of cybersecurity, including budgets. Brauchler described it as “a force multiplier for threat actors” that has the “ability to take a small amount of input and turn it into a very large amount of output.” That trend increased throughout 2025, thanks to rising use of agentic AI by attackers and defenders.

“AI is a powerful tool to automate what would otherwise require a user or an employee to come in and look and say, ‘Okay, this system isn’t operating as we expect it to,’” Brauchler said.

It wasn’t always that way, said Securin CEO Srinavas Mukkamala—decades ago the security framework required a lot more human intervention—but 2025 was a year where AI became more widely used as both a cybersecurity and cyberattack tool. That meant securing systems against attack, even as they help defend organizations.

“When you look at the majority of the new AI models that are released, the agentic frameworks that were released, MCP servers that were released, they all had very common weaknesses you would see in non-exploitable vulnerabilities,” Mukkamala said.