AI agents continue to drive tech industry—for better or worse

AI agents have been a hot topic for IT teams this year, and as they rise in importance, more companies are issuing tools to manage the secure deployment of the technology.

A recent survey from Optro found that implementing AI agents comes with negative effects. Around a third of organizations (34%) said they had acted on improper information from an agent. But stats like that aren’t necessarily impacting how agents are set loose in the organization, as 45% of respondents reported they know or suspect that deployment is often done “without formal approval or review.”

Optro GM of AI Governance Guru Sethupathy told IT Brew that as agents connect across products, compliance networks, and risk frameworks will require layering governance. Confidence in guardrails is key; developers are often faced with a proliferation of agents both inside and outside the enterprise.

“Almost every vendor, third party out there, is building agents,” Sethupathy said. “That is not the differentiator going forward. The differentiator is agents that customers can trust.”

Shifting sands. There’s a disconnect between promise and reality with AI, said Egnyte CTO Amrit Jassal, one that can be confusing and frustrating. Automation promises are running into the reality that in the process of shifting existing workflows you’ll expose your insufficiencies—with the realizations of what’s required that will come alongside that.

“The industry is still not where it needs to be—understanding what content you have, who actually has access to it, etcetera,” Jassal told IT Brew. “And in some cases, you will find that AI, because it’s such a cross-cutting type of a technology, will start exposing those seams that you have in your governance.”

Step by step. Building AI solutions is easy these days, with no-code (or vibe-coded, depending on your preferred terminology) agents a reality. But while the ease with which they’re created adds to productivity, the proliferation of agents has turned into a flood of everything from big SaaS platforms to AI-powered internal tools.

“What we’re starting to see now is a significant gap between how quickly the technology is moving, the capabilities are moving, and even adoption is starting to move, as opposed to where governance is sitting today,” Sethupathy told IT Brew, adding, “That gap is not sustainable; that gap is going to lead to huge risks, to some really bad outcomes.”

Agents tend to have a narrow scope, Jassal said, which means it’s important to have maximum visibility into their operations. Introducing new capabilities means enterprises need to consider how best to deploy them, as Egnyte is doing with its new AI control center and context studio products.

“The whole idea for us, as software providers, is to make sure that we have the right interfaces and the contracts in place where we put these [guardrails], and so we are always monitoring,” Jassal said. “We get this additional complexity because of AI, but then AI is also helping us actually deal with that complexity, so it really cuts both ways.”

Jeff Stewart, group VP of product management at SolarWinds, said that after OpenClaw showed an increased capacity for agents in the enterprise, thinking changed about how to use AI tools to architect systems.

“If you go back 12-plus months ago, we got introduced to OpenClaw and the ability of autonomous agents and actions that they could take—super powerful, but also capable of destructive behaviors,” Stewart said.

Blowing in. SolarWinds recently debuted a new system, Avior, aimed at establishing better guardrails for enterprise agents. Given the company’s history—including the infamous software update in December 2020 that sparked a massive cyberattack—establishing trust is important.

“There’s no better way to understand how to earn trust than when you’ve lost trust…but out of that came some really good things for both our customers and SolarWinds, and frankly the community, because we were very transparent,” Stewart said.