Introducing Red Hat
At VentureBeat’s recent AI Impact event, where the discussion focused on what separates businesses that are measuring agency AI from those that are driving it, Brian Gracely, senior director of strategic portfolios at Red Hat, detailed what companies are actually stepping into once agents reach production.
He focuses on cost control, security blind spots unique to autonomous systems, and organizational conflicts that determine whether agent adoption spreads beyond initial champions.
Businesses overestimate how far behind AI agents are
Many business leaders, especially those following key industry names and AI announcements, worry that they are already dangerously behind competitors deploying agents at scale. But according to Gracely, much of that concern reflects a misconception of how quickly organizations learn when they start building. Teams often climb the learning curve much faster than they expect.
However, that rapid development creates a different challenge. As agent use increases, AI costs rise rapidly, turning cost management from an engineering issue into an ongoing boardroom conversation.
Agent use of AI is orders of magnitude higher than chatbot time, making AI a growing concern for businesses. At the same time, organizations are increasingly aware of their dependence on a small number of model suppliers. According to Gracely, that combination is driving many businesses to explore alternatives that give them the ability to control costs and infrastructure.
"The top two or three suppliers are already telling the market that they are losing money, and they are trying to go public to close those gaps," he explained. "Sometimes, depending on that means you will buy at a higher price, or you will find other ways to control what you do."
Right-sizing AI models are a quick lever to cut agent costs
A major cost problem is that businesses overspend by choosing the most capable model available without the complexity of the job.
"When I’m trying to settle an insurance claim, I don’t need to know about the history of Western civilization in my model, I don’t need to know the World Cup soccer scores," Gracely said.
Semantic routing is a method that many companies use to make that judgment automatically, categorize requests and send each one to a job-sized model without requiring users to choose, while infrastructure techniques such as repeated caching queries determine how often the request needs to access the GPU computer at all. Together, he said, these tools dispel the notion that efficiency and innovation pull in opposite directions.
"There’s more you can do at the GPU infrastructure level, and less you can do about model flexibility," he explained. "Those provide the best choice in terms of the levers you’re trying to pull, whether you need efficiency or need innovation. That shouldn’t be a binary choice."
The financial discipline required for the use of tokens is similar to the FinOps processes that have taken years to mature to manage the use of cloud computing. Those basic structures will be transferred even as the vocabulary changes, Gracely said, especially as organizations push for internal education about choice models so that teams stop switching to more prominent choices for jobs they don’t need.
"In the same way that we started teaching finance people what an EC2 instance is and what an S3 bucket is, you’ll have to start explaining tokens to them," he said. "We don’t always need a Rolls-Royce. We don’t always need caviar, because we try to make basic types of things."
The speed of patching is now critical as AI tools detect vulnerabilities quickly
AI-powered vulnerability detection is forcing businesses to rethink how quickly they can detect, verify and issue patches. Time-based patch management cycles may no longer be fast enough in an environment where AI can uncover – and attackers can exploit – new vulnerabilities very quickly.
"Most companies will probably have a window of between seven and 14 days to stay ahead," he said. "There are groups, including Red Hat, that will develop patches for these, but the embargo window will be short."
AI is also changing what defenders need to look for. Instead of simply uncovering critical flaws in isolation, AI security tools can identify combinations of smaller vulnerabilities that only become dangerous when tied together. As both software complexity and vulnerability discovery accelerate, Gracely argued that the ability to quickly manage and update software becomes a strategic rather than a practical skill.
Subject matter experts and compliance teams decide whether agents are rated
Ultimately, organizational acceptance comes down to the need for deep, ongoing involvement from subject matter experts whose knowledge the agent is meant to encode, making gaining their buy-in a necessity rather than an afterthought.
"You have to think about incentives, what you do for the people who participate in the project so that they don’t feel threatened that you will take away their work, and how you encourage people in the long run to cooperate with that change," he said.
Sponsored articles are content produced by a company that pays to post or has a business relationship with VentureBeat, and is always clearly marked. For more information, get in touch sales@venturebeat.com.
