At VentureBeat’s recent AI Impact event, Brian Gracely, senior director of portfolio strategy at Red Hat, described the practical issues companies encounter when autonomous, or "agentic," AI moves from pilots into production. Gracely focused on cost discipline, security blind spots specific to autonomous systems, and organizational frictions that determine whether agent adoption spreads beyond early champions. The session examined what separates enterprises that successfully scale agents from those that remain in pilot mode.
Executive perceptions: many firms think they are behind
Many enterprise leaders—especially those following industry keynotes and major AI announcements—worry they are already falling behind competitors deploying agents at scale. Gracely argued this anxiety often misreads how quickly teams learn once they start building: organizations frequently climb the learning curve far faster than they expect. That speed, however, creates a new challenge: as usage grows, AI costs rise quickly and cost management shifts from an engineering detail to a recurring boardroom topic.
Costs and dependence on a few providers
Gracely noted that agentic AI usage requires orders of magnitude more resources than the earlier chatbot era. Organizations are increasingly conscious of dependence on a small number of model providers. "The two or three top providers are already telling the market that they're losing money, and they're trying to go public to make up those gaps," he said. That supplier concentration is pushing many enterprises to explore alternatives that give them greater control over costs and infrastructure.
Right‑sizing models and practical levers to cut spend
A primary cost issue is the default tendency to use the most capable model available for every task, regardless of task complexity. As Gracely put it: if you’re simply resolving an insurance claim, you don’t need a model that knows the history of Western civilization or World Cup scores. In practice, companies use several techniques to avoid overpaying:
- semantic routing: automatically classifying requests and sending each to a model sized for the task;
- caching: storing repetitive queries so they do not repeatedly consume GPU cycles.
These measures remove the assumption that efficiency and innovation are mutually exclusive; infrastructure choices and model flexibility offer levers for both.
Financial discipline and token spend
Controlling token‑based spending requires financial discipline similar to the FinOps practices that matured over years to manage cloud compute costs. Gracely said the same foundational frameworks will transfer even as terminology changes, and that internal education on model selection is essential so teams stop defaulting to the most prominent option for tasks that do not need it. "The same way we first had to teach the financial people what an EC2 instance is and what an S3 bucket is, you're going to 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're trying to do basic types of things."
Patch speed: AI accelerates vulnerability discovery
AI‑powered vulnerability discovery is forcing firms to rethink how quickly they can identify, validate and deploy patches. Long‑standing patch management cycles may not be fast enough in an environment where AI can uncover—and attackers can exploit—new vulnerabilities much more quickly. Gracely suggested many companies will have a roughly 7–14 day window to stay ahead of attackers. While groups, including Red Hat, will build patches, the embargo window for disclosure and remediation will be short. He also noted that AI security tools can find not only isolated critical bugs but combinations of seemingly minor vulnerabilities that become dangerous when chained together. As software complexity and vulnerability discovery accelerate, the ability to rapidly manage and update software becomes a strategic capability rather than just an operational one.
Adoption depends on subject matter experts and compliance
Ultimately, whether agents scale across an organization depends on deep and sustained involvement from the subject matter experts whose knowledge the agents are intended to encode, and on compliance teams. Securing their buy‑in is a prerequisite rather than an afterthought. "You have to think about the incentives, what you do for people who participate in this work so they don't feel threatened that it's going to take away their job, and how you incentivize people in the long run to cooperate with that innovation," Gracely said.
Conclusion
Bringing agentic AI into production raises interconnected challenges: rapidly rising costs, accelerated patching timelines, and the need for organizational alignment and incentives. Gracely argued a mix of technical measures—model right‑sizing, semantic routing, caching, faster patching—and organizational and financial practices—internal education on tokens, cost governance and incentives for experts—will determine whether agents move beyond pilots to become widely adopted enterprise capabilities.
