Like many companies these days, cybersecurity firm Armis is keeping a close eye on agentic AI, one of the hottest areas in tech over the past few months. Vendors say tools that rely on the technology can perform more complex workplace tasks with little human intervention, thereby saving businesses on labor costs.
But agentic technology is also rapidly evolving, making corporate tech leaders unsure about which tools are best and whether better products will come along later. Costs can also balloon, particularly for those that get carried away by all the sales pitches and internal pressure to adopt a large number of agentic AI products.
“I fully expect to have approximately eight to 10 agentic solutions by the end of April—and hopefully not increase beyond that,” says Curtis Simpson, chief information security officer at Armis.
It's a balancing act that many companies are facing. A quarter of employers will try out agentic AI this year, a figure that will swell to 50% by 2027, according to consulting firm Deloitte.
Simpson plans a hybrid approach in which he tries one AI agent from a large provider—he says one option is Amazon Q, an agentic assistant that can be used by software developers, customer support employees, supply chain analysts, and other specialized capabilities. He also foresees using narrower agentic products, like Salesforce's Agentforce, which is mainly for sales teams. As the technologies evolve and Armis gets a better sense of what's best for the business, that mix will change.
"We can drop more and more of those agents as we progress through this effort," says Simpson.
Cloud-based security company Netskope is also trying to focus on agentic technologies. Mike Anderson, chief digital and information officer, says his team takes at least two meetings each week with existing partners and outside vendors to closely monitor the space. “Everything we’re doing is short-term commitments because the space is evolving so quickly,” says Anderson.
Most of the AI agents Netskope has adopted have been built into workplace communications tool Slack, including a bot that can handle IT help desk requests and another from Crayon, a software company that monitors real-time competitor pricing to ensure Netskope is pricing its products competitively within the market. But workers can face a flurry of messages from the different bots in Slack, adding unnecessary complexity.
Anderson wants to create a Netskope bot that can ingest all employee-to-agentic conversations with the IT help desk, Crayon, and others, so that the new company bot can be trained to retrieve information from the other bots supplied by outside vendors. “I need the bot to rule the bots,” says Anderson.
Kathy Kay, chief information officer of insurance company Principal Financial, who takes pitches from many vendors of agentic AI tools, says she isn’t star struck by the latest industry buzzword. "I keep going back to, 'How do we make humans superhuman?'" asks Kay.
Principal Financial’s view is that agentic AI, copilots, and other niche forms of AI, are all tools designed to help employees work more effectively. What Kay still has to figure out is which agentic AI offerings will deliver the best business outcomes, while also taking into consideration other factors like speed, ease of integration, and cost.
“I do envision we will have more agents in production throughout this year,” says Kay.
Kim Anstett, the CIO at Trellix, says she met with every department at the cybersecurity provider to discuss possible uses for AI agents and came up with a list of over 100. While internal excitement about agentic AI is high, and Anstett is beginning a few pilots of the technology, she also doesn’t foresee buying agentic products from every vendor Trellix works with today.
“From a strategy perspective, initially we will limit the number of add ons we purchase,” says Anstett. “We’re looking at it from a cost perspective.”
Patrick Richards, CIO of fleet management software provider Motive, says half of his time these days is dedicated to meeting with AI vendors, including about internal uses of agentic AI. He prefers narrower agentic applications, meaning he wants to give these tools limited autonomy so that he can avoid acting on hallucinations, or outputs that are inaccurate or misleading.
Richards also intends to work with multiple vendors, which he acknowledges comes with more controls needed to manage data privacy and security. But it also gives Motive greater leverage when it comes to pricing.
“Do I want to enter into a new world where I have vendor lock with one AI platform, and then they slowly raise the price until I’m a frog in hot water and I’m boiling?” Richards asks. By casting a wide net, Richards believes he can "shop for the best agent provider for the actual business problem I'm trying to solve."
John Kell
Send thoughts or suggestions to CIO Intelligence here.
