Most companies don’t know what they’re doing with AI • The Register
Contrary to popular belief, you can’t succeed in business (or AI) without really trying. Many orgs are jumping on the AI bandwagon without the infrastructure they need to make it work or track results, Cisco says. Most haven’t even defined what they want their AI agents to do.
The networking hardware manufacturer found in its 2025 AI Readiness Index that most companies are planning to deploy additional AI agents in the next few years, and 86 percent expect it to improve employee productivity within three years, but those expectations don’t necessarily match the reality of what it takes for such an initiative to succeed.
According to Cisco, part of that reality is the need to invest in new hardware, particularly networking gear. Per the company, 54 percent of respondents said their infrastructure can’t scale for rising workloads driven by AI adoption, and just 15 percent said that their networks were “flexible or adaptable” in a way that would facilitate the new AI era of business.
That’s not all, though, and Cisco tells The Register that infrastructure is “only one part of the AI readiness index equation.” The rest, a spokesperson explained in an email, includes strategy, data, governance, human talent and company culture.
Take agentic AI, for example. 83 percent of those surveyed said that their companies intend to develop or deploy AI agents, and 40 percent expect AI agents to be working alongside humans within a year. Putting aside the fact that AI agents are still getting stuff wrong most of the time, very few businesses seem prepared to bring agentic AI into the enterprise fold.
Per the report, only 31 percent of companies say that they’re prepared to control and secure agentic AI systems, while only 32 percent have identified which human tasks they want agents to supplement or take over.
“A wave of technology can leave behind a trail of shortcuts, compromises, and underinvestment that later become bottlenecks,” Cisco noted, and it has a name for that phenomenon: AI infrastructure debt, a new iteration of the old-fashioned concept of technical debt. Instead of shaky code and poorly written software leading to a business bottleneck, AI infrastructure debt is all about not having the necessary things in place to actually make AI investments more than window dressing.
“As history with technical debt shows, what looks like an acceptable compromise in the early phases can snowball into systemic drag,” Cisco said in the report. “Organizations know their infrastructure isn’t ready for the surging workloads, they acknowledge that their security measures are still fragile, and their workforce plans are out of sync with the technology.”
Shaky foundations mean that AI value is hard to determine too, says Cisco. Only 32 percent of firms have a process in place to measure the success or failure of their AI investments, meaning that the much-reported lack of AI ROI could be skewed as well.
Nonetheless, companies are pushing ahead, says Cisco.
Follow the leaders
Cisco’s study splits its cohort of respondents into two pools: Companies that are making responsible decisions when integrating AI into their operations, and everyone else. That former category only makes up between 10 and 15 percent of companies, and those are the ones that Cisco says other firms should look to when thinking about how their own AI initiatives should look from the ground up.
Unlike most respondents, which still lag in IT infrastructure, data preparedness, and governance, those are all strong points for what Cisco calls its “Pacesetter” companies. Among these firms setting the pace for AI adoption, 74 percent report high or full readiness in IT infrastructure, 93 percent in data management, and 84 percent in governance – well above the averages for everyone else.
Identifying use cases and figuring out how to measure the effectiveness of AI investments is also a consideration for most pace-setting companies. The result, Cisco said, is far more confidence in the effectiveness of AI for those companies.
But who are the pacesetters? If they’re giant companies able to eat losses on bad AI bets, then this isn’t so much a measure of ambition or planning as it is a measure of resources. Cisco told us that’s not the case – pacesetters are found in a cross section of companies.
“Pacesetters make up 10–15 percent of companies across every size bracket, and even the largest firms have as many laggards as leaders,” Cisco told us. “What really seems to distinguish Pacesetters is discipline and execution: they plan, fund, and measure AI systematically, and get more consistent results.”
In other words, planning AI initiatives is just like planning any other major business transformation initiative: Build those castles on sand and they’re not going to stay upright.
“As we move into an always-on, agentic era, the companies we surveyed expect the strain on networks, compute, and security to rise,” Cisco said in an email conversation. “The evidence suggests that readiness – built on a secure, scalable foundation – might play a big role in helping organizations keep pace.”
Now, if only the AI companies themselves could build systems that don’t suck, we’ll be on our way to a worthwhile new era of business. ®
