Gartner says AI and its vendors aren't enterprise-ready
At Gartner's annual IT Symposium, held first in Australia ahead of later editions in Europe and the United States, the analyst firm delivered a blunt verdict on enterprise AI: the more of it an organization uses, the harder it becomes to control or to get a return on investment. Distinguished VP analysts Daryl Plummer and Kristin Moyer told the Australian audience that AI and its leading proponents remain immature.
Asked to comment on working with the leading AI labs, Plummer said trust in them is not warranted yet: they do not understand enterprise terms and conditions, enterprise liability, or the need for consistency and continuity. He blamed vendors' habit of changing models with little regard for the applications that depend on them. A typical AI model lasts about six months before it is superseded, he said, and a vendor pays little attention to a customer who stays on an older version, which alone shows the vendor is not enterprise ready. At the same time, he acknowledged that the pace of change is now so fast enterprises cannot afford to ignore it.
Moyer cited Gartner research finding that 86 percent of CIOs see AI-created risk growing faster than the value AI creates, in part because early wins with the technology generate more internal demand than IT departments can safely satisfy. Some of that demand is what she called careless consumption: employees using AI when it is unnecessary or inappropriate. She pointed to research showing that 40 percent of workers have run into AI slop, output so poor that untangling it takes about two hours each time, adding up to roughly $9 million a year in wasted staff time at a hypothetical 1,000-person organization. Curbing that is hard, she said, because AI now ships quietly inside software companies already own, making it difficult to even count how many AI agents are in use. She compared the challenge to the decade Gartner watched enterprises spend reining in developers' preference for the most powerful, most expensive cloud computing instances available.
Plummer added that some careless use is invited by vendors themselves, who he said are desperate to monetize AI by bundling it into products or simply selling more tokens, and who want customers to use agents everywhere. As an example of overreach, he said building an AI agent just to query a database is unnecessary when an ordinary function call already does the job. He was equally skeptical of AIOps tools, which use agents to diagnose problems and suggest one-click fixes: such offerings are dishonest, he said, because vendors are more interested in selling their own product than in recommending whatever actually solves the problem, and the market for AI governance tooling is, he added, one of the most fragmented he has ever seen.
The two analysts recommended three concrete steps. First, an AI central bank to oversee AI's systemic impact across an organization, with accountability as its central job: something Plummer and Moyer said is straightforward with conventional ERP or CRM systems, which leave an audit trail of who did what, but far harder with AI, which does not always leave that kind of evidence. It is not easy to know who goes to jail, Moyer quipped, urging companies that lack a way to log AI's actions to build one. Second, guardian agents: AI whose only job is to watch other AI agents and keep their behavior within bounds, with the power to shut down ones that go rogue. Third, dedicated AI disaster recovery teams tasked with cleaning up the messes AI causes. Plummer closed with a warning that blaming the technology will not satisfy anyone: if an organization is called to account and its answer is that AI did it, he said, it is in trouble, because CIOs will be held accountable for AI failures.
Key facts
- At Gartner's IT Symposium in Australia, distinguished VP analysts Daryl Plummer and Kristin Moyer said AI and its leading vendors remain immature and are not enterprise-ready.
- Gartner research cited by Moyer found that 86 percent of CIOs see AI-created risk growing faster than the value AI creates; separate research she cited found that 40 percent of workers have run into AI slop that takes about two hours to untangle, an estimated $9 million a year in wasted time at a 1,000-person organization.
- Plummer said AI models last about six months before vendors move on, so a customer who stays on an older version gets ignored, which alone shows a vendor is not enterprise ready; he also called AIOps one-click fix tools dishonest, since vendors are more interested in selling than in the right solution.
- Gartner's three recommendations: an AI central bank for organization-wide oversight and accountability, guardian agents empowered to shut down rogue AI agents, and dedicated AI disaster recovery teams.
- Moyer said it is hard to even count how many AI agents an organization has, since AI now ships inside products companies already own; Plummer said CIOs will be held accountable for AI failures.
Why it matters
Gartner is one of the most influential enterprise IT analyst firms, and its guidance shapes budgets and purchasing decisions across the industry. A public verdict from two of its distinguished analysts that vendor trust is not yet warranted directly challenges the vendor pitch that faster AI adoption is always the right move. It gives CIOs an analyst-backed argument to slow down, demand accountability, and budget for governance instead of simply buying more AI capacity.
Who it affects
CIOs and enterprise IT leaders deciding how fast to adopt AI and AI agents; procurement and governance teams, who now have a named framework to push back with (an AI central bank, guardian agents, a dedicated disaster recovery function); the AI vendors and labs whose update cadence and monetization tactics are the direct target of the criticism; and, more indirectly, the employees whose careless use of AI is estimated to cost a mid-sized organization millions of dollars a year in cleanup time.
How to use it
Gartner's own prescription has three parts. Create what the firm calls an AI central bank: one function with ownership of AI's systemic, organization-wide impact, instead of leaving each team to adopt AI independently. Build or acquire guardian agents whose sole task is watching other AI agents, empowered to shut down any that misbehave. Stand up a dedicated AI disaster recovery function rather than assuming existing incident response already covers AI-caused failures. Plummer's supporting advice: skip the agent where a plain function call already solves the problem, and treat a one-click AIOps fix as a sales pitch rather than a verdict.
How solid is it
The claims come from two named Gartner distinguished VP analysts, quoted on the record at a Gartner-run symposium and reported directly by The Register's APAC editor, a solid attribution chain for an opinion and a set of recommendations. The two headline statistics, that 86 percent of CIOs see AI risk outpacing AI value and that 40 percent of workers have run into unusable AI output, are credited to Gartner research for the CIO figure and to unspecified, unnamed research for the AI slop figure, with no sample size, survey date or methodology given in the article for either. The $9 million figure is a derived estimate, two hours of cleanup time multiplied across a hypothetical 1,000-person organization, not a measured cost from a real company. No specific AI vendor, lab or product is named anywhere in the piece.
Risks and caveats
The critique never names a specific vendor or lab, so it cannot be checked against any single company's actual update policy or contract terms. The AI central bank, guardian agent and disaster recovery team ideas are Gartner's recommended concepts; the article names no product that implements them and does not explain how a guardian agent would technically detect or stop a rogue agent. The underlying survey figures lack methodology, so they are best read as a directional signal of sentiment rather than a precise measurement. This is analyst opinion and recommendation from a single research firm, not empirical research or a regulatory finding.
“Trust in these vendors is not warranted yet. They are not enterprise grade.”
— Daryl Plummer, distinguished VP analyst at Gartner