Employees shouldn't use AI just because it's available. But they shouldn't be afraid to use it because of what it might mean for their jobs, either.
That's an awkward challenge for CIOs: helping people become comfortable with AI while teaching them when to set it aside. And the business faces a similar decision. Just because AI can do something, does that mean it's the best way to get it done?
Two numbers from last week's reporting are worth considering. Among many companies working with GAI Insights, an analyst firm, just 5% of employees account for 95% of AI usage. Meanwhile, a Gartner survey found that only 13% of CIOs said their enterprises had realized significant or transformational value from AI.
Those are different measures, of course. But for all the enthusiasm surrounding AI, they illustrate two problems enterprises are still working through: getting people to use it and getting enough value from what they're doing.
More AI isn't always better AI
Consider Zapier. Two employees outside engineering attracted management's attention because of unusually high AI spending. One project was shut down because the employee had chosen the wrong model and the cost wasn't justified. The other got the go-ahead because the work was worth the money.
What matters here isn't that both employees ran up big bills. It's that Zapier took the time to decide whether those bills were worth paying.
The company also expects new hires to meet an AI-fluency standard. Its chief revenue officer, Dan Slagen, put it rather bluntly, describing the hiring process as a way to "stop letting enemies through the gate." Meanwhile, at Freshworks, some employees worry that revealing how much time AI saves them could put their jobs at risk.
That's a difficult message to reconcile: Use AI, but don't make yourself replaceable.
There's a similar question at the enterprise level. Rebecca Wettemann, principal at Valoir, a tech research and consulting firm, asked whether an LLM call could have been handled by an ordinary workflow. At Ochsner Health, vice president of IS operations Megan McFarland advised examining processes before introducing AI. Maybe simpler automation would do the job. Maybe the process itself needs fixing.
That shouldn't be a revelation. IT has always had to understand the work before introducing a new tool. AI is software, after all. But agents that interpret instructions and take action across systems are likely to need more explicit controls than traditional automation.
Different approaches, same question: What does the technology need to do, and where should its authority stop?
Three examples from last week's reporting on AI agents in production show how differently enterprises are approaching the problem.
- Meta. An AI agent started applying rules from one subsystem to another. The company responded by dividing the work among 14 specialized agents, with a deterministic orchestrator keeping things on track.
- Lendi Group. This company took a different approach, keeping business rules outside its models and testing agent behavior in production, with people reviewing exceptions.
- Trust Bank Singapore. Letting AI troubleshoot without the structure of prescribed runbooks produced inconsistent results. The bank now uses those runbooks to gather evidence, AI to analyze it and engineers to decide what to do.
Different approaches, same question: What does the technology need to do, and where should its authority stop?
There's a related lesson in David Linthicum's And Then There’s Reality column about IT skills shortages. He describes a financial services firm that couldn't seem to hire qualified cloud engineers. But the real obstacle was a tangle of undocumented dependencies on a decades-old system. Once the company documented and simplified those dependencies, the hiring problem largely disappeared.
Training people to use AI won't fix a broken process or make a bad deployment worthwhile.
Dave wasn't writing about AI, but the connection is worth making. Educating employees is important. So is making sure the underlying processes and systems aren't unnecessarily complicated. Training people to use AI won't fix a broken process or make a bad deployment worthwhile.
None of this means CIOs need to invent an entirely new way to manage technology. Understanding the process, weighing the benefits, establishing controls and learning from experience are familiar IT practices. The challenge is extending them to people throughout the enterprise who can now put AI to work without an IT project team beside them.
The goal isn't to get everyone using AI all the time. It's to help people recognize when AI makes sense, when it doesn't and what needs to change before it will.
James Alan Miller is a veteran technology editor and writer and lead editor for CIO news at TechTarget. He directs coverage of enterprise technology strategy, AI, software, data, infrastructure and the decisions shaping how CIOs manage increasingly complex IT environments.
