6 Guidelines for Governing AI
IEEE Spectrum Sravan Vadigepalli ● Covered by 2 sources
AI work is shifting from building tools to setting the rules they follow. The surprise: most companies have the models, but not the people to govern them.
Based on reporting by IEEE Spectrum, Sravan Vadigepalli — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
For years, a lot of AI work was about making the thing run. Now the harder job is deciding what it may do, when it has to ask for help, and what it must never touch. That’s the argument from an IEEE Spectrum piece by a Lowe’s AI leader, who says the profession is moving from execution to governance.
The backdrop is ugly enough to make the point for him. A 2025 MIT Media Lab Project NANDA report found that, even with an estimated US$30 billion to US$40 billion poured into enterprise generative AI, most organizations in its dataset had not shown measurable profit-and-loss impact. Only about 5 percent of integrated pilots were producing substantial value. The models are there. The returns mostly aren’t.
His answer is a set of six operating habits for people who now have to supervise machines instead of just build them. The first is to notice when you’ve become “human middleware,” stuck relaying data between systems instead of making decisions. The second is to replace brittle rules with principles, written in priority order so a system can resolve conflicts without waiting for a human every time a weird case appears.
He also argues that company values have to be written into code, not left as posters on a wall. That means a top layer of hard limits, a middle layer describing trade-offs, and a lower playbook for specific tasks. On trust, he prefers a “thermostat” to a switch: let low-risk decisions run on their own, route uncertain ones to people, and feed those human decisions back into the system. He wants the whole thing to stay a glass box, with decisions transparent and auditable.
The last two pieces are context and exception handling. If the system can’t see the budget, the contract clause, and the customer complaint in one place, it can’t be governed well. And if humans are still checking every report, nothing really changed. The point is to let people rise above routine work and spend their time on the odd, high-stakes cases. That is where judgment lives now.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI everyone wants to skip because governance sounds less glamorous than prompts and demos. But without decision rights, thresholds, and actual accountability, enterprise AI is just expensive improvisation with better branding. The industry keeps acting surprised that models don’t magically behave like managers.
Read more about this at: IEEE Spectrum