Part 2: Why This Gets Worse With Agentic AI (And What to Do About It)

This is Part 2 of a two-part article. Read Part 1 here: https://unbrokenwork.com/part-1-nobody-decided-who-decides-the-real-reason-ai-adoption-stalls

Why This Gets Worse With Agentic AI (And What to Do About It)

Last time, we talked about why shadow AI and stalled approvals are actually two sides of the same problematic coin: nobody has clearly decided who gets to decide on a given AI use case, so the org either freezes or sprawls. That gap is uncomfortable now, but it's about to get a lot less comfortable, because the next wave of AI doesn't wait around for someone to notice the ambiguity.

Agents don't pause to ask

A person who's unsure whether they're allowed to do something will usually hesitate, ask a colleague, or wait for a meeting; an AI agent given a fuzzy mandate just acts on its best guess, immediately, and moves on to the next task. There's no built-in pause for "wait, should I actually be doing this," because that pause depends on a human noticing the ambiguity in the first place, and agents aren't wired to notice it the way people are.

That means a decision rights gap that was merely annoying when humans were the ones filling it in is suddenly a lot riskier once agents are the ones filling it in, simply because of speed. A person might sit on an unclear judgment call for a day. An agent might make a hundred of them before lunch.

What this looks like in practice

Picture an agent tasked with handling customer follow-ups, and nobody's specified whether it's allowed to offer a discount on its own or needs a human sign-off first. A person in that seat would probably ask their manager before doing anything unusual. The agent doesn't ask; it just makes a call, based on whatever pattern it picked up, and that call is now live in the world, no review, no pause. Multiply that across every workflow you've handed to an agent, and you can see how fast a small ambiguity turns into a real mess.

The fix isn't more policy

The instinct here is usually to write a longer governance document, add more principles, hold another training session. None of that touches the actual gap, because the problem was never that people didn't understand the values. The problem is that nobody made the decision rights visible: who decides, what they're deciding on, how fast they need to decide, and what happens by default when it's unclear.

A simple next step

This doesn't need to be a six-month initiative with a steering committee. It can start as a focused exercise: take your highest-risk or highest-volume AI use cases, and for each one, write down who can say yes and who can say no, not as a department or a function, but as an actual person currently sitting in that role, someone who could pick up the phone and answer the question today. If you can't fill in that name yet, that's fine, that's exactly the gap you're trying to find; write down who should hold it instead, and treat that as your action item. Then add one more line for each use case: what happens if that person isn't available when a decision needs to get made fast. That's it. You're not redesigning your whole org chart, you're just making the invisible visible, one use case at a time, starting with the ones where ambiguity would actually hurt.

Once that's on paper, the bottlenecks tend to loosen and the shadow AI tends to surface, because people finally know whose name to call instead of guessing, or worse, not bothering to ask at all.

The bigger point

AI adoption isn't stalling or sprawling because people don't get it, and it's not a motivation problem or a skills gap either. It's stalling because nobody decided who decides, and until that gets fixed, no amount of new tooling, new training, or new policy language is going to close the gap. Decide who decides first; everything else gets easier after that.

Contact

Reach out for consulting or speaking inquiries

Email

hello -@- unbrokenwork -dot- com

© 2026. All rights reserved.