A pattern I keep running into goes something like this. A company decides it is going to get going on AI. Leadership is excited, somebody buys licenses or turns on an agent, and the message to the team is basically "let's give AI access to everything and figure it out as we go." Within a few weeks one of two things happens. Either people barely touch it because nobody is sure what they are allowed to do, or a few people go all in and IT finds out after the fact and pulls the plug. Both groups end up frustrated, and leadership is left wondering why the rollout stalled.
My take is pretty simple. People get braver with AI when the boundaries are clear, and drawing those boundaries is leadership's job. You can't skip it by buying the right software, and you can't hand it entirely to IT and walk away.
Freedom comes from having lines to play within
I like to use basketball for this. Basketball would be chaos without the lines on the court. Nobody would know where to stand, what counts, or when they have gone too far. The lines are what let players go full speed, because they know exactly where the edge is.
The organizations I see that are the most free with AI are the ones that have safe walls to play within. Their people know which tools are approved, what kind of data can go into them, and what still needs a human sign-off. So they try things. In companies without those lines, the careful people freeze and the bold people improvise, and honestly neither one is what you wanted.
"Give it everything" doesn't move you any faster
The give-AI-everything approach feels fast because it skips the slow conversations. In practice it pushes them to later, when they cost more. Once an agent can read every folder, somebody eventually notices it surfaced something it shouldn't have, such as payroll, a client contract, or an HR file, and now the conversation about access happens in the middle of a cleanup instead of at the start.
It also tends to make the whole program feel unsafe to the people you need on board. When folks hear about one bad incident, a lot of them decide AI is something to avoid. Undoing that takes far longer than drawing the lines up front would have.
Stop treating IT and security like the bad guys
Every company I work with has a kind of angel and devil on the shoulder. One side is pushing to move fast and try everything. The other side, usually IT and security, keeps asking what happens to the data and who is accountable if something goes wrong. I think you need both, and you shouldn't be shaming either one.
I have heard leaders roll their eyes at IT for slowing things down. That is their job. If a security breach happens, who are you going to blame? The better move is to get the people who want speed and the people who want safety in the same room early, and let them agree on what is in bounds. When they have built the lines together, IT stops being the department of no, and the fast movers stop feeling like they are sneaking around.
The lines come out of conversations, not a tool
There is a lot of software out there that promises to handle AI governance for you, and some of it is useful for enforcing rules once you have them. Deciding the rules is different. Who should be able to see what, which decisions an AI can make on its own, and where a person has to approve before anything happens all take real human conversations. A product can't have them for you.
The approval question is a good example. In a TeamViewer survey of 4,200 respondents run by Sapio Research earlier this year, 46% of manufacturing respondents said they want a human involved when an automated action could affect security, data, or business operations, and only 23% wanted a human involved before any action is taken. Those numbers tell me most teams don't want to approve every little thing. They want someone to decide which things matter enough to need a person, and that is a decision your leadership team should make on purpose.
Be honest that you're figuring it out too
Leaders sometimes avoid drawing lines because they are worried they will draw them in the wrong place. I get that. This is new to all of us, and anyone who claims to have the exact formula for every company is overselling. What breaks trust is pretending you have it all figured out and prescribing the right answer from the top.
It works better to say plainly: here are the lines for now, here is why we drew them, and we will revisit them in a couple of months once we see how people actually use this. Your team will respect that a lot more than a perfect-sounding policy nobody believes. It also gives people permission to flag when a line is in the wrong place, which is how you find out where to move it.
What to do this week
If you are rolling out AI, or you already have and it feels stuck, try this:
- Write down, in plain language, which AI tools are approved and what kinds of data can go into them. One page is plenty.
- Pick the three or four actions where a person has to approve before AI does anything, such as sending to a customer, changing a record, or touching money. Let everything else run.
- Get your loudest AI fan and your IT or security lead in the same meeting and have them agree on that list together.
- Check who can see what before you connect any AI to your shared drives. If a summer intern can open the finance folder, so can the AI working on their behalf.
- Tell your team these are the lines for now, and put a date on the calendar to revisit them.
None of this needs a big budget, and most of it fits in a couple of meetings. It does need leadership to own it, because nobody else in the company has the authority to decide where the lines go.
If you would like help drawing those lines with your leadership team and then getting people comfortable working inside them, that is a big part of what we do at Spej. Talk with the Spej team.
