You appointed someone to run AI six months ago. You gave them the mandate, the budget, a seat at the leadership table. You meant every word of it.
Then a vendor shortlist came back and you had a view. Then a decision about which workflows to pilot landed in your inbox and you sat in on the call. Then another one. And somewhere in there, without any single moment you could point to, the programme stopped being theirs and became yours again.
If you’re wondering why the AI programme feels stuck, why your delegate pauses before moving and asks before acting, and why the same decisions keep routing back to you, this is for that moment.
What is the interference trap?
The interference trap is a pattern in delegation psychology where a leader hands over ownership verbally, then keeps intervening on the decisions that matter. The delegate holds the title but not the authority. The rest of the organisation reads this quickly. Within weeks, they have learned that the person with the mandate does not have the real power, and they route decisions accordingly.
Leadership researchers describe this dynamic as reverse delegation, where decision-making responsibility flows back up the chain despite the stated intention to push it down. It has been documented in corporate settings, academic institutions, and founder-led businesses. The pattern is rarely deliberate. The founder does not sit down and decide to take back control. They ask a clarifying question. They join one call. They offer a view that the delegate can see they were expected to accept.
Each small intervention is defensible on its own. The accumulation is what does the damage. By the time the founder notices anything is wrong, the organisation has already adapted. The team has stopped treating the delegate’s decisions as final. The delegate has stopped making them.
Why does it matter for your AI programme?
A stranded delegate, one with responsibility but no authority, will stop making calls and start waiting for direction. The AI programme stalls in a holding pattern where nothing is technically wrong but nothing moves. Decades of change management research confirm that technology implementations commonly fail when the leadership and people work is underestimated, not when the technology itself lets them down.
BCG research finds roughly half of businesses cannot scale AI past proof of concept, stuck in stagnating or emerging stages. A delegate who has lost the authority to move independently is a significant contributor to that stall. The programme has a person in charge, but that person has no room to act.
The exit dimension adds a separate cost. M&A advisors note that owner dependency is among the largest discounts to exit multiples, with buyers applying discounts of 30 to 40 per cent when operations and decisions centre on the founder rather than the business itself. An AI programme that has become founder-dependent in practice, because the founder keeps pulling the real calls back to themselves, compounds that discount rather than reducing it.
There is also a people cost. A capable operator placed in a role with responsibility but no authority tends to make one of two moves. They become cautious and performative, asking permission for everything. Or they leave. The founder who created the interference trap may not realise they have done it until the delegate is gone.
Where will you actually see it?
The clearest signal is where decisions travel. If the AI programme is running with proper delegation, decisions about tools, workflows, and vendor calls land with the delegate. If they are routing back to you, or if the delegate includes you in loops you did not ask to be in, the organisation has already learned that the real authority sits elsewhere.
A second tell is the delegate’s communication style. A delegate with real authority talks about what they have decided. A stranded delegate talks about what they are considering. They hedge, over-explain, and reference your likely view before committing to anything. That caution is not indecisiveness. They have learned the cost of moving without your nod.
A third tell is team behaviour. If the people working on AI are going around the delegate to get your read directly, the reporting line has shifted in practice. The delegate may not have told you this is happening. Telling you would itself be awkward.
The fourth, and most telling, tell is your own calendar. Count how many AI-related meetings you are in that were not your idea. Count how many decisions you have been asked to weigh in on in the last month. If that number is higher than what the mandate described, the drift has already happened.
When should you step back in, and when should you hold off?
Some decisions genuinely belong with the founder. Vendor contracts at meaningful spend, choices with board implications, and anything that publicly commits the business all warrant direct involvement. The interference trap shows up everywhere else, on the day-to-day calls the delegate was appointed to handle, precisely because the founder finds it hard to hold back.
The psychology is well-documented. Founders who have built success on their own judgement find it hard to trust a different person’s call, especially on something as visible as AI. Research on leadership delegation barriers identifies identity investment, feedback addiction, and quality concerns as the mechanisms that make letting go difficult. The founder’s hesitation rarely comes from doubting the delegate’s ability. The pull is towards their own judgement, the one that built the business in the first place.
The practical test is straightforward. Before you join an AI meeting that was not your idea, or before you send a note about a decision that has not reached you, ask yourself one question. Is my involvement going to improve the outcome, or am I here because I find it uncomfortable not to be?
If the honest answer is the second one, stay out. The delegate cannot grow into the mandate if you fill the space.
How do you reset a delegation once it’s slipped?
The reset has to come from the founder. Expecting the delegate to name the pattern upward is asking them to do something awkward and professionally risky. A founder who has noticed they are overreaching needs to call that out directly, restate what they meant to hand over, and close the gap between the role they described and the authority actually granted.
The conversation has a specific shape. Name what has been happening without making it the delegate’s fault. State explicitly what decisions sit with them going forward. Agree the short list of situations where they should loop you in. Write it down, even briefly, because a verbal reset carries the same drift risk as the original verbal mandate.
The decision-rights conversation should also cover what support looks like. The delegate probably needs access to information you hold, or a way to get a quick steer on something genuinely uncertain, without the steer becoming a veto. A standing fifteen-minute check-in where they report what they have decided, rather than asking what to decide, is often enough to give you visibility without reversing the delegation.
Two things will tell you the reset has worked. The delegate starts moving again without asking. And the decisions that route back to you become the ones that genuinely should. That gap, between the decisions you were getting and the decisions you were supposed to be getting, is where the AI programme lost its momentum.
If you are at the point where you need this conversation and are not sure how to structure it, Book a conversation and we can work out where the line between involvement and interference sits for your specific setup.
