You’ve done the handoff. Named the person, had the conversation, said the words. They’re running the AI programme. That part is over in under an hour. What follows runs for the next twelve months, and whether it goes well depends almost entirely on something that conversation rarely covers: which calls are yours and which are theirs.
A lot of founders hand over the mandate without writing down the boundaries. The intention is good: get out of the way, let them run it. But without a written record of who decides what, every borderline call becomes a negotiation. The organisation watches those negotiations and draws conclusions about where the authority sits. By the time you realise the programme has stalled, the pattern is embedded.
What are AI decision boundaries?
Before the work starts, you need to divide your decisions into three groups. Some calls stay with you regardless of how the programme develops. Others belong to your delegate, no check-in required. A third set are the agreed reconvene points where you decide together. That division, written down before anyone starts, is what structured autonomy looks like in practice. Verbal delegation without it tends to collapse under pressure when the first difficult call arrives.
The boundary document doesn’t need to be long or formal. A single page is sufficient for the typical owner-managed business. What matters is that it exists before the programme starts, that your delegate has seen it, and that you’ve both agreed to treat it as the operating frame rather than a starting suggestion.
The reason it needs to be written is precise. Verbal agreements about decision-making erode faster than anyone expects. A call that both parties understood as the delegate’s can feel, a few months later, like one the founder should have been across. Written boundaries don’t prevent that feeling. They give you both a place to go when it surfaces.
Why does building this fall on you, not your delegate?
Your delegate can push for clarity on who decides what, and the good ones will. But the authority to draw those lines sits with you. When a founder sets the frame before the work begins, the organisation reads it as confident leadership. When a delegate asks for the same clarity after the programme has already stalled, the organisation reads it as the delegate asking for protection. The sequence matters as much as the substance.
A number of executive advisers have pushed in recent years for founders and CEOs to stay personally engaged rather than delegating outright [1]. The argument has merit. But staying engaged without clarity on who decides what often produces the worst of both outcomes, a hands-on founder who keeps reaching over their own delegate and a delegate who has stopped running the programme with any real confidence.
The research on failed AI implementation points consistently to leadership decisions, not technical ones [3]. About half of companies find themselves stuck in early adoption stages, unable to move from pilot to scale [2], and Kyndryl found in 2024 that only 14% of organisations had their people, technology and growth goals properly aligned [5]. The founder’s decision about who decides is itself a leadership act, and it needs to happen before the work begins.
Which calls are actually yours to keep?
Budget thresholds are the obvious starting point, and they matter. But the decisions that genuinely warrant founder involvement run wider than a spend cap. Anything that shapes how the business reads to a potential acquirer belongs with you. So does anything carrying data or reputational risk that could surface during due diligence. The delegate owns the how; you hold the what.
If you’re running the programme with any eye on exit, the framing becomes more pointed. M&A advisers consistently flag owner dependency as the largest single discount to exit multiples, with buyer adjustments of 30-40% common in businesses where the founder is the operational centre [8][7]. An AI programme that reduces that dependency is a genuine valuation driver. One that builds AI around the founder’s instincts without documenting the underlying process can increase it.
That means the founder input that matters most is the periodic check that the programme is actually moving the business towards a structure that can run without you. That’s a strategic question, and a quarterly one rather than a weekly one.
Things your delegate should own without coming back to you include tool selection within budget, vendor shortlisting, workflow design, team training sequencing, rollout pacing, and the choice of which functions to prioritise first. These are judgement calls that sit inside their expertise. Reaching back over them after the fact, because the choices turned out differently than you’d imagined, is where programmes break down.
When do you step back in, and when do you stay out?
Structured autonomy means you agree in advance where you reconvene, rather than reacting to whatever feels significant in the moment. Three to five checkpoint moments over the programme’s first year, where your delegate brings you a briefing rather than a permission request, is usually enough. Between those points, the work is theirs to run. Pulling decisions back that weren’t pre-agreed as yours teaches the organisation that the boundaries are approximate.
The reconvene points you pre-agree should be strategic in character, not operational. Board updates on AI progress, a quarterly review of whether the programme is reducing or adding to founder dependency, a check before any external communication about AI capabilities. These are the moments where your view genuinely shapes something consequential.
The discomfort of staying out when a call feels significant is real, and the research on perceived control documents it carefully [6]. The boundary document does some of its best work here, giving you a pre-agreed frame for deciding whether stepping in is warranted or reactive. Without that frame, the pull to intervene on a call that feels important is hard to resist, however good your intentions about delegation.
Reactivity is the failure mode. When founders step in because a call feels wrong rather than because a pre-agreed checkpoint has triggered, the delegate starts pre-checking everything. The programme slows to the founder’s bandwidth, which defeats the purpose of delegating in the first place [9].
What does the written boundary document actually contain?
A single page that answers four questions is sufficient for the typical owner-managed business. Which decisions require your sign-off? Which does your delegate own outright? Which go to an agreed reconvene point? And what is the trigger for escalating a call that was supposed to be the delegate’s? That last question is the one many founders leave out, and without it, uncertain calls default back to you anyway.
The escalation trigger doesn’t need to be elaborate. A practical rule might be that any decision affecting two or more teams, any change to a vendor relationship above a certain threshold, or any output the delegate is uncertain about presenting to the board gets flagged for a one-line steer from you. The point is that the escalation path is agreed and named before it’s needed, not improvised in the moment.
The document also needs a review date. AI programmes move quickly, and what was clearly a delegate’s call at month three may be a founder call at month nine as the scope expands. Build a six-month review into the document so both parties know when the frame is up for renegotiation.
Delegation frameworks in the leadership literature are consistent on this point: structured autonomy outperforms both close control and full abdication when the work is complex, the stakes are high, and the delegate has genuine expertise to contribute [10]. The founder’s job is to build the frame and then trust the person inside it.
The boundary document is a one-time piece of work. A few hours, at most, to get the framework agreed and written before the programme starts. That’s a small investment for what it returns: a delegate who can run the work with confidence, an organisation that knows where the authority sits, and a founder who stays involved at the level where they’re genuinely useful. Write it before the work starts, not after the first difficult call makes the absence obvious.
If you’re working through this as part of a wider AI implementation and want a second view on how to structure the handoff, Book a conversation is the place to start.
