There is a pattern in founder-led businesses that I see regularly. The founder delegates “getting AI into the business” to a senior operator. The delegate has the mandate and the skills. A few months in, the programme is moving, but not as fast as it should. The founder is involved in decisions that are clearly within the delegate’s brief. Each involvement has a reason. Taken together, the reasons add up to something else: a founder who built everything by doing hard things personally, now finding it hard to stop.
The instinct that served them so well is now in the way.
What is the innovation paradox for founders?
The innovation paradox describes a specific trap. The founder who built the business by doing hard things personally develops a deep instinct that the next hard thing also needs their personal touch. That instinct was well-calibrated when the business was young. The same instinct applied to an AI rollout a decade later looks like oversight but feels, inside, like responsibility.
The research on founder behaviour puts a name to what is happening underneath: identity investment. When you have built something meaningful by being the person who figured it out, your sense of what you are for becomes entangled with the act of doing the hard thing. Watkins, writing on leadership transitions, identifies identity investment alongside feedback addiction and quality concerns as the three strongest psychological barriers founders face when trying to let go.
This is not a character flaw. It was a competitive advantage. The same tendency drove you to stay close to early decisions when staying close was precisely what was needed. The trouble is that the business has changed and the instinct has not. When AI arrives as the next hard new thing, the instinct fires again.
Why does handing off AI feel different from earlier delegations?
Many founders have successfully delegated sales, finance, and operations without losing sleep. AI sits differently because it touches how the business reasons about decisions and prioritises effort. That proximity to the founder’s own judgment is what makes it feel special. The feeling is more diagnostic than factual, though. Earlier delegations felt safe because you stayed close to the outcomes. This one can too.
A common framing is that AI is a technology decision, so it belongs with the technical people. That categorisation carries a cost. Once you file AI under IT, you implicitly signal that it is peripheral to strategy rather than central to it. BCG’s 2025 research found that roughly half of companies are stuck at stagnating or emerging stages of AI adoption, unable to scale past proof of concept. The bottleneck is almost never the model quality. It is the people and leadership work being underestimated.
Earlier technology decisions, a new CRM, a data warehouse, could be handed to a technical owner without the founder needing to stay close. AI sits differently because it touches how the business thinks. When an AI tool starts making prioritisation calls, drafting client communications, or surfacing the information that shapes decisions, the founder’s presence in that work is not just about quality control. It is about who shapes the business’s emerging intelligence. That is close enough to how founders see their own value that the pull to stay involved is, frankly, predictable.
Where will you actually meet this pattern in yourself?
It rarely announces itself as resistance. It turns up as opinions. A vendor decision your delegate has already researched. A use-case prioritisation you feel needs one more round of input. A concern about the quality of the output that sits just the right side of legitimate. The tell is not any single intervention. It is that the interventions keep coming, and the programme keeps waiting.
The dynamic has a name in leadership research: verbal delegation with ongoing interference. The founder hands off ownership in principle but keeps intervening on key decisions, which teaches the organisation that AI is really the founder’s call after all. The delegate, sensing this, stops making decisions confidently and waits. The organisation, watching the pattern, stops treating the mandate seriously.
What makes this hard to see from the inside is that each intervention has a plausible justification. The vendor decision matters because the wrong choice will be hard to reverse. The prompt quality concern is real because the outputs go to clients. The prioritisation adjustment seems reasonable because the founder holds context the delegate lacks. These are not lies. The problem is that they are applied selectively, to the one domain where the founder finds it hardest to step back, and not because the risks are actually higher than elsewhere. It is because the identity cost of stepping back is.
One practical signal: compare how you handle oversight here versus in other functions. If you are reviewing AI decisions at a level of detail you would not apply to a similar-value decision in finance or sales, that asymmetry is worth examining.
When is the resistance protecting the outcome, and when is it protecting you?
The distinction matters because not all founder concern is the paradox at work. A misaligned use case, a commercial constraint your board would flag. These are legitimate inputs. The question to ask is whether your involvement changes the decision or only delays it. When the outcome would be the same whether or not you weigh in, the concern is about centrality, not outcome.
A useful exercise is to run the decision forward. If you removed yourself from this call, what happens? If the delegate makes the same decision two weeks later, your involvement cost the programme two weeks. If they make a genuinely different and worse decision, your involvement was protective. Many interventions on the AI side, in practice, fall into the first category.
There is also a second-order effect worth naming. M&A advisers are consistent on this point: founder dependency is among the largest discounts to exit value. PCE Companies estimate buyer discounts of 30 to 40 per cent when operations and decisions remain founder-centric. AI implementation is one of the cleaner opportunities to move the business from founder-led to founder-inspired, because it forces the codification of process. But the same implementation, if it mirrors the founder’s instincts without documenting the underlying reasoning, makes the business more dependent on the founder, not less. The founder who grips the AI mandate is not just slowing the programme. They may be compounding the very problem they are trying to solve.
Naming this is not a verdict. It is an invitation to ask a harder question: is this the outcome I am protecting, or the role?
What shifts when you approach the handoff as a design problem?
Founders who work through this shift where their involvement sits, not whether they stay involved. Designing decision-rights, reviewing outcomes, staying close enough to sponsor well. The day-to-day shape of the programme belongs with the delegate. When you treat the handoff as design work rather than loss, the delegate has room to build and the programme has a clear owner.
The thing the founder does that the delegate cannot replicate is the top-down signal. Research on AI adoption consistently shows that visible, sustained executive sponsorship is among the strongest predictors of whether a programme scales past proof of concept. Nobody in the organisation will take the AI work seriously if the founder does not. The board-facing narrative, how AI sits in the strategy, is a founder conversation.
What the delegate does that the founder should not reach into: the day-to-day decisions about tooling, use-case sequencing, training, and iteration. These belong with the person who has the mandate and the bandwidth to run them. When the founder steps into this layer, they take back a piece of the mandate without intending to, and the delegate’s authority becomes ambiguous.
The practical move is to write down what founder involvement looks like. What decisions need a sign-off? At what cadence is progress reviewed? What would trigger more active involvement? Answering these converts an ambiguous handoff into a clear one and gives the delegate what they need most to succeed: confidence that when they make a call, it holds.
If you have found yourself explaining why the AI work still needs your personal attention, that explanation is worth sitting with. The instinct behind it built something real. The question is whether it is building the next thing, or holding it back.
