How to delegate to people when AI makes you faster
I first notice this in my own work.
For the last three years I have used AI as a working partner across most of what I do, which has increased my capacity in a very real way. I can move from idea to draft, problem to solution faster, keep more workstreams moving, and ship more than I could if I was doing every step manually.
At first, it felt like a personal productivity advantage. Over time, I started to see a different trade-off appearing. The more fluent you become with AI, the easier it is for work to stay with you, because doing it yourself can feel quicker than explaining it, waiting for it to come back, and then reviewing something that has drifted.
I have started seeing this with clients too. Someone becomes the AI-fluent person because they have had more training, felt more comfortable experimenting, or simply spent longer using the AI tools in real work. It might be the founder, a senior operator, or a team lead. Usually, they are the person others turn to for the judgement calls: which customer issue matters most, how to word a sensitive email, what to cut from a proposal, or whether a draft is good enough.
That creates a bottleneck in small businesses and SMEs. One person is using it well, while others are still using AI for quick answers or avoiding it altogether. The problem gets worse when there is no obvious person to delegate to. Hiring is not instant, and a small business may not have the budget, network, or time to find, interview, onboard, and train someone properly. Even with a capable team, there may not be someone ready to take the work and use AI well in that area.
So delegation stops being “can someone else do the task?” and becomes “can someone else take this far enough that I am not rewriting it at the end?” Until the answer is yes, the work keeps coming back to the fastest person.
TL;DR
AI can make one person inside a team dramatically faster, but it can also concentrate capability in one place. Delegation starts to feel inefficient when there is no obvious owner for the work, or when the person taking it on does not yet know how to use AI well in that domain. Hiring more people is not always quick or realistic, so the answer is not simply “delegate more”. It is to train people properly, build AI literacy across the team, and make the work easier for others to pick up without sending it straight back to the same person.
The new delegation problem
Delegation has always been difficult in teams and small businesses where important context sits with a few key people. In founder-led businesses, that often means the founder. In a growing SME, it might be a delivery lead, commercial lead, operations lead, or a senior specialist who knows the customer nuance, the positioning, and the internal standards.
When that thinking is not written down or clearly explained, delegated work often comes back needing more work. It might be technically complete, but it is missing the nuance behind the request: what the person meant to say, how they wanted it said, or the small judgement calls that would have made the output feel right.
That was already true before AI. The difference now is that there is another layer in the middle. You are often delegating to someone who will then use AI to do part of the work. That can be useful if they know how to use it well in that specific context. If they use it superficially, the person reviewing the work can end up dealing with two gaps at once: the human has not fully understood the intent, and the AI output has not been shaped with enough judgement.
Instead of saving time, delegation can create a different kind of work. You have to review the output, untangle the thinking, correct what has drifted, and sometimes pick the whole thing back up yourself. At that point, it is very tempting to think you could have done it faster with AI yourself, and sometimes that is true.
Why AI can concentrate work around one person
The more fluent someone becomes with AI, the easier it is to keep hold of work that would previously have been too time-consuming to carry yourself. Tasks that once needed a proper block of time can now move forward in smaller bursts, so they feel less urgent to hand over.
That can quietly become a problem for teams and SMEs. Because AI makes each task feel easier to move forward, more of the work stays with the person who is already quickest. By the end of the week, they are not just doing more. They are carrying the decisions, the half-finished pieces of work, and the context for everything that has not been properly handed over yet.
For some people, especially those with ADHD or a naturally multi-threaded working style, that pace can feel energising for a while. The cost still shows up eventually, usually through draining context switching, decision fatigue which leads to slower follow-through, and the feeling that everything is technically moving while too much still depends on one person.
AI creates a new strange type of bottleneck here. It makes the most AI-fluent person faster, and it also makes that person the easiest place for too much of the work to land.
The issue is the quality of judgement
It would be easy to say people just need to delegate, and of course they do, but that does not quite solve the problem. The hesitation is not irrational (or at least I’ve convinced myself of this). If one person has spent months using AI inside real work, and someone else is still treating it like a quick answer machine, the gap in quality is obvious to anyone that has experience delivering using AI.
The issue is whether others can use AI with enough context and judgement for the work, in order to achieve your desired outcome. If they are drafting a client email, do they know what needs to sound firm, warm, careful, or direct? If they are shaping a proposal, do they know what matters commercially and what can be cut? These are fairly simple examples, but the problem becomes more obvious as the delegated work is more complex.
That is the difference between using AI as a shortcut and using it as part of the work. Shortcut use usually means asking for the output and accepting the first thing that looks plausible. Better use means staying involved: giving the tool the context, questioning what comes back, tightening the direction, and deciding whether the result is actually good enough.
When AI is used superficially inside delegated work, the output can look polished while still missing the point. The reviewer then has to work out whether the problem is the person’s understanding, the AI’s interpretation, or the gap between the output and what was originally intended.
Delegation now has to include AI judgement
In an AI-supported workplace, handing someone a task is rarely enough. Teams and SMEs need clearer direction on the outcome, the context, the quality bar, and how AI should be used in the work.
That does not mean writing a long instruction document every time. It means defining what the person is actually responsible for. Are they producing a first draft, or deciding what the draft needs to achieve? Are they using AI to move quickly, or to think the problem through before they write anything? Are they making the decision, or preparing options for review?
Those distinctions matter because AI can produce a convincing surface layer output very quickly. Without a clear standard to work to, someone can send back something that looks finished but still needs heavy refinement to make it usable. The work has moved forward, but ownership has not really transferred to the other person.
A useful place to start is one piece of work that keeps coming back half-finished. Take a look at where it breaks, was the prompt missing key context? Did the person not know some nuances that mattered? Did they accept the AI output too quickly? Or was the review standard only obvious to the person who delegated it? That gives you something specific to fix, instead of treating delegation as the whole problem.
Building capability beyond one person
This is where AI training needs to get more practical. It cannot only be a walkthrough of the tools or a few prompt tips. People need to learn how to use AI inside the work they are actually responsible for and be confident achieving the same outcomes or greater.
That looks different depending on the work and the AI tool can help with the output, but it cannot take responsibility for your work.
This is the part that often gets missed. If someone is going to take work off someone else, access to the same AI tool is not enough.
Some work will still need to stay with the founder, senior operator, or subject-matter expert. Strategic decisions, sensitive client situations, positioning, and commercial calls often rely on experience that is hard to hand over quickly. Keeping that close can make sense. The problem starts when everything stays close because AI makes each individual task feel manageable.
The question I keep coming back to is whether a piece of work should still depend on the same person moving forward. If it should, fine. If it should not, then the work needs to become easier for someone else to pick up. That might mean tightening the prompt, building an agent around the workflow, or working through the first few attempts together so the weak spots become obvious. It might also mean accepting that if you cannot hire the perfect person quickly, the people already close to the work need training and practice.
That upfront effort can feel frustrating when the fastest person could just do the work directly. But without it, the same pattern keeps repeating. The business keeps producing more, but the work still depends on the person who already has the most context.
This is also where systems matter. If the way the work should be done only lives in one person’s head, delegation will always feel harder than doing it yourself. AI can even make that gap more obvious, because the strongest AI user can combine the tool with everything they already know, while everyone else is working from a thinner version of the task.
The businesses that get real value from AI usually make the hidden parts of the work easier for others to follow. Not as endless documentation, but as enough structure for someone else to use AI without starting from scratch every time.
Where this points
This is why I keep coming back to training.
If you had instant access to a team of AI-enabled people who were already strong in their own area of expertise, delegation would feel very different. But most small businesses and SMEs do not have that sitting there ready to go. Hiring takes time. Good people are hard to find. And even when you find them, they still need to learn how the work should be done in your business.
So the practical answer is to build the capability closer to where the work already happens. Train people to use AI inside their role. Work through real tasks with them. Show them where the output is likely to go wrong, how to challenge it, and what needs checking before it comes back.
Otherwise, AI does not really remove the bottleneck. It helps the bottleneck person get more done.
If you need to build AI literacy across your team so more people can use the tools properly in their own work, this is exactly the kind of problem worth looking at properly.
