The Decision Engine
Delegation, AI, and the blank page problem
A good CEO is a hard-to-automate decision engine. That’s Cal Newport’s framing, and it’s useful because it explains what the job actually is. Your deep work as a founder isn’t making things. It’s the judgement about what gets made, how it gets made, and to what standard. The output matters, but the decisions that shape the output are where your time is most expensive.
Most founders don’t start there. They start by wearing many hats because they have to. They write the code, design the screens, make the sales calls, and because their taste is good and they care more than anyone else possibly could, the product turns out well. Customers overlook the rough edges because the thing they built solves their problem better than anything else. And for a while, being the person who touches every surface is exactly what the company needs.
The problem is that at some point, this stops scaling. They slowly become the bottleneck, because every decision still routes through them and every piece of output still needs their hands on it. Every hour they spend on a tactical task, fixing a bug, drafting a routine email, rewriting a paragraph that someone else could have gotten 80% right, is an hour the most expensive strategic resource is idling. Their taste and judgement start working against them, not because they’re wrong, but because they can’t be in enough places at once.
This is true whether you’re hiring your first employee or picking up your first AI tool. The emotional friction is surprisingly similar, and the thing you have to let go of is harder to name than you’d expect.
Trust runs in opposite directions
When you delegate to a person, you give trust to unlock them. A good hire needs autonomy to do their best work, and an important part of that autonomy is accepting that they’ll create something that reflects their taste, not yours. They’ll make choices you wouldn’t have made, structure things differently, bring perspectives you didn’t have. If you hire well and then dictate every decision, you’ve essentially hired a very expensive pair of hands. The letting-go, with people, is accepting a different flavour of good.
A good hire grows. Over time, they need less checking, less direction, less of your attention. They develop their own judgement, and eventually they’re making decisions you wouldn’t have thought of, not just approximating the ones you would have made. The bottleneck loosens because they genuinely take things off your plate.
When you delegate to an AI tool, you build trust by verifying output, not by stepping back. You’re trying to replicate your standard, and the setup is how you do it: guidelines, style notes, frameworks, well-formed ideas. You encode your taste so the tool can approximate it, and trust comes from seeing the output match what you would have done. The letting-go, with AI, is accepting that you didn’t make it with your own hands, even though the result looks like yours. If you edit a draft four times until it meets your standard, the standard is the ownership. The process of getting there is just the logistics of the cold start.
The AI bottleneck loosens differently. The tool itself doesn’t learn your taste, but the system around it does. Every time you update a style guide because the output kept missing something, every time you refine a framework because you spotted a pattern in the mistakes, you’re encoding corrections the same way you would when coaching a person. The difference is where the learning lives. With a person, it’s in their head. They absorb it, and eventually they don’t need the style guide at all. With AI, the learning lives in the artefacts, the guidelines, the prompts, the setup. Remove those and you’re back to zero. A person eventually surprises you with something you wouldn’t have done. An AI setup gets more reliable but, for now, never more creative. Both trajectories point up, but the ceiling is different.
But editing and creating aren’t the same weight of effort, and this is where the math changes. The real bottleneck for most founders isn’t the volume of work. It’s the blank page. Whether you’re writing something, building something, or trying to execute a new idea, the cold start is what kills momentum. You sit down, stare at the empty screen, and three hours are gone before you’ve produced anything useful. Not because the work is hard, but because starting is hard.
If something else handles the cold start, you skip the worst part and go straight into correction mode, which is where most experienced operators are strongest anyway. You’re not creating from nothing. You’re looking at a first attempt and immediately seeing what’s wrong with it, what’s missing, what needs to be sharper. The bottleneck doesn’t disappear, but it shrinks from hours to minutes.
The lathe and the chisel
Some people want to hand-carve everything, and there’s nothing wrong with that. Developers who resist AI tools, founders who can’t delegate, writers who won’t work with editors. It comes from the same place: if I didn’t make it with my own hands, is it really mine?
The thing about a lathe is that it doesn’t actually remove the carpenter’s hands from the work. You’re still holding the chisel against the wood while it spins, still reading the grain, still deciding when to press and when to pull back. It’s arguably more demanding than carving a stationary block, not less. But it changes the nature of the skill. Your experience shows up in the decisions you make about pressure and angle and when to stop, not in the marks your hands leave on every face of the wood. The wood is already spinning. Your job is to know what shape it should become.
Where hand-carving becomes a problem is when it’s really about avoidance. Founders sometimes gravitate toward building things themselves because execution work has clear requirements and a known path. It feels like progress, and it is progress, but it’s progress on the comfortable stuff. The scarier work, the creative decisions, the strategic bets, the things that might not land, those are easier to put off when your calendar is full of execution.
That doesn’t mean all execution is avoidance. Encoding your taste into guidelines and frameworks and clear direction, that’s real, necessary work. It’s the difference between insisting on writing every line of code yourself and investing in the system that lets your team write code that meets your bar. One keeps you busy. The other actually scales.
The messy middle
None of this is clean in practice. When you start delegating, there’s a period where everything gets worse. The first hire produces work that isn’t as good as yours, and you spend more time giving feedback than you would have spent just doing it yourself. The first AI-assisted draft reads like someone else wearing your clothes. You look at the output and think: I could have done this better, faster, if I’d just done it myself.
And you’re right. In the short term, you probably could have. That’s the trap. So a lot of founders try delegating, feel the pain of worse output, and pull everything back in. They conclude that nobody can do it as well as they can, which is true right now and completely irrelevant to whether they should keep trying. If you keep doing everything yourself because you’re better at it today, the company stays single-threaded. And a single-threaded company eventually dies of exhaustion, not competition.
Some founders know they should delegate, they agree with the logic, and they still don’t do it, because the work they’d delegate is the work they love. That’s honest, and there’s no clean answer for it. For some, the move is to bring on a CEO or CTO and step into a different role entirely. For others, it’s learning to find satisfaction in the directing and editing and shaping rather than the making from scratch. Not everyone can make this shift. Some founders need to step aside, not because they failed, but because the company needs something different now.
Either way, the decision engine doesn’t run itself. Somebody has to operate it. The question is whether you’re going to spend your time on the decisions that need your taste, or on the execution that doesn’t.