AI is a Bicycle, Not a Cyclist

It was Steve Jobs that first called computers, “a bicycle for the mind.” And it was Satya Nadella that applied it to AI in May of 2023.

It’s a good metaphor. Where bicycles improve the speed and efficiency of your body, AI improves the efficiency of your mind.

AI is, among other things, an incredible aid to your thinking. There is a lot it can do as your bicycle, and there’s one important thing that it cannot.

Just as a bicycle cannot decide where to go, AI cannot be your employee for complex work.

Why? It comes down to how AI functions structurally, not something that technology will quickly solve. The reason you hire a human is to expand context management and make decisions, and only humans can do that. AI cannot do complex work because it cannot self-direct nor gather its own context.

Human Employees

Think about your side hustle if you have one, or one that you’d like to have. It’s going well: customers are increasing, revenue is going up, and you’re beyond breaking even. You think it may be time to hire an employee.

Let’s pause there. Why do you want to hire someone? Borrowing from the Cynefin Framework, I can think of a few reasons:

• Simple: There’s a boring part that you don’t want to do yourself. Perhaps that is answering phones or shipping packages.
• Complicated: The business needs an expert to handle a well-understood aspect of the business that you don’t know about, like accounting.
• Complex: As you expand, there is too much to attend to. You want someone to take initiative and make decisions.

We can consider how AI can serve as each kind of employee.

Simple

What is a simple job? A simple job has a straightforward process and little variation. These are the jobs that don’t require training, like answering phones, making appointments, or checking out groceries.

These tasks are increasingly becoming automatable. And if your side hustle or job has a task like this, do whatever you can to automate it.

You don’t necessarily have to use generative AI to do this. In fact, using AI should be your last resort. Use workflows and programmatic integrations as much as you can; they are faster, more reliable, and much cheaper.

Simple jobs are great; but life is messy and complicated.

Complicated

A complicated job is one that requires expertise, but the task is still defined upfront and the result is easily judged. If your side hustle needs a surgeon, you should hire an expert surgeon.

Complicated tasks are a perfect match for AI. You define the task, and AI provides the expertise. When the task completes, the AI brings you the result or the decision.

You should do this for your website and software, as well as many other tasks. The bicycle is getting better and better, and it’s quickly becoming capable of doing nearly all complicated tasks. Don’t walk; ride your bicycle!

If you do, you’ll be dramatically more efficient and effective at your work. It may be tempting to keep pushing the envelope with AI, but that will be a mistake. AI fails once work gets complex.

Complex

Complex work also requires expertise, but it is your expertise. Your expertise is the reason your side hustle is going so well. But now it is time to expand; you’re opening a second potions and alchemy shop down on 3rd street.

You can’t be in two places at once, nor could you pay enough attention to all of the problems that come up at two locations at once. You need a store manager. The work of the manager is not just scheduling shifts, but foreseeing problems and making decisions without consulting you.

Complex work at scale

You see this at scale as well. Why does Anthropic, the developers of Claude, have 100 software engineering positions open right now? Because they need people to prompt Claude Code using deep knowledge and context about a domain.

There’s also only so much one mind can understand and keep track of. At even bigger scale, consider why there are 10,000 product managers at Microsoft. Because Satya Nadella, even if he is smarter and more talented than any one of us, cannot keep context of each tiny sub area of the company and product in his head.

AI can’t do your complex work

AI cannot do the complex work of managing a storefront because the manager’s job is to predict and act. Managers identify tasks to do on their own.

The problem is that AI only works on tasks that you give it. But the entire point of complex work is that you don’t know the task in advance! Think of it this way: if you ask an AI to predict whether Sally, your best focaccia dimpler, will be on time today, it will give you a prediction. But if you don’t ask, AI won’t answer. AI only does what you tell it. If an AI is in charge of your complex work, your focaccia will go undimpled.

You may try to work around this. Perhaps you’ll schedule cron jobs to have the AI wake up and plan ahead, but anything you didn’t anticipate won’t be handled? You want someone to make a call—someone that has all the relevant context.

That’s a problem too, because AI can’t hold the context of a domain area in the same way as a human. Tech commenters now talk about “context engineering” instead of prompt engineering, because most of the work of prompt engineering is carefully specifying the context necessary to do the task. That store-specific context is why you hired a manager of your second drive-through farriery.

Once again, you could try to work around this. You could think hard about an area for a while and set up an AI agent to handle it based on what you’ve learned. But you’ll move on to another topic and the area will stagnate. No AI on the horizon is continuously learning and generalizing from experience.

If you try to “hire” an AI as your employee for complex work, it will be a disaster. At best you’ll burn out fixing things, and at worst, the lifeless focaccia will accidentally make a love potion, distracting the surgeon and spooking a horse that will send a hoof right through the front window.

Not a cyclist

Artificial Intelligence is a wonderful way to increase your impact. There is a lot of simple and complicated work to do! When well-applied, AI will make your macaroni noodle necklace company much more successful.

But the entire point of complex work is that you don’t know the task ahead of time.

If you misapply AI to your complex work, you’re actually just increasing your own cognitive load. Once the simple and complicated work is out of the way, there’s still a limit to how much complex work you can do. I think that’s what’s going on with the recent discourse on AI fatigue. The work a software engineer can complete in 1 hour could have taken an entire day before, but no one can repeat it eight times in a day.

Crucially, AI is a bicycle and not a cyclist. The bicycle needs a rider in order to do anything. You could add e-assist and GPS navigation to a bicycle, but the cyclist decides where to go and why.

Use the bicycle. Automate your simple work. Hand off your complicated work. When you have too much complex work and you’re getting burned out, hire another human cyclist onto the team. You’ll be much faster if your second bicycle has a rider.

---

Photo credit: User v230gh on Flickr. CC-BY-ND 2.0