How do you decide what to do next with AI?

Why do you need a thinking system for AI?

AI has a huge amount of possibilities, and it's quite easy to get lost in this. You can just use it and start with the first thing that comes to your mind. But what comes to your mind can be organized and filtered, so that you come up with the right ideas and prioritize them before you actually work on them.

That's what a thinking system is for. It gives you a predefined lens for looking at reality, and that lens increases the chance that you see the things that are necessary for you. Mine stands for Automate, Enable, Stabilize and Accelerate. It's set out for a business that is generally working: some revenue, some history, a few employees. If you're starting from scratch, other thinking systems might serve you better.

What should you automate with AI first?

Things that you are already doing and that contribute to the success of your business. Both parts matter. Things you already do that don't contribute very well are not the right things to automate. Things that would contribute but that you don't do yet are not automate either. They are enable, and they come next.

Automate means you take a task that you do by hand today and automate it with AI. The end goal is that you don't have to do it anymore, or only a fraction of it. You automate 70, 80, 90 percent. Maybe the last 10 percent stays with you, because you still need to check if everything is okay, or it needs the human touch. Human in the loop is still very important.

Think about things you do all the time. A customer wants an offer from you. You have a call, they tell you what they want, and you agree on something. Then you have a transcript from that call. You feed it into AI and connect it to how you create offers and what your value propositions are. From the transcript it can draft the offer so that it matches what your customer wants, without the things they don't need and without missing what they asked for.

You don't have to do it yourself, and you get consistent quality. Consistent quality comes a lot from standardized processes. In reality you often have a process, but then you don't have as much time as you want, or you forget something. With automation, part of the goal is sticking to the process and keeping things tidy.

What can AI let you do that you couldn't afford before?

That's enable. Once you have automated a few things, you have some experience in how to make AI do the work you want. With that, you can do things you haven't been able to do in the past, or that weren't viable from a return on investment perspective. They still need to add value. You do something that serves your business, but that you couldn't do before.

One example for me is my YouTube channel. I never had motion graphics or sound design, and I very rarely did snippets for social media, because it was just too much effort. If I had paid a cutter to do it, it would have been a lot of effort. Thumbnail design was partially in that range too. So I just didn't do it. AI enabled me to build something that gets the motion design and the snippets done. I didn't do it before. Now I do it.

You don't have to automate or enable everything. Automate a few smaller, simpler things, then enable a few smaller, simpler things.

Why should an AI system be stable before you speed it up?

Because it's very dangerous to accelerate something that's not stable. Think about a car. You wouldn't want to accelerate a car that isn't stable to 200 kilometers per hour. You stabilize it first. You make sure there are no security issues and that it won't blow up in your face.

One part of that is spec-driven development. You describe how the output is supposed to look. Based on this description the system is built, and a lot of tests are built as well, tests the system has to pass again and again. When you keep developing the system, it doesn't suddenly lose capabilities or security functions it had earlier. It's an enhancement of what was earlier called test-driven development.

Another part is evals, short for evaluations. That's a set of things a system has to pass, graded from, let's say, one to 10. Only if it passes them at nine of 10 is it considered okay to run. It's similar to tests, but it's not yes or no. With AI it's more often a question of to what degree, judged by other agents that test it.

Is security part of a stable AI system?

Yes. As you have probably noticed, there's currently a huge influx of cyber crime, and hackers have access to AI a lot. Making sure your system has a certain level of cyber security, and that you're not using outdated and very vulnerable systems from the 90s, is very important. You don't want to scale a system that has security vulnerabilities, because then it will blow up in your face.

Why cut complexity, even things you like?

Stabilize has another component, and that's eliminate. Any system that becomes too complex, with too many features and elements, will become unstable over time. So think about which of the things you have are actually necessary, and take out the others that don't contribute much in a complexity to value ratio.

If something is too complex given the value it creates for your business, it's often better to remove it. Even if you like it and love it and it looks fancy. Making a system more simple adds a level of stability by removing complexity.

What does accelerating with AI look like?

Once you have a system that is stable, you can accelerate it. How can you make it work faster, without friction points that increase the time it takes? Take offer creation again. If every offer has to go through multiple iterations with you, it automatically takes more time. If your offer creation gets to a point where you look at it once, and usually don't have to correct anything or only once, that's very fast.

You get off the call with the customer, and within five minutes they get a mail: this is your offer, just as you wanted. And they say, this is exactly what I wanted. That speed adds another layer of value. Speed is also part of the whole scalability game. If you can do in 15 days what took you 30, then of course you can do twice the amount in 30 days. That's the basis for getting more customers.

In what order do you run the four steps?

You go through it again and again. You automate something, you enable a new opportunity, and you don't need to do both at the same time. I would prioritize automation and always start with that first. If you don't have anything good to automate, look at something to enable. Always run stabilize in a loop. Accelerate is something you can park for some time, until automation has gathered most of its value.