AI is not new, but “good generative AI'' is.
Now, anyone can build an AI system — you just need to know how. Before we had ChatGPT, you needed a massive dataset, a PhD, or know how to code. In the new, post ChatGPT world, you don’t need any of that to take advantage of AI. Every time you open a new thread, you’re training AI to understand the context, and what you input is actually what gets it to come up with the right output.
It’s this change in accessibility that’s driving the new way of work that’s emerging. One that’s AI-first. I recommend you start thinking about how you can start taking advantage of AI — especially this early in the game.
And who am I exactly? I’m the CEO of the AI Exchange. Our work in AI Operations helps people and businesses better understand how the application of AI can run a company more efficiently, and what efficiencies you can unlock while doing that. We have served 200+ businesses directly in our first year, and indirectly served thousands more.
When you work AI-first, and think about how to use AI systems to their fullest abilities, you’re not only going to get work done faster, you’re going to get it done more reliably, and eventually, at a higher quality. Let’s take a look at this more closely.
We now know that AI can improve an employee’s performance by as much as 40% compared to employees who don’t use AI. In a similar way manufacturing optimizes workflows to increase their output, employees in every industry should be thinking about how AI can transform their output. In addition to the speed of the completion of work, AI unlocks more reliability and standardization.
When you work AI-first and master how to use AI systems, your work output is higher quality. Not only that, you’re getting a large percentage of your time back — time you can use to consider how to make processes even better. You have more time to innovate and push the envelope. Here’s how to get started.
I’m going to walk you through the playbook that our company uses, not only for our customers, but internally, to optimize our operations with AI. We treat ourselves as customer #1 in all of our research. We push the envelope on how to run our company as AI-first as possible.
The number one question I get asked is, “What are some use cases of AI?” Instead, I encourage you to ask yourself: What is the work I’m currently doing? What are the problems that I’m trying to solve in the business? Where are the bottlenecks?
To help you get started, try the AI task hierarchy. Not all work is created equal, and you need a good representation of what AI is good at. Take everything you do and put it into the below three categories.
There’s a lot of work we’re all doing, especially when you start to break down your day-to-day, that is objectively successful or unsuccessful. Objective work is the type of work that you can look at the outputs of and agree that it was either done right or done wrong.
This work is exactly how it sounds: work that just needs to be good enough. Sending emails, summarizing meeting notes, writing internal documentation, plans, and reports, for example. It doesn’t need iterations or creativity (that’s tier 3 work). With AI optimizing your operations, you may be asking yourself: How do I spend as much of my time as possible on tier 3 work?
When you audit your daily tasks, you’re going to find that you have a lot of tasks within those first two tiers of work. This is a good place to start when building your own AI systems, because this is the work AI will do.
With AI helping you complete tier 1 and 2 work, you have the opportunity to do more of the excellent work — the work that requires expertise, judgment, and creativity, like coming up with the slogan for a new marketing campaign, building out a strategy for how to roll out new support docs to customers that can get them answers faster, or creating an engaging TikTok video to boost engagement.
Once you’ve identified opportunities for AI to assist with operations, you have to build the systems. For someone who doesn’t have an engineering background that can seem pretty intimidating — I promise it’s not.
A lot of us are still using one time, single use tools, and when we go back into them, they likely didn’t learn from what we did before. Finding systems that actually improve on themselves and that you have the ability to improve on is a huge unlock for starting to get to the place where you’re working AI-first.
The best way to build an AI system is to start with a Standard Operating Procedure (SOP). Start breaking down a plan for how to get AI to do the work the same way that you would break down delegation to someone on your team. The key to getting AI to do what you want, is knowing what you want done.
One thing that I hear a lot about SOPs is, “We have SOPs for the obvious stuff, like how to refund a customer, but what about all this other stuff that’s more complex and more creative?” I’m going to challenge you to SOP tasks that you didn’t think were SOP-able before.
For example, we write a newsletter twice a week where we cover the most important updates and perspectives on AI. At the beginning of the year, this would take me 4-5 hours every single time I would write it. It was a really creative experience for me, and I was really trying to hit that excellent work. But I want to grow my company, I want to scale my company, and I also want to challenge myself to be as AI-first as possible. So I sat down to try to see if I could take something that I thought was really creative and put that into some SOPs to see if I could get AI to help me with some of that.
Following the newsletter example, a simple SOP might look like:
When we convert that into an AI system, it looks more like this:
In this process, I thought about if I wanted someone else to write the newsletter, what would I want them to do? With that thinking, you can find opportunities for AI to intervene and take on some of the work. Each SOP we have is a different AI system that we’ve set up. Now our newsletter takes me 20 minutes.
But what needs to happen to not only make the system, but get it to a place where it works really well? You need to get really good at integrating your expertise.
The more specific and fool proof you can get with how you’re describing what you want, the more success you’re going to have with AI. The best way I can describe this is to be bossy. You need to tell AI exactly what you want done and figure out how to translate that into these systems, so they can refine and get better over time.
Here’s a tip for you: You already have all the tools you need to start trying to get it to do what you want — it’s called a prompt.
We have over 150 prompts that power our business . We don’t have a single fine-tuned custom model. We’ve gotten all the results we’ve been pushing for with prompts. This means that as you’re thinking through ideas and what you want, you can write prompts too. You don’t need a heavy investment in fine-tuning something custom.
Treat your prompt as if you’re creating a project brief — that’s going to get the AI to go further. Here are five of the top tips we give to our customers when creating prompts:
The number one reason AI projects fail is poor expectations and not planning for reality.
Once you have those first three steps in place — you find out what you’re trying to do, you start building a system, and you’re starting to write prompts — you get to this place where you think you know what will work. This is the most dangerous place to be because your expectations are really high.
More likely, once you start using or delegating to AI, there are unforeseen circumstances and issues beneath the surface. Disappointment Is common. As with most projects, there’s a lot to uncover during the process.
Most people jump to the conclusion that AI can’t get the job done. Instead, I want to challenge you to anchor yourself to lower expectations, and to realize that you’re going to have to put work in to get it to the place where it’s working really well. You need a human in the loop.
Figure out how and where you can have this person give feedback to the initial prompt and continue to iterate on it. You’re going to find that the 5th, 6th, or 7th time you update the prompt is when it’s going to get way better. That is how you get this technology to do what you want.
Think about it in the context of self-driving cars and how that industry has progressed. They have 5 different levels of autonomy outlined to show exactly what level of expectations they have on the AI and how they are designing for it.
Any use case that you have in your work, figure out where to be in this progression, and then don’t be disgruntled if you think you can start at level 5.
The next goal in work isn’t going to be doing it, it’s going to be delegating it — and figuring out how to delegate to these systems.
I encourage you to slow down where there's an opportunity to try this. Think about how you, your team, or your company can get the most out of these AI systems. Dig into how you can build a system and put a strategy in place to encode your expertise. And finally, figure out how you’re going to get to the place where there’s a human in the loop and you’re refining the system to be more valuable over time.
In thinking about how you really get to the place where you as an individual, your team, or your company can get the most out of these AI systems, I encourage you to slow down and think about questions like: What are the opportunities? How do we build a system? How do we start to have the practice of encoding our expertise? How do we finally get to the place where we have a human in the loop and we’re refining to get it to be more and more valuable over time?
You don’t need to become AI-powered overnight. But if you start now, you’ll get there faster.
Evolve your team, strategy, and tech stack for the AI-first future.
Get the toolkit
