My wife named her Chat GPT app. It’s called Sabastian. It was a brilliant move.

I’m a slow AI adopter. We know about so many problems with it. It tells you what you want to hear, even if it’s bad advice. It confidently lies to you (See Note 1). It invents false “facts” to support what it wants to say. It sometimes struggles with math. It misinterprets information.  It isn’t careful with information you share with it. The problems are even bigger for people who use it for mental health advice (research article here). But it was a superpower to start thinking of AI as a “person” with strengths and weaknesses, even if that’s not strictly accurate. Here’s why I say that:

I reread the business classic book 7 Habits of Highly Effective People (Covey) almost every year, and I’m reading it again now.  This read-through, a quote stuck out:

“Many people refuse to delegate to other people because they feel it takes too much time and effort and they could do the job better themselves. But effectively delegating to others is perhaps the single most powerful high-leverage activity there is.”

Covey describes effective delegating as the main difference between an individual contributor and a “manager”, between getting things done and creating a system where others can help you do more of what matters most. Yet managers have a difficulty especially in the early days — delegating to others feels initially slower and worse quality than just doing the task yourself, but leaders have to overcome that initial reluctance to get the benefit of your team’s strengths and skills as they develop. That was the experience in the 80’s when Covey was first writing and it’s the same today.

Here’s what struck me: What if it’s also true with AI “team members”? So let’s do a thought experiment: Imagine I have an AI assistant. Let’s call him Sabastian. I should ask myself what do I want my AI assistant Sabastian to do for me? For sure there are some things that my AI assistant can’t do as well as a human assistant could, but some things it could to better. Laying that out is for another post, but others have made suggested lists here and here.

Here’s a key point: I need to check Sabastian’s work. He’s not good enough to rely on his work as is. One day he might be, but not today. How big of a weakness is that? Think of him as a junior team member, or perhaps an “apprentice” working under the supervision of a “master” (see note 2).

AI may or may not ever get that good enough to do work completely independently. Lots of people think it will, but I’m a skeptic. Even so, let’s think of our AI assistants as apprentices–they do whatever you ask them to do, and they do good work, but it needs our supervision and guidance to do the work well. That means you will need to:

• Work with it on prompting to get your deliverable in a format and style appropriate for your industry
• Verify its claims, citations, and calculations for accuracy
• Ask it for an opposing perspective of the advice it gives you
• Give feedback for how it can serve you better, and give it a chance to apply it

In other words, treat your AI assistant something like a new team member who needs guidance and feedback to reach their full potential. And now, no one has an excuse not to delegate

From here on out, I want to say that Sabastian is a low-cost member of my project team. He has some big weaknesses, but he also has some big strengths. If I choose to let him sit idly, that’s on me as a PM. But he’s ready and willing to be put to work on whatever I ask him to do. Let’s not miss the opportunity to delegate work to him. He’s getting better and better every day.

---

Notes

1. Just this week I asked if there was a train between two cities that were a 10-hour drive apart; AI emphatically said there was and it told me the name of the train, but upon looking into this train, I found that the train it suggested would only take me about 45 minutes to the bus station, where I had to sit on the bus for 10 hours anyway. Definitely not what I was trying to ask.
2. My background is in engineering, so originally I thought about it that way, but it felt like a tangent to explain it in the main body. For those interested, here’s how the system works: In the US, the official practice of engineering works something like an apprenticeship system. If a project needs formal approval by an engineer (usually this is construction projects; not all projects need it), an engineering company with licensed Professional Engineer (PE) is needed to approve designs, but experience is needed to get this licensure. A new graduate engineer can be designated “Engineer in Training” (EIT), and he will do important work, but he will do it under the supervision of a licensed PE, who is ultimately responsible for his work. So although EITs perform valuable work for engineering firms, the firm can’t use their work unless it’s reviewed first. So a major part of a PE’s work is reviewing and correcting the work done by the EITs under their supervision. In the early days of an EIT’s work, there are a lot of mistakes, but over time the EIT will learn and get better and better, and eventually the EIT will qualify for PE licensure of his own. In this context, Sabastian is a bit like the EIT, and I’m the PE. I can use Sabastian’s work, but I need to review it for accuracy, and I’m ultimately responsible.