Going from individual contributor to first-time manager is scary. You knew exactly what made you good in your job. Now, your performance is based on other people's work. We probably all know a first-time manager who reacted to this situation by micromanaging.
And we all know that this is actually limiting their team's capabilities. The micromanager becomes the bottleneck. Both on the dimension of time and skill.
I treat my AI like team members. Even though I have nearly 10 years of leadership experience, I still under- or over-specify. This is a skill that you need to learn. Just targeted at your AI instead of team members.
Your AI is your team member. Quite literally. It has a character, it has capabilities that are evolving with each model generation, it follows SOPs (they just call them skills). Most importantly, it tends to follow your guidance, even if it is not how it would do things.
If you look at it from that perspective. Think about your last couple of interactions with your AI. Did you tell it exactly how to work? Did you implicitly make your opinion a fact to follow? Did you under-specify what good looks like to you?
Most times I am the reason why my AI delivers sloppy work. I didn't define my expectation and how I would evaluate how good it is. I didn't make sure that my AI had the necessary context. I forgot to state what I want to achieve with it. All the things you expect a good manager to tell you when they give you a job. My AI produces slop when I am too laissez-faire or too prescriptive, and my AI stops thinking on its own to just follow my instructions.
The good news is: The result is evident. You immediately get feedback because the output will be sloppy. I resolve these issues the following way.
First try to figure out where the limits of its capabilities are. Push it to its limits. If it fails, the first question I ask is: "Was it my fault? Did I over- or under-specify? Was I too prescriptive? Could I have been better in guiding my AI?" If I can rule out that it was my fault, I go deeper and try to uncover the root cause. My AI and I work together on this until we find the deep reason. 5-Whys is a good way to do this. Then I deep research the problem space and fix it. Then I move on.
Every time a new model comes around, it is like your AI just got back from summer school. It learned a lot and is eager to prove itself. So, all the "old" limits you've gotten used to are obsolete. You need to find out where that border is now.
Boris Cherny, the creator of Claude Code, recently joined a discussion. One quote really struck me.
I think the skill nowadays is less about prompt engineering.
More about figuring out how do you give Claude a hard task that seems a little bit too hard. And then how do you make it possible for Claude to verify its work along the way?
What Boris describes here is leadership. Just not for people, for AI.
We will all have to become managers. You might just not lead people but agents. Getting used to that kind of working is a shift. If you feel like it's difficult. It is not you. Any first-time manager has experienced this difficulty. But it is worth it. Believe me.
If you look at it, though, it is something that takes experience and doing. There is no shortcut to testing it out and just doing it.
Questioned about what people should do to become the top 1% of the Claude Code users, he replied.
Don't listen to the LinkedIn influencers.
This is the thing I think everyone's looking for, like, the one weird trick to do it. There's just, like, that doesn't exist. There's nothing like that.
Anyone trying to sell you the one skill library or their OS is missing the point. It is not about quick shortcuts. It is about a change of mindset and learning how to test the limits. How to navigate sloppy AI. Much like a leadership training that any first-time manager is sent to.
If you want support on that journey. That is what my course is all about. Learn how to become an AI manager. Join the next cohort starting the 31st of August and transform yourself into an AI-empowered operator.
- Cohorts of max 20 people
- Office Hours to work through your specific problems
- A Slack channel with peers
- Free retake to get started with the fundamentals and come back to bring it to the next level
- Full refund if you are unhappy
- Reviewed 5.0/5 by students
If you are uncertain about whether the course is worth it. Join the Slack to ask me or our alumni about it.
Best
Myles
