You can't ask AI to manage AI (+discount on Full Stack Founder)
Managing AI agents is a skill and it'll stay relevant until there's another breakthrough in AI development, which may take years.
(Discount code for Full Stack Founder AI Bootcamp at the end of the post.)
The reason jokes about stupid managers will never disappear is because management looks easy but is actually hard.
When I was a CEO, managing others was one of the hardest parts of the job. When I think about my coaching clients, many of the problems they bring into the coaching space are related to management. When I reflect on my experience building software with AI agents, I notice that all my work is management: setting context, defining goals, ensuring high standards, owning the vision. I still have to work to build software just like before AI, but the nature of work is different.
It’s very easy to get excited when trying to use AI agents: they can do a lot of work in no time almost for free. However, the job of the human shifts to managing the agents.
Sadly, we can’t easily delegate this task to another “manager AI” because that would also mean management: explaining how that first AI needs to be managed. At some point, we need to start thinking and communicating clearly ourselves.
, a CTO, writes about one of the reasons this is the case: humans deal with possible, whereas AI deals with probable.LLMs cannot design, because they remix the past. They're incredible at doing so, and the results are sometimes magical, but by their very nature, they deal with the art of the probable.
Humans, on the other hand, deal with the art of the possible.
They can hold a vision in mind that can exist wildly outside the bounds of the statistically likely or reasonable.
No LLM can be trained on what we see in our mind's eye.
Dealing with the probable, holding the vision, determining what’s important and what’s not are firmly within the human domain for now. And, if anything, this is becoming more difficult than before because AI is raising the performance bar, as
, also a CTO, observes:AI raises the bar for speed to "good enough", but actually, value comes from injecting your distinctive human perspective back into the work.
Far from making me go sip matcha lattes at 11am while my computer does all the work, AI is pushing me even harder to do two things. First, understand and articulate what my distinctive human perspective is on the project I’m working on. Second, manage my AI agents well, ensuring they actually do a good job. Neither is easy.
So, while AI is giving us more and more powers, it requires us to be good managers to make use of them. If management were easy, there would be far more good managers and far fewer jokes about bad managers. But it’s hard, especially if you never done it before.
New managers (of people, but also AI) tend to make two mistakes:
Assuming the job is easy, along the lines of “just tell them what to do.”
Assuming the job is so hard, they can’t do it, so won’t even try.
In fact, it’s neither easy nor impossible. But it certainly requires training, dedication and skill to be a good manager of people. I imagine the same holds for AI agents. We can’t just take for granted that we can open a chat interface and “tell AI what to do”. We need to learn the skills of AI management; it won’t come for granted.
At some point, AI will probably get to the point that it’ll be beyond human ability to manage it. Maybe even soon; likely within our lifetimes. But for now, and that status quo may easily last for a few years, the skill of managing AI agents will make the difference between success and failure.
I’ll be speaking and running a practical workshop at Full Stack Founder AI Bootcamp in London in a couple of weeks.
I’ll share my perspective on how to think about AI from a strategic perspective if you’re a startup founder, and share what I learned about managing AI agents, using Claude Code and Replit as an example. The event is designed to be as practical as possible, with all facilitators being practitioners, not theorists:
Rachel Carrell, Koru Kids CEO (and a subscriber!)
Peter Nixey, AI Builder and Coach (and a subscriber!)
Glenn Smith, CEO of fueld.ai (and a subscriber!)
Rob Elkin, founding partner at Rational Partners
David Hathiramani, Talisman Tech CEO
Rod Banner, AI Czar at Positive Change Group (and a subscriber!)
- , CEO at DraftPilot (and a subscriber!)
- , CEO at Alfa AI (and a subscriber!)
Ian Broom, CEO at Fliplet
Tarun Gidoomal, CCO at Blue Light Card
Scarlett McCabe, CEO at Debate Mate (and a subscriber!)
Emma Sinclair MBE, CEO at EnterpriseAlumni
Use the code EVGENY100 for £100 off at checkout. The event is capped at 60; most tickets are already gone.
Hope to see you there!