For most of my career, managing people was simply the next step. Across management, executive and founder roles, I have managed and led my fair share of people. For an ambitious professional, that was the normal career ladder to climb, so I climbed it. And I did quite well at it, which surprised a few people, as I can have a somewhat abrupt style.
Almost 3 years into a very different path, and after most of a year working with AI agents every day, I've come to a conclusion the younger me would not have expected: I probably don't want to manage people anymore.
Where my drive is
My heart and my drive have always been in being hands-on. Doing things, building things, being on the front line. The higher you go up the ladder, the less of that you get. More of your time goes into managing people, internal presentations, numbers, reporting and politics, and less into doing things yourself. You ask others to do the work instead, which, given the high bar I set for myself and for others, was a frustrating experience at times.
So I checked the box. I proved to myself that I could be a successful executive in SaaS startups. I also built something from scratch with my co-founder, and sold it: someone saw value in what we had built. That proved it to others too, and it added some interesting lines to my CV. But it didn't change where my drive is.
What OfficeBots taught me
My second startup, OfficeBots, was about automation for businesses, before AI as we know it. The platform let you create email-based assistants: you sent a bot a predefined task by email, and a few minutes later it answered you with the work done. I built the platform myself, and it taught me a lot.
Selling it as a solo founder taught me even more. The premise was simple: email is the common denominator in every business, the lowest point of friction there is. But to get the bots running, I had to set them up myself, for each client. That's where I met the real challenge. People couldn't answer my questions, or gave me wrong answers based on wrong assumptions, or couldn't describe the ideal state they wanted. Sometimes they asked for a faster horse instead of a car.
That is the pain of selling technical solutions to non-technical buyers. Getting from what someone has in their head to something you can build is hard work, and it was always mine to do.
Going back to individual contributor
In 2023 I read Tech-Powered Sales, by Justin Michael and Tony Hughes, a book about technology and automation in sales. It made me ask a simple question: what if I went back to an individual contributor role in enterprise SaaS, and used my automation and AI skills to increase my own productivity and impact, instead of trying to help others increase theirs?
These were the early days of AI as we know it today. I was trying to fine-tune my own model with code, which, spoiler alert, I never finished, and actually gave up on pretty quickly. But my mind was already in AI, and I could see a fraction of the possibilities. So I accepted a role at Kaltura with a clear ambition: become a 10x, or 100x, employee, using technology.
Building NicAI
The first 2 years were about building my own knowledge base, NicAI, and all the workflows and pipelines to keep it as close as possible to everything I know. That meant tons of time, trial and error, and tests of new models, new harnesses, new tools and new APIs.
The unlock came with Claude Opus 4.5 at the end of 2025. I jumped on it in January 2026, and that's when the agentic era started for me. Today I spend most of my days in Herdr, a terminal multiplexer, with between 20 and 40 agent windows open, and 1 to 5 of them active at any given time. For the last 6 months I've been on the highest Claude subscription, $200 a month. It buys me AI credits in rolling windows: a weekly limit, and a shorter window limit within it. Hit one, and you wait for the reset.
I'll skip the techniques I learned along the way. The frustrating part is that they expire quickly: models change, tools change, and I change my setup often. What you learn deep in AI that lasts is the logic and the systems. The specific knowledge has a short shelf life.
Intelligence on tap
Those agents give me an army of superintelligent anything at my disposal: market researcher, writer, designer, builder, IT, marketing, DevOps, whatever else I need. And I can direct and shape that intelligence exactly as I want it.
This is exhilarating. I'm deep in AI, and still, every few days, I have a WTF moment where my mind is blown by what's possible. Even more when I take a step back and remember how much time, resources, effort and money it took before AI to build an app, a website, a process, a presentation or a one-pager. Now all of it comes with a subscription. Intelligence on tap.
The overhead
Looking back, I find managing people adds a lot of overhead. To manage and lead people properly, you need to understand their background and their experience, spend time with them, and understand what drives them and what blocks them, so you can unlock their power and their impact for you and for the team. That is real work, and I did it for years.
With agents, all of this goes away. I can ask for something very specifically and precisely, and get exactly what I want in return. Or at least closer and closer every day, as the models improve. And already today, closer and better than what people would have given me. In minutes, or a couple of hours at most: a laughably short time to get documents, presentations, apps, websites, anything.
Probably
I proved I could manage and lead people. But my drive was always in doing things myself, and for the first time I have a way to get the output of a whole team without leaving the front line. Things move too fast to say "never", hence the "probably" in the title. For now, I'd rather direct agents than manage people.