The More AI Agents There Are, the More I Disappear

These days, it feels like everyone is talking about AI agents. Sleep one day and wake up to a brand-new agent. The tool you used yesterday already feels outdated, and every blog and news outlet says "collaboration is the future."

Splitting work across multiple agents — one gathers information, one writes code, one reviews. It sounds impressive. Hearing people who actually do it, you can't help but think, "That's amazing."

But I don't work that way. Or rather, I can't. Honestly, I wouldn't recommend it.

When You Collaborate, You Disappear

Anyone who's tried delegating work to multiple agents knows what happens. You lose grip on the flow of work. You end up busier tracking what each agent did and why than actually doing the work yourself.

That's what it means to lose the person coordinating the work. The work isn't being done for you — you've been demoted to someone who just monitors the tools' progress. Think about it the other way around: are you using the tools, or are the tools using you?

The industry calls this "productivity," but when you actually experience it, it's something else. You constantly lose track of whether you or the tool is making the judgments.

I Know Because I've Been There — When the Agent Dies, I'm Powerless

This is my story. One day, whether it was a server issue or something else, the agent just stopped working.

That day I literally couldn't do anything. The funniest part? I couldn't even remember the commands I used all the time. Commands I'd typed every day for months just slipped out of my fingers.

In that moment, a chill ran down my spine. When did I stop being able to do things with my own hands? For 30 minutes while the agent was down, I just sat there staring at a blank screen.

That was the price of delegating everything to the agent.

So I Don't Run Automation

Since then, I never run automation. Sure, I still use agents. But not as a boss who does everything for me — more like an assistant.

Gathering information from here and there? That stays the same. Reading docs, checking logs, listening to what others say. I synthesize it, I make the call, I decide. The only thing I delegate to agents is collecting the raw materials I need to make that judgment.

When you delegate everything you've experienced to an agent, problems arise. That's not a choice — it's a consequence. I know because I lived it.

The Right Number Is One, Maybe Two

In my view, one primary agent is enough. If you must, two. Master that one agent deeply, and let the second fill only the gaps the first can't cover.

Many people seem to believe that working with ten agents is better than mastering one. I think the opposite. Managing ten agents means managing ten variables. The more variables, the less control you have.

It's true that agent information floods in like a deluge. But instead of trying to ride that flood, these days I'd rather focus on steering one boat well.