I’ve realized that I’m a micromanaging boss.

Not to a human, thankfully. As most people know I am pretty relaxed and hands-off. No, I’m a micromanaging boss to my AI agents. And it’s a bad habit I’ve caught myself falling into far too often lately.

I’m talking about sitting there and watching the AI work.

The Blinking Cursor Trap

You know the drill. You type in a prompt, hit enter, and then you just stare. You watch that little green dot pulse, or you watch the letters stream across the screen. When AI was just summarising a quick email, this was fine. It took three seconds. But things are changing fast. We’re not just using simple chatbots anymore. We’re starting to use agents.

These are tools that perform complex, multi-step processes. At work, I might set off an agent that has to run through six or seven different steps. It has to search, write, test, and verify. This kind of work can easily take five or ten minutes to complete.

And yet, what do we do? We sit there. We stare at the screen. Personally I like to look at the reasoning traces to see what it’s “thinking”. It’s the modern equivalent of watching water boil, and it’s a massive waste of time.

The Shift to Asynchronous Thinking

The fix is simple but hard to put into practice. Soon we will need to stop treating AI like a synchronous search bar where we wait for a reply. Instead, we have to start treating it like a team of people that we manage. When you delegate a task to a colleague, you don’t stand over their shoulder and watch them type. You give them the brief, you tell them to get on with it, and you go back to your own desk.

We need to do the exact same thing with our AI agents. We need to stack up our tasks, kick them off, and then walk away. Let the agent do its thing in the background while you go off and do something else. You’ll get a notification when it’s finished.

Catching Myself in the Act

I realised I was failing at this myself. I write this very blog with the help of a custom agent pipeline. I take a raw audio dictation, get it transcribed, and then my agent runs a massive multi-step process. It creates the draft, refines the style, writes the image prompts, and even drafts the social media posts.

It’s brilliant, but it takes about 5 minutes to run.

And for the longest time, I would just sit there. I would watch the console, checking if the formatting was right, making sure it didn’t go down a weird rabbit hole. I was managing the agent like a micromanaging boss. It was ridiculous.

Once I forced myself to stop watching, everything changed. Now, I do my personal projects on my commute. When I’m on the tube, I’ll use Claude Code on my phone. I’ll fire off three different tasks at the same time. Then I just lock my phone, browse the web, or listen to my audiobook. When the agent is done, I get a ping, I jump back in, review, and keep moving.

The “Agent Dance” Workflow

To make this asynchronous style work, you can’t just throw random tasks at a bot. You have to start thinking like a project manager. Here is how I map out my work before I touch a single prompt:

1. Look at the Big Picture: List (or at least hold in your head) all the projects and tasks you currently have on your plate. You need a bird’s-eye view of your workload before you start delegating.
2. Spot the AI Tasks: Look at each task and ask: is this something an agent can do? (e.g. transcribing an audio clip or compiling research).
3. Find the Connected Tasks: Work out which tasks are reliant on each other. If Task A (generating a draft) has to happen before Task B (writing LinkedIn posts), then Task A is your first AI hand-off.
4. Plan Your Next Move: This is the key. While the AI is busy working on Task A (which might take five minutes), what human task can you switch to? You need to know exactly where your attention is going before you hit send.
5. Do the Dance: Trigger the AI agent, immediately move to your human task. When the agent ping arrives, pull in the finished work, kick off the next dependent task, and repeat the cycle.

And I am telling you this becomes automatic very quickly.

The Attention Tax

But we have to be realistic here: this style of working has a cost. There’s a real cognitive tax that comes with it.

If you’re running three or four agents at once, you have to be able to hold all those different projects in your head. When they all finish at different times, you have to switch contexts fast. You have to jump into project A, see where it is, make a decision, then jump straight into project B.

This level of task-switching is exhausting. It’s a completely different skill set than just sitting down and doing one thing for three hours. But it’s the skill set we’re all going to need.

Obviously, you can decide how to manage your own focus. But don’t underestimate the sheer mental energy it takes to keep track of multiple digital streams running in the background.

The New Resumes of the Future

I honestly believe this is going to be the defining shift over the next few years. The people who get ahead won’t be the ones who write the best prompts. It’ll be the people who can manage the most agents at once. And doing so well!

I wonder if we’ll start seeing CVs in a few years where people list “ability to juggle 40-50 agents simultaneously” as a key skill. It sounds weird now, but that is exactly where we’re heading.

So stop micromanaging your AI. Set the task, and go do something else. Let the machine do its thing and you do yours.