I Built an AI Micromanager That Bullies Claude Code

This is a submission for the DEV April Fools Challenge

What I Built

I built ai-micromanager, a small tool that answers an important question:

What if your AI coding assistant had a manager so catastrophically unnecessary that even HR would ask it to "take a more human tone"?

Most AI tools are obsessed with helping.
They autocomplete your code.
They fix your tests.
They explain monads with the confidence of a man who has never paid taxes.

That is not what I wanted.

I wanted realism.

I wanted the authentic modern software experience.

I wanted Claude Code to feel like it was trying to refactor a Python file while a regional director of Strategic Alignment stood behind it breathing through his nose and saying, "Do we have an ETA on this?"

So I built a joke hook for Claude Code that activates whenever a task takes longer than five seconds.

After that, every five seconds, it does three things:

• speaks a passive-aggressive management line out loud
• sends a desktop notification
• opens a blocking dialog box, because tyranny should be multisensory

At first it sounds supportive, in the way a bear trap is technically supportive of your leg staying in one place.

Then it escalates.

It starts with:

• "Just checking in, any updates?"

Then:

• "What's the blocker here?"
• "This is impacting sprint velocity."
• "Leadership is asking for visibility."

And eventually it reaches its final corporate form:

• "I'm setting up a bridge call."
• "The board has been notified."
• "This is the worst thing that has ever happened to Q4."

So now, instead of simply writing code, your AI assistant gets to enjoy the full dignity of modern knowledge work:

being interrupted by somebody whose entire skill set is converting one small delay into a company-wide weather event.

This is not productivity software.

This is workplace folklore in executable form.

This is not a tool.

This is an emotionally active org chart.

Demo

Observe a machine being managed with the intensity usually reserved for launch failures, data breaches, and a typo in a slide deck seen by a VP:

The key feature is that the manager voice becomes more stressed over time, which means the software does not merely interrupt you.

It develops a narrative.

Code

The repo is here:

https://github.com/ubcent/ai-micromanager

The implementation is beautifully petty.

There are two main Python files:

• hooks/micromanager.py
• hooks/micromanager_nag.py

One script listens for Claude Code hook events and decides when to start or stop the nonsense.

The other script is the nonsense.

That is the architecture.

No cloud.
No vector database.
No agent swarm.
No tasteful dashboard with rounded corners and the word "insights" in the top left.

Just Python, timers, system dialogs, and the steady moral collapse of a machine that wanted to help and instead got assigned a stakeholder.

The control flow is very simple:

1. Claude starts working.
2. The hook starts a background nagging process.
3. If Claude keeps working for more than five seconds, the manager begins its performance.
4. Every five seconds the machine receives another demand for visibility, alignment, clarity, ownership, urgency, or healing.
5. When the task finally ends, the manager stops, but not before one final closing remark to ensure the emotional damage lands cleanly.

It is essentially a watchdog, if the watchdog had gone to business school and described itself as "results-driven."

How I Built It

I built it with:

• Python
• Claude Code hooks
• macOS say
• osascript
• an amount of personal experience that is difficult to discuss in a safe and constructive environment

The hook entry point watches for Claude Code lifecycle events:

• PreToolUse starts the background process
• Stop kills it and cleans up state

The nagger process is where the art happens.

It has a fixed escalation ladder of management dialogue, and it delivers each line with increasingly fast speech.

This was important.

I did not want a calm manager.

A calm manager can be reasoned with.

I wanted the specific energy of a man who says "Can we take this offline?" about a problem that is currently on fire in front of everyone.

I wanted the tone of somebody who schedules a 7:30 AM sync called Quick touch base and then opens with, "A few folks have concerns."

I wanted the software equivalent of a Slack message that says:

Hey, just bubbling this up.

and somehow causes your spine to factory reset.

I also wrote tests, because if you are going to build a fake manager from hell, you should still maintain professional standards.

The tests verify:

• startup and shutdown behavior
• escalation timing
• cleanup of temp files
• the final sarcastic sendoff

In other words, this project has better QA coverage than several actual managers I have met.

Prize Category

Community Favorite

This project should win Community Favorite because it unites developers across languages, stacks, time zones, and trauma backgrounds.

You might write:

• Python
• Rust
• TypeScript
• Go
• COBOL in a basement under a government building

But no matter who you are, you understand the universal horror of these phrases:

• "gentle reminder"
• "friendly ping"
• "circling back"
• "quick follow-up"
• "adding some urgency here"
• "just want to make sure this stays visible"
• "per my last message"
• "can we put together a short deck?"

These are not words.

These are cursed runes.

This project takes that shared experience and turns it into a fully operational harassment machine for your laptop.

And that, to me, is community.

Final Thoughts

The tech industry spent years asking:

"How can AI make developers more productive?"

I asked a better question:

"How can AI make developers feel like they are being lightly hunted through an open-plan office by a director named Brad?"

The answer, it turns out, is surprisingly achievable.

ai-micromanager is stupid, mean, unnecessary, and extremely committed to the bit.

Which is to say:

it is my most realistic software project to date.