Every few weeks a new model comes out. It is faster. It writes better code. It scores higher on some test I have never heard of.
And every time, my real problem stays the same.
The problem is not that my agents are dumb. The problem is that I lose track of them.
Three agents and no idea what they are doing
It starts small. One agent fixes a bug. Another one writes tests. A third one updates docs, because I finally feel brave.
Ten minutes later I have three terminal tabs. One is waiting for permission. One finished and is just sitting there. One is stuck in a loop, reading the same file again and again.
I do not know which is which. So I click through tabs like a nervous parent at a playground.
That is not a model problem. That is a management problem.
A board is older than AI
The fix is not new. It is a kanban board. Cards. Columns. Things move from left to right.
Teams have used this for decades. Sticky notes on a wall. Nothing clever about it.
When I started seeing tools that put agents on a board, it felt obvious. Of course. If I treat agents like workers, I should manage them like workers.
The columns are simple. Working. Done. Idle. And one more.
The column that matters
The most important column is not "Done". It is "Needs You".
An agent that finishes is fine. I can look later. An agent that works is fine. I can let it work.
But an agent that waits for me is wasting time. Its own time, and mine. Every minute it waits, the whole plan slows down.
So "Needs You" is the only column I really watch. It is my inbox. If it is empty, I can think about other things. If it has a card, I go there first.
This changed how I feel about running agents in parallel. Before, parallel meant chaos. Now it means a short list of moments where I matter.
One card, one place to work
The other thing a board gives me is boundaries.
Each card has its own space. Its own branch, its own copy of the code. Agents do not step on each other. One cannot break what another just built.
Without that, parallel work turns into a fight over the same files. With it, each agent has a desk, and I am the one walking between desks.
It is boring structure. Boring structure is exactly what makes the exciting part work.
The manager is still me
There is a temptation to add another agent on top. A coordinator agent that watches the others and decides things.
Sometimes that is useful. It can split a big task into small ones. It can hand them out.
But someone still has to say yes to the risky parts. Someone has to decide what "good enough" means today. A board does not remove that job. It just makes it visible.
I like that. I do not want agents that pretend I am not needed. I want agents that tell me clearly when I am.
Smarter is nice, clearer is better
I will keep trying new models. They do get better, and that matters.
But when I look at what actually saves my time, it is rarely raw intelligence. It is knowing where everything is. Knowing what is stuck. Knowing what is waiting for me.
A smarter model makes each card a bit better. A board makes the whole day calmer.
If I had to pick one, I would pick the board.
Anyway. Time to check that "Needs You" column.
