TELL THE TRUTH
Make the problem visible.
AI CHANGES THE
SCALE OF AGENCY.
One person can now explore, design, implement, test, and coordinate work that once needed a much larger team.
THE QUESTION What happens when we give that agency a real engineering problem?
Bridge — about 1 minute
AI changes the scale of agency. A person with a clear goal can now run several lines of inquiry or implementation at once.
But more capability also creates more opportunities for drift, duplication, and failure. So I wanted to test a structured version of that agency.
CURRENT AI WIN
THE EXPERIMENT.
Hermes agents claimed bounded hbsguard tickets across a shared Kanban workflow.
The experiment — about 1 minute
hbsguard is a standalone Handlebars linter I built. I used Hermes to place a set of focused engineering tickets on a shared Kanban board.
Agents could claim ready work, move it through running and review, and expose state to one another. The goal was to learn where coordination genuinely improves focused engineering work.
THE SWARM AT WORK.
Workers begin in parallel. The dashboard and terminal make claims, handoffs, heartbeats, and status changes visible.
← → BROWSE IMAGES
Workflow — about 1 minute
The useful part was not simply having several agents. It was making their work legible.
I could see who claimed what, whether a process was alive, when work changed state, and where review was needed. Parallel work is only useful when coordination does not become invisible.
USEFUL FAILURETHE SYSTEM
THE SYSTEM
HIT A LIMIT.
More agents did not mean more progress. My laptop became the bottleneck.
THE LESSON The failure made an invisible constraint visible.
The failure — about 1 minute
Then my laptop became nearly unresponsive. More workers were not producing more throughput; they were competing for a fixed amount of memory and CPU.
This was a useful failure. The constraint moved from human typing speed to orchestration and system resources. The experiment made that visible.