Next
In progress
Loops, Attention, and the Agent Operating System
How recurring work gained triggers, evidence, escalation, and a way to decide what should run next.
A field study in AI-assisted work
One useful prompt opened a door.
Then every new possibility asked for a little more attention.
Act 02 / Multiplication
The output got faster.
The work did not get smaller.
Act 03 / Open loops
Every possibility became another unfinished decision.
Open loops01
Act 04 / Selection
Attention decides what runs next.
Not more output. A way to choose, preserve context, and continue.
Act 05 / Durable work
The loop closes when the learning can outlive the prompt.
Field notes from the work
Writing through the work.
Notes from building software, working with agents, and learning where engineering judgment matters most.
Read the latest field note More writing on DEVCurrent thread3-part series
01
Published
8 min
02
AI Agents Made Me Faster. Then Attention Became the Bottleneck.
Useful agents accelerated the work. They also exposed the operating layer I was still carrying myself.
Published
7 min
When Prompts Become Infrastructure
How I stopped rebuilding the same setup in every agent session by moving the development process, reusable procedures, and project memory into the repository.