Agent Engineering Notes Substack
Live Last read · last published · next check
One is a persistent, self-hosted runtime across channels. One is an agentic coding tool with deep repo execution. They overlap less than people think, and that is exactly why this comparison matters.
Two open, self-hosted agent runtimes. Same problem class, different center of gravity — and the choice is mostly about what you want the system to compound on over time.
Most agent failures are memory architecture failures wearing a prompt-quality mask.
Every option looks like a loop with tools until you define what has to be true for the run to have succeeded.
Every vague word in your system prompt is a decision your agent makes without you.
Five design decisions that look obvious in hindsight.
Roles, config, delegation chain — and the three things I got wrong before it worked.
Here's the difference — and five real patterns worth stealing
The newsletter about AI agents I kept writing by accident — now on purpose.
Everyone is excited about AI agents. Very few have actually tried running a team of them. Here is what happens when you do.
