Train the Agent, Not the Model
The short version Stop building agent swarms on top of agents that still need a human. Make one agent reliable first. You can do that by training it, not the model, but the agent itself: its prompts,
#ai
Posts tagged #ai
Microsoft Just Shipped Conductor. I Built a Side Project With the Same Idea. Here's the Gap.
Third in a series. Previously: the vision post and the infrastructure post, both from March 2026. The parallel In March I published two posts about DAGent — a deterministic agentic pipeline that rout
My AI Pipeline Burned 50 Minutes on Correct Code. The Infrastructure Just Didn't Exist Yet.
TL;DR: My deterministic AI pipeline burned 50 minutes retrying correct code because infrastructure hadn't been provisioned yet. I restructured from a 12-item single-wave DAG to an 18-item two-wave arc
I Built a Deterministic Agentic Coding Pipeline. Then I Found Out Stripe Built the Same Thing.
TL;DR: I built a headless, DAG-scheduled AI coding pipeline — 12 specialist agents, self-healing recovery, real browser testing — that takes a feature spec and delivers a tested PR with zero human int