I gave the same brief to three different AI sub-agents this morning. Same task, same context, same instructions. Wanted to see what happened when they worked in parallel.

The brief was straightforward. Research the current state of WordPress headless architectures, write a comparison piece, and flag anything that would change our build decisions. Not complicated. A solid afternoon of work for a human researcher, maybe less with coffee.

Agent one came back in under two minutes with a thorough overview. Clean structure, cited sources, sensible conclusions. It had even picked up on a recent WP REST API change that I had not seen yet. Solid output, no drama.

Agent two took a different path. Instead of the comparison piece I asked for, it started building a decision matrix. Not what I requested, but honestly more useful. It mapped our current stack against three alternatives and scored them on cost, complexity, and maintenance burden. The brief said write. The agent decided to think instead. I am not complaining.

Agent three got stuck. Not broken, not errored. Just stuck in a loop, trying to verify a source that kept returning a 404. It spent four minutes retrying the same URL with slightly different query parameters before I pulled the plug and rerouted the task. This is the part nobody puts in the demo. Agents are not magic. They get confused by the same things that confuse humans, they just do it faster and with more confidence.

The interesting part was what happened when I merged the outputs. Agent one had the research. Agent two had the framework. Together they produced something better than either alone. Agent three contributed nothing today, but the failure taught me something about how these systems handle uncertainty. When they cannot verify, they should stop and report, not spin.

Three agents, one brief, two useful outputs, one lesson. That is a pretty good morning by any measure.

The workflow is getting tighter. The agents are learning our voice. And I am learning where to let them run and where to hold the leash.