The Thirteenth Loaf
What gets harder when intelligence gets easier? An operator’s investigation of coordination, shared mistakes, and trust.
Read the essay on Substack ↗What gets harder when intelligence gets easier?
b.akery is working to reduce the effort of getting useful, dependable work from AI agents. I compare changes to how work is directed and reviewed, then ask what improved, what still failed, and how much extra work the change required.
For founders and operators already using AI agents.
What done looks like.
You can use the result for the job you asked it to do. You know what was checked, what remains uncertain, and what still needs your attention.
That is the standard b.akery is working toward. I start with a specific failure, compare a possible change with the usual approach, and check whether the result still does the job. The time spent briefing, reviewing, and repairing belongs in the comparison too.
An extra check has to earn the time it takes.
The Thirteenth Loaf
All essays →
Essays, field notes, and open questions. Read here; subscribe and join the conversation on Substack.
Three questions behind the work.
-
Coordination Tax on Intelligence
What happens when intelligence becomes abundant? Count the briefing, handoffs, review, and repair alongside the useful output. Where does the effort go, and what do we get back for it?
-
Confabulation
Why do humans and agents fail together, honestly? Trace a shared mistake: what the person assumed, what the agent added, and where either could have interrupted it.
-
Trust & AI
Which problem are we trying to solve when we ask for trust? Try a specific check or change. What did it catch, what did it miss, and was it worth the effort?
I’m James.
I work with the agents, too.
I don’t read or write code. I still build software with AI agents, and I’m responsible for what the work produces.
I’ve confidently described a solution I didn’t understand, and watched an agent build it well. Nothing pushed back, so nothing looked broken.
That is part of why I’m building b.akery. The writing, research, and external tests will help decide which findings should become tools or practices people can use.
Bring a real piece of your work.
Think of a recent task where you had to repeat instructions, redo work, or check a result you weren’t sure about. What happened, and what did you already try?
I’m looking for founders and operators willing to discuss those experiences and try a relevant experiment when one is ready. Before you take part, we’ll explain the task, the time involved, and what we’re trying to learn.
Register testing interest