Build notes from SaaS, agents, automation,
simulations, and the decisions that changed
the outcome.
Selected work · An open laboratory
Products, experiments, and decisions—with the evidence attached.
Can reusable prompts become a workflow, rather than a folder of text?
An open question.
A product in progress.
Illustrative workflow study
The question: what should happen between saving a prompt and using it again?
This sketch explores capture, refinement, and reuse. It is a design study—not a shipped-feature claim or a measured result.
Does a second opinion improve the decision—or just make it more expensive?
Two perspectives.
One decision to examine.
Illustrative agent study
Compare a single response with two independent responses and a review step.
A real experiment would track correctness, latency, cost, and the cases where review makes an answer worse. No benchmark results are claimed here.
What changes when cooperating becomes the rational move?
A small model.
A different way to ask why.
Illustrative simulation study
Start with two players choosing whether to cooperate or compete. Change the incentive, then compare their choices.
The model is a thinking tool, not a prediction of real people. Assumptions and limits belong beside any eventual results.
Every subject is another way
of asking how systems behave.
What changes when mathematics makes cooperation rational?
Follow this questionA useful answer should
survive its assumptions.
The decision before the outcome is often the useful part.
Sharing the branches, trade-offs, and failed assumptions makes the work more useful than a polished victory lap.A good prompt makes the decision criteria visible.
Useful prompts clarify the goal, context, constraints, and evidence needed—then leave room for the model to do the work.Automate after the judgment becomes repeatable.
A clear trigger and a reviewable output matter more than removing every manual step. Some friction protects the decision.Evidence begins where the demonstration can fail.
The setup, comparison, result, and limitations should be visible enough for someone else to challenge the conclusion.Shorter feedback loops beat fuller calendars.
Reduce the distance between building, observing, and deciding. Movement without feedback is only activity.Progress is a useful uncertainty removed.
A shipped feature can teach nothing; a small test can change the entire plan. Measure what the work helps you decide next.Short, honest notes on what I am building, testing, breaking, and learning—sent only when there is something worth sharing.
Not ready to subscribe?Read the latest field note