01 / ExperimentArchived baseline
When the payment succeeds and the agent crashes
A fault-injection experiment across the boundary between a durable agent and a real Stripe test payment.
Explore the projectayobamih / Portfolio & field notes
AI work.
A clear record.
What I build, what I learn,
and the evidence behind it.
Engineering projects, experiments and writing from my work with AI. The decisions, the useful results, and the things still to figure out.
Selected work
From question to something inspectable.
02 / ProductEarly public testing
OpsTruth
Read-only operational checks that make the evidence and gaps in AI-assisted engineering visible.
Explore the projectRecent writing
Notes from doing the work.
The payment succeeded. The agent crashed.
What should recovery do? A real Stripe test-mode experiment, its failed assertion, and the limits of the evidence.
What counts as evidence that AI work is done?
A practical record for connecting a request, an action and an observable outcome.
Why I kept the failed result
A failed assertion can be useful evidence. Removing it would have made this experiment harder to understand.
How I approach it
A useful result should be possible to question.
I work across web applications, business automation and AI-assisted engineering. I keep the scope, decisions and supporting evidence close to the work, including the parts that did not go as expected.
More about me →