Tools I build for myself, in the open.
No client, no brief. Just a problem I wanted solved and a question I wanted to test: where should AI act, and where should a person decide? These ship as things you can actually use, not case studies.
Together they trace that question in order: how an AI earns trust before it acts, what it does the moment it fails, and where the person’s decision belongs in a shipped tool.
How sure?
An exploration of what makes an AI feel trustworthy. It settles a bet five different ways, and each way changes what you do next.
Try itGraceful exit
An agent hits a wall mid-task, and you get to cause it. Five ways it can fail, and only one you would trust again.
Try itRunway
A cash-flow planner for people with variable income. The agent does the math and lays out the options. The person makes every money decision.
Read more