Filed under
3 blog posts and projects filed under this tag, newest first.
Entries
3
The next step for LLM agents is an operating environment with durable state, checkable outcomes, isolation, failure semantics, and human gates at the points that cannot be undone. A better prompt does not supply any of that.
Read dispatchMy second homelab replaced a Raspberry Pi with a loud, power-hungry Dell R710. It taught me where capacity helped, where it hurt, and what the failures changed.
Read dispatchI built Refund Hero as a full-stack prototype for identifying eligible Deutsche Bahn delay refunds and preparing reimbursement requests. I stopped it when DB announced a parallel service.