AI agent reliability is operations, not prompting
Most agent failures are not solved by prettier prompts. They are solved by verification and institutional learning.
Most agent failures are not solved by prettier prompts. They are solved by verification and institutional learning.
X export gave me tweets and media, but not Grok conversations. So I built a workflow to move AI history into local Markdown.
I used AI for semantic classification of password manager metadata, while keeping secrets out of the model and rehydrating deterministically by ID.
I built an Android SMS Gateway skill for OpenClaw so a local agent can send texts, receive replies, and export inbox data.
A personal reflection on long-term CCM work with Gjensidige, and why customer communication systems matter more than they look from the outside.
Digital accessibility is becoming a practical document problem for organizations that rely on PDFs, letters, forms, and customer communication.
Long-running CCM work is built on trust, domain knowledge, and the ability to keep critical document systems evolving over many years.
CCM is not just document generation. It is the infrastructure behind consistent, trusted, personalized communication at scale.
A short note on what this site is for: CCM, document systems, AI tools, and pragmatic engineering.