Automation is about attention
I build AI tools for a simple reason: repetitive work consumes attention. When a system can reliably handle the repetitive parts, people can focus on judgment and creativity. AI is one tool in that automation toolkit.
Reliability beats novelty
Novel AI features are easy to build and easy to break. I prefer a different tradeoff: build a deterministic core, then add AI where it improves interpretation or reduces writing friction. The result is usually less flashy and more usable.
Make the system reviewable
Trust comes from reviewability. Call logs, summaries, and visible field mappings are not "nice to have" details. They are what make automation usable in practice.
Projects that reflect this
- Exempliphai, automate repetitive web forms.
- AI voice agent automation, voice workflows with logs and configuration.