I work on the part after the demo.

I build AI systems at FieldPulse, a field service management platform in Dallas. Four of them are in production: an agent drafting support email, one handling payment notices, an assistant answering product questions in Slack, and the pipeline behind our customer status pages.

Getting a model to produce something impressive is a demo. Knowing what it does on the day retrieval comes back empty, or the provider changes a model underneath you, or the output is about to reach a customer who will act on it — that is the engineering, and it is most of the job.

I came into this through marketing and automation work, building n8n workflows before I was building agents. The through-line is the same: work out what a process actually does, then make the repetitive part run itself without anyone having to trust it blindly.