AI
AI that does a job, not a demo
Turn the questions customers ask every day, and the copy-and-paste your staff does every week, into a system a person supervises. I build narrowly scoped assistants and automations from your own approved information, with a human taking over when the system shouldn't answer. It runs on infrastructure you control, at a cost you can see line by line.
Two kinds of work
- An assistant that knows your business. It answers from your catalog, your documents, and your rules, not from the open internet. It links only to things that exist in your current catalog, declines what it shouldn't answer, and emails you the conversation when a question needs you. It doesn't name competitors.
- Automations that remove a chore. New writing that has to be copied from one platform to another. Events that live in three places. A report someone rebuilds by hand every month. If someone on your staff does the same tedious thing every week, it's a candidate.
Where it's running
Live on wisewomenbotanicals.com as a soft launch. Every conversation is logged and reviewed, so the owner can see what visitors actually ask, and mistakes get corrected instead of repeated.
AI product consultant · Sonoma County
Wise Women Botanicals
Before. A naturopath with decades of formulations and 475 products in her store, and a website that could only answer the questions she had already written a page for. Every real question, “what do I use for this,” came to her by email or not at all. Off-the-shelf chatbots would have invented products, named competitors, and given health advice with no guardrails.
After. A private assistant, “Ask Connie,” that answers from her catalog, her own blends and protocols, and reference material on 80 oils and 60 herbs. It explains why before it recommends, links only to products in the current catalog, applies written cautions for pregnancy, children, and medications, and emails her the conversation when a question deserves a human. Runs on Cloudflare with Claude, at a cost she controls.
Live now as a soft launch, with visitors using it while the conversations shape the next version. It went live only after a compliance review of every product page, and every conversation is reviewed so mistakes get corrected rather than repeated. None of this was in the original brief: the engagement started as taking over a half-finished store from the previous developer and getting it live. The search data, the problem-first taxonomy, the Google Shopping feed, and the assistant were all proposed along the way.
Automations already in use
- Substack to website, every two hours. Four Worlds International Institute publishes on Substack. Their site picks up each new piece, builds it into the archive, checks every link, and deploys, with nobody logging in. The full case →
- The work behind this site's prices. PHP compatibility scanning, WordPress-to-static conversion, link and image checking. The tedious middle of each project is automated; the judgment isn't. See how that shows up in Convert →
What I won't build
An assistant that gives medical, legal, or financial advice without guardrails. An assistant that pretends to be a person. Anything that sends your customers' conversations to a vendor you haven't agreed to.
For sensitive work, approved sources, written safety rules, logging, and human handoff are requirements, not extras. The Wise Women Botanicals assistant went live only after a compliance review of every product page in the store. The goal is to constrain bad answers and catch the ones that get through, not to pretend they're impossible.
Pricing
From $2,500 for a first assistant or automation, scoped to one job. Then $50 a month for hosting and monitoring, plus the AI usage itself at cost, which you can see line by line. The first conversation is about whether it's worth doing at all.