Practical writing on AI systems, agents, automation, and lessons from building.
AI marketing automation pays on the work before publishing: research from real conversations, first drafts, repurposing across channels, and reporting. It fails when the model publishes. Here is the design for a lean team.
An AI receptionist for a small business handles first contact: answers the common questions, books what it's allowed to book, takes a message for the rest — and makes clear what's actually free and how to judge the fit before voice gets involved.
AI sales automation pays on the steps between lead and call: research, qualification, first-reply drafting, follow-up and CRM hygiene. It loses on the steps that carry the relationship. Here is the line and the build.
A technical look at what AI SEO and GEO really require from a website: crawler access, server-rendered text, and structured data and why an llms.txt file isn't one of them.
Most businesses have more AI subscriptions than AI workflows. A stack audit lists every tool, ties each to a measured workflow or to nothing, and decides. Here is the method, the sheet, and what to do with the result.
An AI sales agent that sends on its own damages the domain and the brand. One that researches, qualifies, drafts and books, and stops before anything reaches a prospect, pays. Here is the design and the pause list.
An AI research agent gathers what is known about a company, a contact or a question, with sources, before a person needs it. The design problems are cost, confidence and knowing when not to run. Here is the shape.
AI web design produces layouts, variants and first drafts fast. It produces generic pages just as fast. The difference is the brief. Here is what to hand the model, what to keep for a designer, and how to check the output.
An AI website audit can crawl every page, flag the mechanical failures and summarise them in minutes. It cannot tell you whether the site does its job. Here is how to run one, what to trust, and what to decide by hand.
AI website copywriting is fast and it is usually flat. The model averages every site it has read. Here is how to feed it real claims, real proof and a written voice, and which pages to keep a person on.
