Watch and check
I explain the choices behind useful AI assistants and automation: where the work runs, how to check an answer and what a usable result costs. Each video has a linked article for the detail.
The videos use my authorised AI likeness and cloned voice. Demonstrations use fictional material and state their limits. Selectable English captions are available on YouTube.
Video ·
ChatGPT Enterprise is private by contract: OpenAI says business data is not used to train its models by default, and the rest runs inside OpenAI. A private AI stack keeps the chat interface, model access, your documents and the logs in a workspace you own, so you can see where every step runs.
Short ·
ChatGPT Enterprise is private by contract: OpenAI says business data is not used to train its models by default, and the rest runs inside OpenAI. A private AI stack keeps the chat interface, model access, your documents and the logs in a workspace you own, so you can see where every step runs.
Video ·
An automation engineer is the person who embeds inside a company to turn an AI tool into a working part of a real process, not a demo. The role needs technical skill, an understanding of what AI can actually do, and an understanding of the workflow it has to fit into.
Video ·
To judge whether AI reduces your costs, divide the full cost of the agreed test by the number of results that pass your acceptance checks. Time released is capacity; it becomes cash saved only when spending actually falls.
Video ·
An internal knowledge assistant is usable only when you have agreed what a good answer must contain, which sources it may use and when a person takes over. This video looks at when an assistant could help with recurring questions, and when better documentation is the right first step.
Video ·
The model should be chosen by the work, not by you: a small classifier sizes each request, cheap models do the research, and the expensive models are reserved for the hard work. I tested this on my own laptop, and this video shows the setup, what broke on the first run and what a business gets from the same shape.
Video ·
I build AI assistants and automations for recurring business work. I scope the work in writing, agree the acceptance checks, build it and hand it over with a runbook.
Video ·
In my three tests on fictional documents, the assistant gave one supported answer, one correct conclusion with citation links attached to the opposite statements, and one honest answer that the information was not in the documents. Three tests do not establish a reliability rate, so check the sources and measure cost per acceptable result before you build.