AI Training.

Trained on the work, not on slides.

A team that has sat through an AI demo and a team that has built three prompts for their actual work are at very different points. The session is built so that the second team walks out of the room: prompts that work on real inputs, a workspace they will open on Monday morning, and a clearer sense of when not to use AI than they came in with.

How a training is shaped

Every session starts with a short skills assessment. Where the team is, what they have tried, what is already working, what has been frustrating. Sessions for a team that has been using Copilot for a year look different from sessions for a team that has been told they should be using AI but has not been told what for. We adjust the format and the examples before the session, not during it.

The session itself works through real examples your team faces. Real client emails, real source documents, real classification problems, real drafting jobs. We build prompts on those examples in the room, watch the output, iterate, and end with a prompt that the person who brought the example can run themselves the next time the situation arises.

The format and length are scoped to your team size and goals. Half-day sessions work for a small team focused on one or two use cases. Full-day or two-day sessions work for a broader team covering more of the operation. The principle is the same in either: hands on keyboards, real examples, working output by the end of the session.

The workspace and the prompt library

The custom prompt library is the deliverable your team will use after we go. Built around the actual work, organised by who does what, written in plain language so it can be edited, extended, and shared with new joiners. Each prompt comes with notes on when to use it, what to expect from the output, and what the team should still check before sending anything on. The library lives in a workspace your team owns: a shared document store, a prompt-management tool, or wherever fits the rest of your stack.

Reference materials cover the principles behind the prompts, not just the prompts themselves. The reason a prompt works the way it does, the structure that makes a new prompt easier to write, and the limits of what the model can do reliably. Teams take this back to colleagues who were not in the session and bring them up to a useful level without needing us in the room.

Judgement, not just prompts

The harder skill, and the one most AI training does not cover, is knowing when not to use AI. A team that confidently delegates the wrong tasks to an AI ends up generating more downstream work than it saves. We spend time on where AI helps (drafting, summarising, classifying, extracting, brainstorming, structured reformatting) and where it does not yet (high-stakes judgement, work with no source material, work where every detail must be right and there is no human review).

Most AI tools are wrappers over the same models a team can already access. The skill the training builds is the one no wrapper supplies: knowing what to ask for, what to push back on, and what to keep human. We train on whichever AI tools your team already uses or has access to. Claude, ChatGPT, Copilot, Gemini, Perplexity, vendor-AI features inside other software, whichever fits the work in front of you. The principles transfer across all of them, so the training stays useful as new tools appear and old ones get retired.

The training pairs well with an AI Integration build, where the prompt habits the team has built get embedded into automated workflows with Judges around them, or with the AI and Automation Audit, which often surfaces training as one of the recommendations. We are independent AI and automation specialists. The training is on the tools that fit, not the tools we are paid to push.