FAQ

A skill is a set of instructions and reference material that teaches an AI model to do one job the way an expert does it. It comes with a method for running it, defined below. Ours run as an interactive interview: the model asks, researches, drafts, and challenges its own output, and you steer.

Why it matters: the model stops being a blank chat window that needs to be re-briefed every time. The expertise is in the skill, so the tenth run is as good as the first, whoever on your team is running it.

Yes. We build them on Agent Skills, an open standard developed by Anthropic and supported by tools from OpenAI, GitHub, Google, Mistral and others. A skill is a folder of instructions, and optionally scripts and reference files, in the same format whichever tool runs it.

Why it matters: you are not locked in to one provider. A skill you run on Claude today can move to another compatible tool later. How well it performs still depends on the model, so test it again after a switch.

The expertise inside it. A skill is only as good as the know-how written into it and the points where it stops for a person to decide. When we build a skill with you, that know-how comes from your own expert: we interview them on real work, AI drafts the skill, they check it, and we test it on your live work.

Why it matters: the format is free, and anyone can write a skill in it. What you pay for is whose expertise it carries, and how well it marks the calls your team has to make.

Building skills the AI runs. The phrase reads both ways, so it is worth being exact: a skill here is an artefact, a file your team owns, that tells a model how to do a job your experts already do well. Human method in, machine-runnable skill out.

Why it matters: the capability stops depending on who is at the keyboard. Personal fluency with AI helps the person who has it. A skill is the method itself, captured once and run by anyone on the team.

The method is how your team runs the skill and improves it: where to set the intent, where to challenge the output, who owns the decision, and how each run feeds back into the next. Ours is the Third Vector Method. At its centre is the Third Vector Judgement Framework, the six calls that stay with people, from setting the intent to the gut check, and on the Platform the Judgement Layer puts it into practice. The best outcomes come when human judgement is part of the flow of the work, and the method is what puts it there. AI output is a draft to interrogate, and the method makes interrogating it a habit.

Why it matters: your team's judgement is drawn out on every run, so its AI fluency grows with each one. That can mean working differently, and that is deliberate. It is why every skill we build comes with a method.

Yes, and it is the core of what Applied AI does. We sit next to the people who hold the method, take apart how they actually work, and write it into a skill an AI model can execute end to end. You get the skill, tested against your live work, and the handover, so your team can run it and change it without us.

Why it matters: the method stops living with a few individuals and becomes something the whole team runs. Your best people are freed for the work only they can do.

We build every skill for the company that runs it. Some follow a pattern we have built before, such as company writing, job postings and proposals. We start those from the pattern and tailor them to how your company works, as part of Applied AI.

Why it matters: a skill only earns its keep when it carries your rules, your examples and your way of deciding. Starting from a proven pattern takes less of your experts' time, and the result is still yours.

It is where a skill goes once your team relies on it. The Platform runs the same skill on several models and compares them on a scorecard per model, so you can pick the one that does the job best. It keeps a gold set for each skill, the cases it has to get right with answers your experts approve, and tests every new version and every new model against it. And it sends your team's notes on each output back to the skill, to shape its next version. It is in early access, by invitation, with the teams we build skills for.

Why it matters: models change every few months, and a skill that worked last quarter can drift without anyone noticing. The Platform tells you which model to run it on, and catches a drop before your team relies on it.

Applied AI is bespoke: we build custom skills and integrations, and scope the work to your operation. It is quoted after a call. Academy is productised: a set programme at a fixed price, bought online, where your team learns to build for themselves, live on your own stack. Build & Run, the programme on sale today, covers standing up an AI-native system, connecting your tools, and working across them day to day. Writing skills is not part of it.

How to choose: by who does the building. Applied AI when you want it built into your operation with you. Academy when you want your team able to build in-house from here on.

The same idea as above, in Anthropic's own packaging: a folder of instructions, and optionally scripts and reference files, that Claude loads when the task calls for it. The skills we build ship as Claude skills today, and the format is simple enough that your team can write its own.

Why it matters: a way of working becomes a file you can version, review, and hand to a colleague. Knowledge that used to live in one person's habits becomes something the whole team runs.

MCP, the Model Context Protocol, is an open standard for connecting an AI model to your systems: your CRM, your file store, your ticket tracker, your database. One integration per system, and any MCP-speaking model can then use it.

Why it matters: the model stops working from whatever you paste into it and starts working from your actual data. That is the difference between asking for a generic pricing framework and asking about pricing given the last two years of your own deals.

Claude Code is Anthropic's agentic tool for real work in a real environment: it reads and edits files, runs commands, and works through multi-step tasks rather than answering in a chat box. Engineers use it in the terminal and in their editor. It is not only for engineers, and this website is built with it.

Why it matters: the model moves from advising on the work to doing the work, with you reviewing. That is a different relationship with the tool, and it takes some getting used to.

No. The skills we build run on Claude today and are portable by design, so the same skill can run against another model, or through a proxy or gateway you already use, or against a model you host yourself. The Platform runs one skill on several models side by side, so you can see which does the job best before you choose. On Applied AI engagements and in Academy programmes we go deepest on Claude and also deploy Mistral, OpenAI, or open models where they fit your data, budget, or sovereignty needs.

Why it matters: you are not holding a capability that only works in one place. Models come and go faster than capabilities do.