This 5-step system turns Claude into your hardest-working employee.

Most people prompt Claude. Brad gave it a job description.

That is the whole difference, and it explains why two people with identical subscriptions get results that are not remotely comparable. One is asking questions. The other has onboarded a member of staff — with context, a department, a documented process and a defined scope of work.

Here is exactly how each step works, and why the order matters more than any individual piece.

The short version

• Context = the brain. Departments = the org chart.
• Workflow maps = the SOP. Skills = the employee who never clocks out.
• Three hours of competitor research becomes hitting run and getting it back in minutes.

1. Identify revenue-generating tasks

Before anything is automated, you need to know what must not be.

List your top one to three activities that directly make you money — calls, content, whatever it is for you. These are the things where your personal involvement is the product, and delegating them is how businesses quietly get worse.

Then look at everything else. Everything that is time-consuming but non-revenue — that is what goes to Claude.

The reason this is step one rather than an afterthought: most people start automating whatever annoys them most, which is not the same list. The annoying task and the low-value task overlap, but not completely, and automating a revenue task because it is tedious is a real way to lose money efficiently.

2. Give it real context

A new employee does not start with a task. They start with an induction.

Write one document covering identity, revenue model, offer stack, team, market position, and 90-day priorities. Not a paragraph — the actual working picture of the business, the kind you would give a competent hire on day one so they could make decisions without asking.

Crucially, this becomes Claude’s single source of truth, not a one-off prompt. The difference is that a prompt is discarded after one conversation, while a source of truth is present in every conversation, forever, informing outputs you did not think to connect to it.

Almost every complaint about generic AI output traces back to this step being skipped. The model is not being generic; it is being given nothing specific.

3. Organize by department

Marketing, sales, finance — each one gets its own folder in Claude’s workspace.

Every workflow and output for that department lives in one place. That is the whole rule, and it does two things at once.

First, retrieval: the model knows where to look, so it stops searching and starts fetching. Second, and less obvious, it forces you to decide which department a piece of work belongs to — which is a clarifying exercise in itself, because work that fits nowhere is usually work nobody actually needs.

The org-chart framing is not decoration. Businesses organise by department because it works, and an AI workspace is subject to the same forces.

4. Map the workflow like an SOP

This is the step that does the real work, and it is the one that feels like a detour.

Pick one painful, non-revenue task and map every step from start to finish — like writing an SOP for a new employee. Not a summary. Every click, every decision, and examples of what success looks like versus what failure looks like.

The examples are the part people leave out and the part that matters most. “Produce a competitor brief” is not a specification. A good brief and a bad brief, side by side, with the difference articulated, is.

There is a side effect worth mentioning: writing the SOP frequently reveals that the process itself is bad. You cannot document a workflow honestly without noticing the steps that exist for no reason. Fix those first, then automate what remains.