From Claude Cowork to automation

Let your team build their own automations, then promote the ones that matter

Individual routines first, because that is where the real work shows itself. Then the ones that touch a system of record, or that more than one person needs, become automations the business runs: scheduled, logged, and waiting for a human before anything leaves the building.

How work gets promoted

Start where the work is

The best automations begin as one person’s morning routine. Your team builds these themselves once they have Claude Cowork.

The promotion test

More than one person wants it, it touches a system of record, or it writes to the outside world. Any yes and it gets built properly.

Simplify, then build once

Ten steps become three before a machine touches it. One loop: fetch, derive, file, log, draft for a human.

Runs in your accounts

Your GitHub organisation, your cloud project, your secret store, your database recording every run.

Alongside the manual process

In parallel until you trust it. Measured on hours saved, when it is ready, and whether the meeting uses it.

Back to the team

Once a verb exists, your leaders call it from Claude: “another view of the sales report”. New views are two-hour jobs.

Tell me what you're dealing with

Three lines is plenty. I reply personally, usually within a day, and the first call is free.

Frequently asked questions

Why not automate the obvious processes straight away?

Because a process nobody has run by hand with AI first is a process nobody understands yet. The routine stage is cheap discovery: the real steps, the exceptions, and who cares.

What is the difference between a routine and an automation?

A routine runs on one person’s machine, under their login, and stops at systems without a connector. An automation runs in company accounts, on a central schedule, through APIs, with every step logged and every outward action approved. A habit versus infrastructure.

We use ChatGPT Work, not Claude Cowork. Does this apply?

Yes. Claude Cowork and ChatGPT Work do the same kind of job: an assistant that works through files and connected apps and finishes a task rather than answering a question. The promotion path is identical. A routine that more than one person needs, that touches a system of record or that writes to the outside world gets rebuilt as an automation the business owns.

Which systems can you connect?

Anything with an API or a scheduled export: ERP, point of sale, e-commerce, accounting, CRM, help desk, spreadsheets. Where the API is slow to arrive I build on exports and swap the API in later.

Who supports it afterwards?

Two weeks of bug fixing is included. After that a support arrangement sized to what is live, or your own people, because it is documented and in your repository.

What does a project cost?

A fixed price per project, quoted after the audit or a short call, with the measure agreed up front. Days, not months.

Got a routine that has outgrown one desk?

Three lines: what it reads, what it produces, who waits for it. I will tell you whether it is ready to be promoted.

Back to the form