AI-assisted workflows

Freelancers (zzp’ers) do their own bookkeeping, taxes, planning and customer contact next to the work itself. For our own freelance practice we built an end-to-end AI-assisted workflow: a team of specialised Grok bots (xAI’s AI model), each connected to one system: Moneybird, the mailbox, the calendar.

You talk to the bot team. It puts the appointment in the calendar and confirms it to the customer. At the end of the workday you say the finish time; it counts the hours and kilometres and drafts the invoice in the same format as the last one. You say “approved, send”, and the invoice leaves from your own bookkeeping system. Nothing irreversible leaves without you.

We set up the same workflow for other zzp’ers, in the tools they already use. Grok bots by default, something else where that fits better. Read how the workflow for zzp’ers works →

This website is the example. It is built, maintained and deployed in one AI-assisted flow with Claude Code, Anthropic’s coding assistant that works inside the project itself: it reads the files, makes the changes, runs the checks and explains what it did.

• All text lives in plain files per language, and every build checks that nothing is missing.
• The house style is kept in one place, so the look can change without rebuilding the pages.
• Publishing, the domain and the search engine files are part of the same flow.

A person decides what goes on the site and reviews every change. The assistant does the building.

Before we invest in an app or a flow, or pay someone to build it, we build a working test version with Grok Build, xAI’s coding agent for the command line. It plans the work, searches the code and builds, with several agents working on a task in parallel.

If the test version proves its worth, it gets built properly. If it doesn’t, it cost hours instead of a budget.

1. One task first. Repeated, well-defined, and worth the time it saves.
2. One connection per bot. Each bot gets access to the system it needs, and nothing more.
3. Agreed checks. What the assistant may do on its own, and what a person approves first.
4. Data agreements. Which information may go in, and which model and provider handle it.
5. Handover. The instructions in plain language, so you can adjust them yourself.

• The same tasks, emails or documents come back every week.
• The knowledge exists, but is spread over systems, inboxes and people’s heads.
• An idea needs testing before it is worth a budget.

• Decisions that must be right every time without a check.
• Sensitive data that may not leave your own systems under the provider’s terms.
• A process that is not clear yet: an assistant only speeds up what is already there.