How we tidied our own projects into one system

01 · Problem

The information exists, just not in one place

A spreadsheet, a folder, messages, a document with notes. Every piece of information is somewhere, but the answer to a simple question has to be assembled. And if only one person can assemble it, the business rests on that person's memory.

02 · What we observed

What it looked like for us

• The state of projects was spread across repositories, shared folders, documents and messages.
• We answered the question of what the next step on a project was from memory.
• While preparing this part of the website we found that research from August 2026 existed only as unsaved files in one working copy. One clean-up and it would have been gone.
• Older documents contradicted each other and it was not clear which of them applied.

03 · Method

The record first, then the software

We did not start by programming. We started by agreeing what one project record should contain: who the client is, what state the project is in, where its sources are, what the next step is and whose turn it is.

Only once that record was clear did we build a system around it. We called it Command Center.

• one record per project
• one place that applies when sources contradict each other
• a backup before every change to the register
• if we do not know something, the record says “unknown”, not an estimate

04 · Tool

Where AI helps and where it does not

Our AI assistants work on top of these records: they summarise the state, prepare material, find contradictions between documents. They get through a volume of text quickly that a person would spend hours reading.

Their limit is the same as a new colleague's: they know only what is written down. If the record is inaccurate, the result is convincing and wrong. That is why order in the records mattered more than the tool itself.

05 · Human verification

Who decides

Changes that cannot easily be undone are approved by a person: publishing, deployment, a change to the register, anything a client will see. The assistant prepares a proposal and shows what it was based on. The decision stays with us.

Important conclusions are also checked by a second, independent view that did not see the first one's process and whose job is to look for errors.

06 · What changes

What is different today

• One place answers the question “where does this stand and what is the next step”.
• When documents contradict each other, it is defined which source applies.
• The August research is stored in a versioned repository and marked as input, not as a decision.
• A new project starts with creating a record, not with searching for where things are.

07 · What we learned

What we would do differently next time

• The most value came from agreeing what the record should contain. The software came only after that.
• An unclear process should not be automated. Automation only makes it faster, errors included.
• “Unknown” is a more useful piece of data than a good-sounding estimate.
• Start with one type of record, not with the whole business at once.

How much time this saved us we did not measure, so we give no number.

08 · Related service

If you recognise yourself in this

We use the same approach with clients: first find out how the work really happens, then build the smallest system that is enough. More is under business systems and automation.