Notes on automation systems, n8n workflows, AI agents, integrations, content pipelines, and the logic behind making messy processes work reliably.
May 22, 2026
After testing multiple AI tools for invoice processing, expense categorization, and anomaly detection — the conclusion is always the same: it's the combination, not the model.
May 21, 2026
At 5-10 minutes per invoice, 300 invoices a month is 25-50 hours of manual entry. The automation exists. The part most people skip is building the GL mapping table that makes it work.
May 20, 2026
Switching from e-Boekhouden to Moneybird isn't just a UX preference. It's a choice about what you'll be able to automate later — and what will stay manual forever.
May 19, 2026
A bookkeeper's advice for Dutch starters: set up your own administration, use the bank connection, and bring in a professional for a quarterly review. Automation reduces frequency, not the need for expertise.
May 18, 2026
A student built an n8n invoice controller with a cheap model as the first filter. That filter — the 'bouncer' — is actually the most valuable node in the workflow.
May 17, 2026
Most automation fails not because of bad code, but because the builder started too early. The discovery work — understanding where things actually break — is the hard part.
May 16, 2026
An accountant who automated his way through PwC, Python, and ChatGPT — and the question it left: why did I spend years studying to end up doing this?
May 15, 2026
The most useful accounting automation isn't AI or agents. It's removing the data-moving, file-copying, and format-cleaning so accountants can do actual accounting work.
May 14, 2026
Credit card transactions don't auto-import into Moneybird. Not because nobody built it — but because the card provider controls the data and has no incentive to share it.
May 13, 2026
The question isn't whether to use bank feeds. It's whether the person running them knows enough to catch what the automation misses.