Automation. The boring bit first.
Business automation starts with one repetitive step: data copied from Excel into a system, a report assembled by hand or the same check done every morning. We build internal tools and automations that have a clear owner and tell you when something goes wrong.
Is this you?
SMEs where Excel is the system
An important process lives in spreadsheets, email threads and one person’s memory.
Operations teams
The same manual work repeats every week, and mistakes are spotted late.
Companies considering an AI agent
You want to try whether an agent could handle part of the work — without it making decisions it shouldn’t.
Automating the whole process is too big a first step.
Many automation projects try to replace an entire process at once, and the exceptions eat the schedule. A better approach is to pick one bounded, repetitive and measurable step.
Unmonitored automation is dangerous: it can make the same mistake quickly and many times. So every run has a log, every failure an alert, and a human can always check.
Small automation that actually works
Describe the current state
Where the data comes from, what happens to it, who checks the result and what happens in an exception.
Scope the first step
One repetitive step with a clear input and output. We measure it before and after.
Build it to be watched
Logs, run history, error notifications and, where needed, human approval before an important action.
AI where it helps
Classifying text, extracting data from documents or drafting replies. Decisions with consequences stay with a person.
Senior plans. Agents build. Humans verify.
Agents run in parallel where dependencies permit. They don’t make production decisions.
- 01Goal & scope
- 02Architecture
- 03Bounded agent tasks
- 04Pull requests
- 05Human review
- 06Tests
- 07Release & monitoring
Example automations
- ↗ Excel and CSV processes turned into an app
- ↗ Automatically assembled reports
- ↗ Data transfers between systems
- ↗ Internal admin and tooling apps
- ↗ Document data extraction with AI
- ↗ Internal AI agents for bounded tasks
- ↗ Scheduled data updates and checks
- Python
- TypeScript
- Scheduled jobs
- Queues
- LLM APIs
- PostgreSQL
We pick the stack per project. These are common choices, not a menu.
Generated isn’t done. Verified is.
AI models make mistakes, and so do people. Our process is built to catch them before your users do: human review, tests suited to the risk and validation before release.
See what “done” means ↗How do we get started?
The first automation often fits a bounded build sprint: €5,000 + VAT for the first week, fixed price. After that you’ll know whether the next step is worth it.
Fair questions.
What should we automate first?＋
A step that repeats often, has a clear input and output, and whose mistakes are easy to spot. Not necessarily the one that annoys you most.
Can we replace Excel completely?＋
Sometimes. Often a better start is automating the manual steps around the spreadsheet and turning the process into an app once it’s clear.
Will an AI agent make decisions for us?＋
Only the ones agreed on. An agent can classify, extract and suggest. Actions with financial or legal consequences are approved by a person.
Isn’t Power Automate or another no-code tool enough?＋
Sometimes it is, and then we’ll say so. Custom automation pays off when the logic is complex, the data volumes are large, the systems lack ready-made connectors or you need full control over error handling.
Read before you buy.
Which step repeats every week?
Describe the process as it is today. Screenshots and example spreadsheets help.
Describe the process↗