Insights · As of September 2026

Six months of a company run on AI agents: what works, what does not.

AI Studio is our product lab. For six months, administration, receipts and analyses have run on automated workflows that follow documented rules. People review and approve. This is what we learned for accounting and finance.

What works

Recurring workflows with a clear rule: matching documents, tracking deadlines, preparing reconciliations.
Documented rules instead of tribal knowledge: every decision traces back to a rule and its source data.
Human approval as a fixed step, not an exception.

What does not work

Edge cases without a rule: what was never described is handled wrongly or not at all.
Autonomy without a log: without the four-eyes principle and an audit trail, a result cannot be verified.
Cloud dependency: external services mean data leaving the premises and outages; everything runs in-house.

What this means for your accounting

The lab is neither a client project nor an offer. It shows which rules belong in a knowledge base and how approvals must work so your team stays in control. That experience feeds Step 2 and Step 3 of our three steps.

→ Services: Three steps · → Security & Data Protection · → AI Studio (product lab)