Answer from your knowledge
Policies, product information and procedures, answered with sources your team can check.
AI agents · Copilot Studio, Claude and more
Agents that do real work, on data you already govern.
Most AI pilots never reach production. We design agents around one real task your team does every week. In Microsoft, with Copilot Studio on your governed data. In a custom product, with Claude, OpenAI or the model that fits. Then we stay until people trust them.
YOUR NUMBER THIS WEEK€1,240
The coach
The coach is looking at your weeks…
Hi Sam. Costs and tax are covered this month, but your buffer is thin. Three things you could do:
Raise your hourly rate to €80About €1,100 more by December
Add two hours a weekCloses the buffer gap by March
Ask your boekhouder about a provisional assessmentSpreads 2026 tax over the year
Looking ahead: December has two quiet weeks. Plan them as time off now and your Number stays steady:
Book 21 Dec to 3 Jan as time offYour Number becomes €1,310 from Monday
Move one big invoice into NovemberSmooths the dip before the holidays
Your tax bucket is on track for the 30 April 2027 deadline. To stay there:
Keep €310 aside each MondayCovers income tax and Zvw at your rate
Check in again after your next big invoiceI will update the estimate straight away
The coach never moves money and never files anything. You decide. It is not tax advice.
What agents can take off your plate
Start with one job, done well.
Act on records
Create, update and summarise records in Dataverse or Dynamics 365, within each user's permissions.
Take on routine requests
The agent collects the details, sorts each request and routes it to the right person, so nothing waits in an inbox. Anything important still needs a person to approve it.
Coach your own customers
An agent inside your product that reads each user's own figures and suggests the next step, like the coach in ScaleBill. Built on Claude, OpenAI or Microsoft Foundry.
For founders and productsSafe by design
What stays under your control
- Agents use each person's own sign-in and permissions
- Your data policies and admin controls still apply
- A person approves actions that change important records
- Test and live environments, with documented releases
People first
Why agents fail, and how we avoid it
Agents usually fail because nobody changed how the work gets done. We map the task with the people who do it, design the agent around their day, and train champions who keep it improving.
A coach for one chiropractic practice, which became ScaleBill
New Life Chiropractie wanted to know, every week, what they could safely pay themselves once tax and buffers were covered. We built a weekly planning app with a coach agent that reads their figures and the Dutch tax rules and suggests practical next steps. It never moves money and never files anything; they decide. Their use case became ScaleBill, our own product, now open to every zzp'er in the Netherlands.
Built on Anthropic's Claude, called from a secure server function, with the same DTAP street and reviewed pull requests we use for clients.
More about ScaleBillAgent Pilot
From€7,500excl. BTW · 2–4 weeks
- One agent for one real task, in production
- Connected to your governed data
- Training, champions and 30 days of hypercare