01AI & consulting

AI agents that get tasks done.

An agent reads, decides, acts and asks when it is unsure. I build agents that work inside your existing systems, may only do what you allow, and whose work you can trace at any time.

What an agent is

From chatbot to temporary team member

A chatbot answers, a workflow follows a fixed sequence. An agent is given a task and completes it in several steps: it reads the enquiry, looks things up in the shop or the ERP, writes the reply and submits it for approval. For recurring tasks with small decisions, that is the difference between “saves a bit of time” and “done”. The strategic groundwork is done in AI consulting; fixed sequences I build as AI automation.

Boundaries first

Only what you allow

Before an agent does anything, I define with you what it may do alone, what needs approval and what is off limits: no payments, no deletions, nothing sent externally without your okay. Every step is logged, costs are capped. The agent processes personal data under the Swiss Data Protection Act, on request with models located in Switzerland or the EU.

Services

Agents that prove themselves in SMEs

Inbox agent

Sorts incoming enquiries, answers standard cases, prepares replies and quotes for approval and adds new contacts to the CRM.

Microsoft 365 & Gmail·Replies for approval·CRM maintenance

Shop and ERP agent

Maintains product data, attributes and translations, answers order-status questions from shop and ERP and reports stock problems before customers notice them.

Shopify, JTL & WooCommerce·Product data & translations·Order status & stock

Support agent

First reply from your knowledge base, multilingual and around the clock. What it does not know for certain goes to a human as a ticket, with a summary and a suggestion.

Knowledge base·Ticket system·Handover to humans

Document agent

Reads invoices, orders and contracts, checks them against orders, prepares bookings and reminds you of deadlines.

Invoice checking·Accounting preparation·Deadlines

Knowledge agent for the team

Answers internal questions from SharePoint, manuals and contracts, and only with what the person asking is allowed to see anyway.

SharePoint & OneDrive·Permissions preserved·Source references

Research and sales agent

Qualifies enquiries, gathers background on prospects, prepares meeting notes and drafts follow-up emails.

Qualify enquiries·Meeting preparation·Follow-ups

Operation and monitoring

An agent runs like any other system: with a log, a cost cap, a monthly review and adjustments when your business or the models change.

Log·Cost cap·Monthly review

Process

How a project runs

01

Choose the task

Recurring, clearly scoped, measurable. One task, not ten.

02

Set the boundaries

Which systems, which actions, what only with approval. Data protection under the Swiss act included.

03

Shadow mode

The agent proposes, you confirm. Only when the quality is right does it work independently within its scope.

04

Operation

Log, cost cap and monthly review. The agent gets better because someone is watching.

FAQ

Frequently asked questions

What is the difference to AI automation and chatbots?
An automation follows a fixed sequence, a chatbot answers questions. An agent decides for itself, within its boundaries, which steps are needed, uses tools such as mailbox, shop or ERP to do so, and asks when it is unsure. Much starts as automation and only becomes an agent once decisions are added.
Which tasks are suitable?
Recurring tasks with many small decisions and clear rules: inbox, standard enquiries, product data, invoices, internal questions. Not suitable are tasks where a mistake is expensive and nobody double-checks, such as payments or contractual commitments.
What happens if the agent makes a mistake?
That is what boundaries and approvals are for: everything that goes out or moves money is confirmed by a human. Every step is in the log, mistakes can be traced and the rules are adjusted. In shadow mode at the start you see how the agent decides before it does anything on its own.
Does the agent see our data, and where is it processed?
Only the data it needs for its task, and only with the rights you give it. I work under the Swiss Data Protection Act: business versions that do not train on your data, on request models located in Switzerland or the EU, or self-hosted when data must not leave the building.
Which technology do you use?
Models from Anthropic (Claude) and OpenAI, Microsoft Copilot Studio for agents in Microsoft 365, n8n and Make for integrations and MCP as an open standard for connecting tools cleanly. I am not tied to any vendor and choose by task and data protection.
What does an agent cost?
A pilot agent for a clearly scoped task starts at CHF 3,000, including shadow mode and handover. Running model costs for most SME tasks are below CHF 100 per month and are capped. You receive a quote in advance with effort and expected time savings.

Contact

Which task should an agent take over?

Tell me the task that costs you the most time in daily business. I will tell you whether an agent is suitable for it, what it should be allowed to do and what the pilot costs.

Phone
Email
Reply
Usually within 24 hours. Your details stay with me.

How your data is handled: privacy policy.