Skip to content
Contact

AI Agents for Business

An AI assistant that finishes the task, and a person who approves it

An AI agent for business receives a task, goes through your data, uses tools (e-mail, CRM, calendar, web) and prepares a result for approval. We build agents with limited permissions, logging and an approval step, so that AI automation saves time without causing damage. From €1,600.

Ordinary automation follows fixed rules: when an order arrives, issue an invoice. An AI agent also handles steps that cannot be described by rules: it reads an enquiry, understands what the customer wants, looks up prices and puts together a draft quote. The difference is that the agent makes decisions. That is why we build it to decide within limits you set.

Every agent we build has three safeguards. Limited permissions: the agent sees only the data and tools it needs for the task and has no right to pay, delete or publish. Approval step: the output goes into a draft, to Telegram or into a queue, where a person approves, edits or rejects it with one click. Logging: every action, every input and output is stored, so you can trace afterwards why the agent did what it did. This is how our own content pipeline runs: an agent picks a topic, writes the article, a second agent grades it and the text is rewritten until it scores at least 8 out of 10. It is published only after a manual approval of the pull request. We build agents in n8n on top of Claude and other models, depending on the task and the data requirements. In practice this means sorting e-mails, preparing quotes from enquiries, regular research, and content that no longer starts from scratch.

  • An agent with limited permissions: no payments, deletions or publishing without a person
  • Approval step in Telegram, by e-mail or in a pull request
  • A log of every action for later review
  • Output quality evaluated by a second model before it reaches a person

What's included

What You Get

E-mail sorting and draft replies

The agent sorts incoming mail (enquiry, complaint, invoice, spam), pulls out the key details and prepares a draft reply. You send the reply or edit it.

Quote from an enquiry

From an e-mail or a form the agent assembles a draft quote: items from the price list, quantities, deadline, text. A salesperson checks it and sends it.

Research and summaries

Every week it goes through the sources you specify (competitors, market, legislation) and sends a short summary with links. No invented sources, only what it actually found.

Content pipeline

Topic selection, article, quality grading, revision and handover for approval. We use the same pipeline for our own blog.

Agent tools

The agent gets only the tools it needs: reading the CRM, search, calendar, creating drafts. Every tool has clear permissions.

Evaluation and tuning

For the first 30 days we track how many outputs you approve without changes and tune the instructions accordingly. We measure, not guess.

Perfect for

Who AI agents are for

  • Companies where a salesperson spends hours turning enquiries into quotes
  • Customer support with dozens of similar e-mails a day
  • Marketing that needs regular content without writing from scratch
  • Management that wants a weekly overview of the market and competitors
  • Companies with a large document archive that is hard to search

Our Process

How We Work

1

Task selection

We find a task where an agent makes sense: it repeats, has clear inputs and a person checks the output.

2

Roles and safeguards

We define what the agent may do, what it may not, who approves and what is logged. You get it in writing before we build.

3

Prototype on real data

The agent runs on your data in test mode. You grade the outputs, we tune the instructions.

4

Deployment and measurement

Production operation with approval. We measure the share of outputs approved without changes and the AI costs.

Pricing

Pricing Tailored to Your Project

AI agent from €1,600, operation and monitoring from €100 a month plus the actual cost of the AI model.

One agent

From €1,600

  • Design of roles and tools
  • Approval step
  • Limited permissions and logging
  • Test on real data
  • 30 days of tuning

Content pipeline

From €2,200

  • Author and critic (two models)
  • Revision to target quality
  • Handover via pull request or CMS
  • Social media posts
  • Telegram notifications

Operation and monitoring

From €100 / month

  • Server and updates
  • Output quality tracking
  • Instruction changes
  • AI fees at cost
  • Monthly overview

FAQ

Frequently Asked Questions

A chatbot answers questions. Automation performs fixed steps. An agent receives a goal, breaks it into steps on its own, uses tools and hands over the result. That is why it needs safeguards: limited permissions, approval and logs.

The agent's output always goes into a draft or an approval queue, never straight to a customer, into accounting or onto the website. The mistake stops at a person. Every run is logged, so you can find where it came from and adjust the instructions.

The agent runs on n8n on a server in the EU. The AI model receives only the content needed for the task. We choose models whose API data is not used for training. We write down in the documentation what data flows where.

Monitoring from €100 a month. AI fees are paid by usage: for our content pipeline it is $0.15 to $0.50 per article including grading and revisions. For an e-mail agent it is typically a few cents per processed e-mail.

We measure the share of outputs you approve without changes, the time from input to output and the cost per run. You see the numbers in a monthly overview. If the share of approved outputs does not improve, we adjust the instructions or tell you the task is not suitable for an agent.

Have a task an AI agent should be doing?

Describe it in a few sentences. Within 48 hours we will reply whether it makes sense, what the safeguards would look like and how much it will cost.
Free consultation