DE Get notified
Services · AI agents

Custom AI agents. And honest when you don't need one.

An AI agent takes on tasks that cost desk time today: reading, sorting, preparing, following up. We build agents that work by your rules. And before we start, we check honestly whether your problem is better solved with something simpler.

Get notified
  • By your rulesThe agent works according to your operating procedures
  • You chooseFour autonomy levels per process
  • HonestIf something simpler is enough, we tell you
Example: agent in customer serviceAwaiting approval
  1. Request“When will my delivery arrive?”

    Customer email about order 24-118

  2. ProcedureSOP “Answer delivery request”

    Your rules, step by step. If one is missing, the agent asks.

  3. ToolsPre-built tools only
    • Look up order
    • Fetch shipping status
    • Draft reply
  4. Check layerChecked before anything happens
    • Read-only access to orders
    • Reply follows the SOP
    • Every step logged
Send the reply to the customer?
Examples

What an agent can take on for your business.

Typical tasks from businesses we know. Decisions that no rule covers stay with you.

Service

Monday morning, full inbox

TodayA complaint, a callback request, a delivery question and a newsletter all sit at the top. Whoever reads first decides what is urgent.

With an agentThe agent has sorted everything before you arrive: the complaint on top, reply drafts ready, the newsletter archived. You check and send.

Accounting

Three documents, one mismatch

Today40 ordered, 38 delivered, 40 invoiced. You only notice if someone puts order, delivery note and invoice side by side, line by line.

With an agentThe agent puts the three documents side by side and shows you only the line that doesn't match, with a note on what is missing.

Purchasing

The confirmation that never came

TodayThe order went out two weeks ago. Nobody notices the supplier never confirmed until the day the goods are missing.

With an agentThe agent notices the missing confirmation after the deadline you set and follows up politely. If nothing comes back, it lets you know.

Sales

The 2019 quote as a template

TodayOpen an old quote, delete items, look up prices, change the greeting. And hope the old customer's name doesn't appear anywhere.

With an agentThe agent turns the enquiry into a draft with your current items and rules. You read it, adjust and send.

Engineering

Is it in the data sheet?

TodayA customer asks whether the part can handle 120 °C. Someone searches three PDFs and the manufacturer's site and finds two different figures.

With an agentThe agent searches data sheets and manufacturer details, names where it found the answer and flags it when sources disagree.

Management

The Friday report

TodayEvery Friday someone copies figures from the inventory system, the shop and accounting into a spreadsheet. By the time it's done, the afternoon is gone.

With an agentOn Friday morning the report is already there, with a short note on what was different this week.

Root cause before symptom

Do you really need an agent?

Some people come to us saying “we need an AI agent” and actually need something else. We tell you honestly, before you spend money.

Agent

An agent fits when …

Cases vary and can't be captured in a rigid rule, but the decision follows your rules.

See an example
Varying cases
Every enquiry is a little different, but there are recognisable patterns.
Text and language
Emails, documents or notes need to be read, understood or written.
Judgement by rules
It takes an assessment that can be described in an operating procedure.
Simpler

Something simpler is better when …

The root cause can be removed. Then the work never arises.

See an example
Orders by email
An online shop, a customer portal or an interface: the order arrives in the system correctly from the start.
Always the same steps
A fixed rule or automation. It behaves the same every time and is easier to check.
Data in two places
An interface between the programs instead of an agent that retypes.

Often it is a mix: first remove the root cause, then an agent for the cases that remain.

Get notified

How we build agents

This is how we approach every agent, however big the task.

  1. 01

    Procedure first

    The basis is an operating procedure (SOP) that we write down with you. If a rule is missing, the agent stops and asks instead of guessing. Your answer becomes a new rule.

  2. 02

    Only the rights it needs

    The agent gets its own service account with exactly the rights it needs for its task. Every request to your systems is checked.

  3. 03

    Everything traceable

    What the agent does is logged. You can check at any time what it did and when.

  4. 04

    Autonomy and approvals

    You set one of four autonomy levels for each process and can change it at any time. Our default is level 2: “Software does it, a person approves”. At level 4, software decides according to your rules. The level also applies to money, contracts and customer contact. Your solution checks it before every action and logs what happens. You handle approvals directly in your solution, for example by email, Teams or app.

Autonomous, according to your rules

If you choose, some processes can run fully autonomously. Even then, the software does not decide freely: it follows your rules, just as an employee would follow the same operating procedure. You decide which processes may do this and where the limits lie. Everything is logged, and you can change the level at any time. The agent handles a replacement shipment for an incorrect delivery on its own. It submits a refund to you for approval.

Your agent or ours

We are not tied to any particular tool. If you already use an agent or an AI tool, we integrate it and give it the rules and interfaces it needs. Otherwise we build the agent for you. It runs in the cloud or on a dedicated server in Germany, depending on what suits your business and your data.

Frequently asked questions about AI agents

Does the agent replace employees?

The agent takes over routine so your team has more time for customers, decisions and special cases. What you do with the time gained is your decision. Approvals and judgements that no rule covers stay with people.

What happens to our data?

The agent only sees the data it needs for its task, through a service account with exactly those rights. Its access is logged. Whether personal data is involved, where it is processed and which privacy filter sits in front of it, we agree with you in writing before we start.

What if the agent makes mistakes?

Mistakes happen, with AI too. That is why the agent works by fixed rules, stops on unclear cases and requests approvals according to the chosen autonomy level. Every step can be traced through the log, and from every mistake we derive a new rule.

Once the portal launches, you see your fixed price there before you commit. The account is free; registration is still closed.

Get notified at launch