AI agents for your business software. Ours are already at work.

We have been building business software since 2001. Today we connect it with AI agents: they handle the steps you have agreed and hand everything else to your team to decide. We start with one process that costs your team time today.

Discuss your process

Give us a call. Lena, our AI, will answer.

+49 6735 941616-0

You reach us on our regular company number. Lena greets you in German, takes your message and passes it on to us.

Call privacy notice

Listen to a sample call

In this fictional call, Lena greets Daniel in German and switches to English at his request. Both voices are AI-generated.

Callback note

Daniel Weber, Warehouse integration

Tomorrow from 2 pm, Callback from: Hans-Peter Schelian

Incoming number, confirmed by the caller

Read the call transcript
  1. Lena
    Guten Tag, hier ist Lena, die KI-Telefonassistentin der Schelian IT-Beratung. Was kann ich für Sie tun?
  2. Daniel Weber
    Hi. Could we speak English, please?
  3. Lena
    Of course. How can I help?
  4. Daniel Weber
    Hi, this is Daniel Weber. Hans-Peter was going to call me back about the warehouse integration. I’m free tomorrow from two in the afternoon.
  5. Lena
    Of course. Is this the best number for him to call?
  6. Daniel Weber
    Yes, the one I’m calling from.
  7. Lena
    Got it. I’ll let Hans-Peter know you’re free from two tomorrow about the warehouse integration.
  8. Daniel Weber
    That’s right. Thanks, Lena. Goodbye!
  9. Lena
    Goodbye, Daniel.

What an agent takes on for you

It handles the agreed steps itself
Correcting an address, matching a receipt, preparing a reply: whatever you have defined in advance runs without sign-off.
It hands everything else to your team
A price change or an exception is still decided by a person. The agent prepares the decision and cites its source.
It works inside your systems
Inbox, ERP and ticket system stay in use. The agent is connected and gets only the permissions it needs for its step.
InboxERPTicketsAI agentunder your rulesDoneAddress correctedYour teamApprove price

What runs in our own business

We recommend what we use ourselves.

Software development
Coding agents handle analysis, implementation, testing and review. Since March 2026 our Agentic Framework has split the work across seven roles.
since 2025
Phone reception
Lena answers calls, records the request and when the caller can be reached, and prepares the callback.
since May 2026
Customer support
An AI agent handles incoming support requests. In some areas it resolves around 90 percent on its own, with access to the systems involved. In others it clarifies with the sender what is missing and hands the complete ticket to a person.
since January 2026
Bookkeeping preparation
AI reads bank statements, matches each transaction to its receipt and lists the receipts that are missing. It names incoming documents along the way.
since October 2025

Three weeks instead of nine months

A transport planning system we developed ourselves has grown for more than 20 years. With AI agents we brought it fully up to date in three weeks. The new versions have been delivered and are in use.

The three weeks are calendar time: during them the agents worked on all seven days of the week, evenings and nights included. The nine months are our estimate for the same rebuild without AI, with two developers, five days a week and about ten hours a day.
Starting point
The overall system had grown across several generations: from .NET Framework 4 through .NET Core 2 and .NET 6 to .NET 8. It comprises 92 projects and around 800,000 lines of code, plus a generated data access layer.
Tests first
Before the rebuild we covered the existing behaviour almost completely with automated tests. That let us check after every step whether everything still worked as before.
Then the rebuild
With coding agents we brought the application and its security technology up to date; the software now runs on .NET 10. Agents also operated it through screen and browser and checked inputs and outputs from end to end.
The human share
It did not work without people. One developer accompanied the rebuild for around 60 hours: setting the tasks, checking the results, retesting by hand. Without AI it would have been around 3,900 working hours by our estimate.

We start with one process

You do not have to rebuild your business for this. A single process is enough, and as a rule you can see after a few weeks, from your own cases, what the agent delivers.

Your teamAgent

  1. 1 (Your team)

    Show us the process

    Your team shows us a process that often gets stuck or needs the same entry in several systems. We walk through it together, exceptions included, and agree with you what the agent handles itself and when your team decides.

  2. 2 (Agent)

    The agent works alongside

    We connect the agent to the systems involved. It then works on your cases: it handles what was agreed and hands over everything else.

  3. 3 (Your team)

    Test it on your own cases

    Your team checks the results against its own business cases and decides whether the agent goes into operation.

Two processes to start with

Turn tender documents into proposals

A bill of quantities arrives with hundreds of line items. An agent extracts quantities and units and suggests matching entries from your catalogue. Your bid team sees the original line and the source for each match.

Rules check units, and a classification model can learn from confirmed matches. Your team resolves open items, sets prices and signs off on the bid.

Lead generation for sales teams

Your sales team is looking for manufacturers with several warehouses. An agent checks company websites for firms that fit the profile and identifies publicly listed logistics contacts. Each match comes with its source.

Your team gets a shortlist with the reasoning behind each match and a suggested conversation opener. It checks the findings and decides whether and how to make contact.

Why it holds up in production

A prototype is quick to build. What counts in production is what comes next.

More than a prototype
Vibe coding produces a prototype within hours, and that is a good start. Once a department depends on it, it needs requirements, tests and someone who maintains the solution.
Data protection from the start
With customer or employee data, the GDPR applies. We first clarify where data flows, who may do what, what is logged and how access requests and deletion work.
Test first, reviewed independently
One agent writes the test before anything is built. A second agent reviews the code without having seen it being built.
Reviewed by agents, backed by us
In our own development we define what gets built and how far a coding agent may work on its own. Where tests clearly show that something works, it may carry the work through to delivery. Everything else is tested by a person, and the responsibility stays with us.
How we work in detail

From hand-written code to AI agents

We have been building software since 2001, 17 of those years without AI, because no AI tool yet existed that helped with programming. The first arrived in 2018, and since 2021 another step has followed almost every year. That is why we know both sides: how long something takes without AI and what gets faster with it. Anyone who started two or three years ago is among the first today. With every year of waiting, they pull further ahead.

Each point is a step, and its height counts the steps. The time axis is to scale.
The individual steps

Which process takes up too much of your team’s time?

Bring us a process that regularly stalls, generates too many follow-ups or requires the same data to be entered in multiple systems. We will discuss where the effort goes and which first step makes sense for your business.

Arrange an initial conversation

Your contact

Hans-Peter Schelian

Founder and owner

Hans-Peter Schelian founded Schelian IT Consulting in 2001 and runs the company. A qualified master technician in information technology, he has been in business since 1985.

More about us