An agent that actually works the process through
A chat replies. An agent does the whole step: reads the attachment, looks the details up in your systems, prepares the document and passes it on. Most often I do this on the operational admin of manufacturing and logistics companies, where the retyping piles up.
Why choose me?
What you get out of working together, beyond a working solution
A whole process step, not one reply
The agent is given a task rather than a question: process an incoming order, handle a complaint, prepare an invoice. It looks up what it needs, checks its own result and writes it where it belongs. When it is unsure, it escalates instead of guessing.
The operational admin that hurts most
Invoicing, quotes and enquiries, incoming email, complaints, service tickets, delivery notes. Work that moves nobody forward but has to be done every day, and in peak season eats half of your people's capacity.
A human stays on the decisions
Wherever a mistake costs money or reputation, the agent prepares and a person sends. How far its autonomy goes is set by the risk of the specific step, not by what demos better.
Every step is traceable
For each case you can see what the agent read, what it looked up, what it did and why. Without an audit trail, automation is a black box that nobody on the floor will let near real documents.
Connected to your systems
The output does not end in a chat window. It goes into ERP, CRM, accounting, WMS or the helpdesk, where people carry on working with it. An unconnected agent just produces text for somebody to retype.
Measurable savings
Before we start, we measure how much time the work takes and how many cases flow through per day. After go-live we repeat it, so you know what the agent gave back and what running it costs.
Types of solutions
Choose the solution that best fits your needs
Invoices and documents
Incoming paperwork extracted, checked and filed.
- Extraction from invoices, delivery notes and orders
- Matching against orders, stock and suppliers
- Flagging discrepancies instead of silent approval
- Writing into accounting or ERP
Quotes and enquiries
A draft assembled from price lists, templates and history.
- Quote drafted from past orders and price lists
- Checks on completeness, margins and availability
- Missing details chased up with the customer
- Human approval before anything is sent
Email, complaints and tickets
Incoming work sorted and driven to a conclusion.
- Sorting and assignment to the right owner
- A draft reply referencing the case history
- Complaints handled by your own procedure
- Service tickets enriched with asset and contract data
Where to find me and where I work from
I am based in Prague and in Tálín near Písek. Consultations run online or in person. I work with clients across the Czech Republic and abroad.
Prague
Letňany and surroundings
Tálín near Písek
South Bohemia
Online
Czech Republic and worldwide
Frequently asked questions
Answers to the questions clients ask most often
01What is agentic AI, and how is it different from the chat we already use?
A chat answers a question and its job ends there. An agent gets a goal and the tools to reach it: it can read an attachment, call into your ERP, look up history, create a document and decide that a given case belongs to a human. You can tell the difference because the process finishes, rather than because you got a nicely written paragraph.
02What if the agent gets it wrong?
That is designed for from the start. On steps with financial impact the agent prepares and a person sends. Where autonomy goes further, we measure the success rate on real cases and set the threshold at which the agent asks for review instead of guessing. Every step is recorded, so a mistake can be traced and corrected rather than merely inferred from the outcome.
03Which models do you use?
Commercial models from OpenAI, Anthropic and Google, chosen by task, cost and data requirements. I build so the model can be swapped, because today's best will not be next year's.
04What does it cost to run?
Infrastructure plus model usage, driven by the number of cases processed. An agent consumes more than a simple query because it takes several steps, and you should see that up front. An estimate for your volumes comes before any development, as part of the audit.
05Will it cope with our exceptions?
Exceptions are usually the reason an off-the-shelf tool is not enough. We go through them during the audit and decide which ones the agent covers and which are cheaper left to a person. The goal is not a hundred percent of cases, but the majority that eats the time.
06Do we have to change our existing systems?
No. The agent connects to what you have through an API, an MCP server or a secure data exchange. The point is to take work away, not to add another system to maintain.
07Will it replace people?
In practice it looks different: the retyping and deadline-chasing goes away, and review work plus the cases nobody had time for comes in. In the companies where I deploy this, the usual driver is that routine roles cannot be staffed, not that anyone wants to send people home.
Got admin that eats hours?
Tell me what gets retyped by hand in your company and how many cases flow through each day. On a free intro call I will estimate whether an agent can take it on and what that would involve. Online or in person in Prague and South Bohemia.
Currently capacity for 2 new projects. We start with an audit, not development.
Before I build anything, we go through your processes and I write up where AI pays off, how much it saves and what running it will cost. The audit fee is deducted from the project that follows.
Or go back to the contact form