← Articles

How the assistant helps describe a problem before the first conversation

5 min read (updated: August 24, 2026)

A form does not understand the answers

The message “I need a system for handling orders” is a good start, but it is not enough to assess the work. You also need to know who will use the solution, how the company copes today, where the errors arise, what data is processed and what is supposed to change after the project.

An ordinary form can ask all the questions at once. Then it resembles a specification to fill in, and the client has to foresee which information matters. The other option is several exchanges of messages stretched over a few days.

I built a third way: a short description of the matter guided by an assistant. It does not replace a conversation with me. It helps gather the material before we start one.

Two equal ways to get in touch

On the site you can write a message right away or use a conversation that is guided step by step. The assistant is useful to someone who knows what hurts, but does not know the names of technologies and has no ready document with requirements.

The conversation starts with a description in your own words. The assistant asks one question at a time, reacts to the answers and finishes when it has a sufficient picture. The upper limit is 16 exchanges, but a well-defined matter can close much earlier.

What the assistant collects

The questions are mainly about the company and people’s work:

  • what is done by hand today,
  • where the process stops or produces errors,
  • who uses the current solution,
  • what systems and data the company already has,
  • what a successful project would mean,
  • whether there are legal, security or deadline constraints.

The budget and the technology may matter, but they are not the first question. First you have to understand why the company wants to change something.

What happens after the conversation

The system creates an organized summary for me and — if the client gives an address — a confirmation for them. No automatic quote and no promise of a deadline is produced.

On the basis of the material I choose the next step:

  • if the scope is clear, I can give a price range or a specific quote,
  • if a few pieces of information are missing, we arrange a short conversation,
  • if I need to go into the code, the data or the contracts, I propose a paid audit,
  • if the biggest unknown concerns how it should work, I can prepare a demo.

This way automation shortens the road to a decision, but does not make the decision for me.

Two different tasks for two models

The first message goes to a light model that recognizes whether the inquiry concerns the services at all. It does this in the background and does not run the conversation. This limits the cost and simple attempts to flood the system with junk.

If the message describes a real problem, a more precise model takes over the conversation. It remembers earlier answers, does not ask about the same thing twice and prepares a structured summary at the end.

The split was not made to multiply tools. Each model has a different, narrow task: the first protects the entrance, the second helps a person make the matter more precise.

Rules that matter more than the model

The assistant does not pretend to be me

It says plainly that it is Artur’s tool and uses the third person. A client should not discover after the fact that they were not talking to a human.

It does not price and does not promise

It does not know the client’s whole system, so it gives no price and no deadline on the basis of a few sentences. It can explain the possible routes, but I prepare the final assessment personally.

It asks in plain language

The client does not have to know what a programming interface, a data model or a job queue is. It is enough to describe: “an employee retypes orders from e-mail into a program” or “two people can book the same hour”.

It asks only for the information needed

The conversation should not collect data just in case. With a sensitive matter the client can first describe the problem in general terms and pass on the details only after confidentiality and access have been agreed.

What this system really saves

It does not replace a consultant and does not give a ready diagnosis in a few minutes. It saves the first, repetitive part of the conversation: it organizes the symptoms, the people, the data and the expected result.

Thanks to this the client does not have to write a technical specification, and I do not start with questions that have already been answered. The most important decision — whether the project makes sense and in what form to begin it — still belongs to a human.

For the technically interested

The site runs on Astro. Sessions and messages are stored in a Supabase database, and Claude models handle the initial classification, the conversation and the summary. Limits per network address and on the pace of sending protect the service against abuse. Summaries and contact details are handled on the server side, not in the user’s browser.

You can describe your matter to the assistant or write directly.

Case Study

An assistant that helps describe the problem before we talk

See the case study →

Have a similar problem?

Describe it in your own words — you do not need to know the names of technologies or prepare a specification.

Describe the problem