t.me/der_vlad
an AI agent for one process · for businesses whose enquiries live in chat

Reply to customers in a minute, not tomorrow

I take one process in your business — replying to enquiries — and lift one number in it. Upfront payment is zero: first a free measurement, and your starting point written down. You pay only a share of the uplift — and only once the number has grown and a second measurement shows it.

Free measurement of your reply funnel Your reply funnel is the path an enquiry takes from “they messaged you” to “they booked”. Message me on Telegram — I answer within the day. The measurement commits you to nothing: the numbers are yours to keep.
what it looks likeconversation log
21:42 customerHi! How much is a course for complete beginners? It’d be two of us — me and my wife.
21:43 agent — on behalf of the businessGood evening! For two beginners the basic course is the right fit: 6 sessions, starting on Saturdays. The next dates are the 14th and the 21st. Shall I hold one of them for you?
21:45 customerThe 14th, please.
01 · how it works

Three steps. I’m the one risking money, not you

step 1

Free measurement

I measure your current number as a mystery shopper: I write in the evening and at the weekend, time your replies and count what’s lost. We put the starting point in writing — so there’s nothing to argue about later.

It costs zero. The numbers are yours even if we don’t work together after that.

step 2

A week of work

I put an agent on your process: it replies on behalf of the business along scenarios we’ve agreed, works out what the person actually needs, and moves them towards booking. Anything the agent isn’t sure about goes to a human.

step 3

Paid from the uplift

After an agreed period we measure again. If the number reached the threshold, you pay the agreed share of the uplift, with a cap and a deadline. If it didn’t, you pay nothing.

02 · numbers, not promises

What has already been counted

I label every number with its honest status — the same way I do in client reports. You won’t find promises of growth “several times over” here.

235 facts / week

That’s how many checkable facts (prices, links, promotions) my agent verifies every week on a paid travel-guide site. It has surfaced 26 discrepancies, each with evidence; six were critical, including a discount promised in two different cities at once.

✅ fact — running in production
2,500–16,800 PLN / year

An estimate of what a landlord with six flats loses to a manual monthly cycle: hours of routine, errors in utility reconciliation, late payments. Three scenarios; the rates and some of the figures are market assumptions.

🟡 hypothesis — to be confirmed by measurement
10 companies

A sector-wide measurement will go here: I’ll go through 10 companies as a mystery shopper and publish how many of them lose enquiries sent in the evening. Your company can be one of the ten — free of charge.

📌 planned
✅ fact — measured, verifiable 🟡 hypothesis — a calculation awaiting measurement 📌 planned — it will appear right here
03 · live agents

Not screenshots — working systems

Every agent can be seen in action. The demos run on made-up data — break them, ask awkward questions.

sailing school · enquiries “Watch Officer”

Answers incoming enquiries on behalf of the school: works out the person’s level and goal, suggests a course and a date, moves them to a booking. Anything unclear goes to the office.

Try the demo → 🟡 demo · in Russian
paid travel guides · content control “Freshness Auditor”

Once a week it checks 235 facts on the site: prices, affiliate links, promotions. It sends the owner only the discrepancies — each with evidence. When everything is in order, it stays quiet.

▶ short video: the weekly report — coming soon ✅ in production
flat rentals · monthly cycle “Czynszownik”

An assistant for a landlord with six flats: bills, messages to tenants, payment tracking, closing the month. The money is counted by ordinary code — the language model is not allowed near the arithmetic.

▶ short video: the monthly cycle — coming soon 🟡 prototype

Before this: 25+ agents in production use at a marketing-analytics platform (2025–2026, my previous job — plus 11 automated routines there, from competitor monitoring to integration-error control) and three apps of my own in the App Store and Google Play — built the same way. Links to all of it are below, in “who does this”.

04 · when the agent isn’t sure

The most important scenario is the failure scenario

Every agent eventually meets a question that isn’t in its script. The difference between a toy and a working system is what happens next.

▶ short video: the agent gets an off-script question and hands it to a human — coming soon

  • The agent doesn’t touch money. No payments, no price changes, no confirming refunds — not without a human approving it.
  • Anything unclear goes to a human. A question outside the agreed scenarios isn’t guessed at; it’s passed to you or to your office, flagged.
  • Every action is logged. What the agent said, to whom and when — you can open the log at any moment.
05 · price

In the open: zero now, a share of the uplift later

first clients in a sector

0 zł upfront

  • A fixed sum — payable only once a threshold agreed in writing is reached (for example: “occupancy is up by 10%”).
  • If it doesn’t grow, you pay nothing. The measurement and the report stay with you.
  • Instead of part of the fee I take a written reference and permission to publish the case with its numbers.
after that

10–20% of the uplift

  • A share of what’s gained or saved — with a cap and a deadline (usually 3 months), so the percentage doesn’t run forever.
  • All three numbers — the share, the cap, the deadline — go into a written agreement before work starts.
  • Maintenance of what’s been built is a separate, plainly stated monthly sum, if you want it.
06 · what I need from you

Three things — and none of them are about money

What I’ll need
  • A. Access to customer conversations — connecting the bot to your business messenger (Telegram / WhatsApp) or to the inbox where enquiries land.
  • B. 30–40 minutes talking through the process: how enquiries arrive today, who answers them, what counts as success.
  • C. Your price list and schedule — what you already send customers becomes the agent’s knowledge base.
What I won’t need
  • Access to your bank, till or payment systems.
  • Passwords to your CRM or admin panels at the start — if they’re needed later, we agree that separately and in writing.
  • Your time for “implementation”: for a week the work is mine; you answer questions as they come up.
07 · security

What the agent cannot do — and how to switch me off

The agent is built so that it cannot
  • Spend or transfer money.
  • Change prices, confirm refunds or enter into contracts.
  • Answer outside the agreed scenarios: an off-script question goes to a human rather than into improvisation.
  • Delete your conversations or your data.
Your guarantees of control
  • Everything built is yours: the code, the bot, the knowledge base and the logs stay with you even if we part ways.
  • Access is revoked in a minute: one access key — you switch it off yourself and the agent stops immediately.
  • The log of everything the agent does is open to you at all times.
  • The agent and its logs run in an isolated environment in the EU. If we part ways, I hand them over to you or delete them — your choice.
  • Business data is never sold or published. Only the necessary minimum reaches the language model — on API terms, with no model training on your data.
  • I’ll say “no” if the task isn’t one for AI — before I take it on.
08 · who does this

I work alone — and that’s a feature

Vlad Derkach — product manager and builder of AI agents. Whoever answers your message on Telegram is the same person who builds the agent and answers for the result. No “team of experts”, no handing the work to subcontractors.

  • 25+ agents in production use at a marketing-analytics platform (2025–2026).
  • Three apps of my own in the App Store and Google Play — from the first line of code to release, alone.
  • Product Discovery Flow, an open tool — 15 agents for product research, source code open under the MIT licence.
  • No employees and no subcontractors: no markup for a “team”, no broken telephone, and no one else’s projects queued ahead of yours.
  • I work from Gdańsk and take one process at a time — which is why I sometimes say plainly, “not right now”.

Check me instead of taking my word for it

There isn’t a single testimonial or screenshot below. Only addresses that open right now — apps in the stores, source code and a working bot. The links lead outwards, to platforms I don’t control.

apps · published under my name

Three apps released in 2026. Each one passed Apple’s and Google’s review — an outside check, not my word about myself. All three are free and open without registration.

✅ fact — published, downloadable
open tool · source code open

Product Discovery Flow — a pipeline of 15 agents that carries product research from a raw brief to a finished report. It installs with one command. The code is open under the MIT licence: you can see exactly what each agent does, rather than “trust the black box”.

  • Open the tool The write-up and one-command install: what all 15 agents do, step by step. product-discovery.dervlad.com
  • Source code The whole repository, MIT licence — read it, copy it, use it yourself. GitHub
  • In a public catalogue The same tool in a third-party catalogue — proof it doesn’t only live on my own disk. skills.sh
✅ fact — publicly available
try it yourself · and who I am beyond this site

The demo bot runs on made-up data for a sailing school: ask it anything, try to throw it off. It replies in Russian — I built it for a Russian-speaking client. The other addresses are about my path before agent work.

  • Demo bot “Watch Officer” The same agent as in “live agents”. The data is invented — break it freely. Replies in Russian. t.me/vahtenny_demo_bot
  • My professional track record A separate site about my product work: what I did, what I built, with which numbers. dervlad.comLinkedIn
  • Message me directly I answer myself, within the day. No enquiry form and no account manager in between. t.me/der_vlad
✅ open right now
09 · common questions

Three honest answers

What if the number doesn’t grow?

You pay nothing. The before/after measurement and the report stay with you — that’s worth something in itself: you’ll know your reply funnel in numbers. I lose a week of work; that’s exactly why I measure before starting and only take on cases where I can see the number really can be lifted.

Why a percentage rather than a fixed price?

So that I earn from the same thing you do. A fixed price pays the contractor even when the result is zero — a share of the uplift doesn’t. And the cap and deadline in a written agreement protect you from a percentage that never ends.

My process isn’t about enquiries — would this work?

Possibly. The processes that automate well are the ones that repeat daily and live in messages, calls or spreadsheets: bookings, reminders, payment tracking, checking data. Write to me and I’ll tell you honestly whether I’ll take it on. I do say no: if AI won’t solve your problem, you’ll hear that on the first call, free.

10 · first step

Find out your number — it’s free and commits you to nothing

Message me the word “measurement” — I’ll go through your reply funnel as a customer, in the evening and at the weekend, and send the numbers back in writing. What you do next is up to you.

Free measurement of your reply funnel

telegram: @der_vlad email: info@uladzislau.com