Half-day workshop · online or in your building
AI That Actually Does The Work
Asking an AI a question saves you a minute. Giving an AI a job saves you a week. The difference is not the prompt. It is whether the thing can reach your systems.
This is a workshop for teams who have had the chatbots for a year, use them daily, and have quietly noticed that nobody’s workload actually went down. Half a day, built around four AI systems that are running in production for paying clients right now, shown live and doing real work.
The offer
Sit through the first hour. If your team is not going to get its money back out of the other three, say so and we stop there, and you pay nothing.
I would rather lose a booking than take a fee from a room that is not going to use it. You are deciding after seeing the material, not before.
Why this is not another prompting course
Almost every AI course on the market teaches you to talk to a chat window. That is a genuinely useful half hour, and the internet already gives it away. It also leaves you doing the same job you were doing before, with an extra tab open.
The work that disappears is the work you can hand over completely, and that only happens when the AI can reach the place the work lives: the mailbox, the store, the phone line, the folder of documents. That step is the whole subject here.
I am not a trainer who learned these tools this year. I build and run these systems for paying clients, and the examples in this workshop are those systems, shown live against real work. If a demo cannot be run in front of you, it is cut from the material. That rule is the whole reason this is worth your half day.
What gets shown, live
Four systems, each doing a different kind of job, each running in production right now. Nothing here is a mock-up or a screenshot.
A voice assistant answering a real business phone line
AI acting in real time, in front of a customer, with rules it is not allowed to break. Hear it.
A checker that reads a nineteen-chapter manuscript
It reports what is wrong and, more importantly, refuses to guess at what it cannot judge. This is the example that teaches the difference between a rule and a model.
A store assistant a stranger has to trust
Answering customer questions on a live shop, quoting real prices from a live catalogue instead of a number that merely sounds right. See the agents.
A system that reads hundred-page documents and cites its sources
Every finding carries the page it came from, under a rule of verify rather than summarise. See Deal OS.
The half day, part by part
Three hours thirty of material, thirty minutes of questions with no agenda. Two of the five parts are exercises on your own work rather than mine.
1. Why your AI is not saving you time
35 minThe chat window is a dead end, because you are still the one carrying data in and out of it. The three shapes of AI work are answer, assist and act, and almost everybody stops at the first. We run the same task both ways with a stopwatch on screen.
2. Giving an AI a job
55 minWhat access actually means: a file, a mailbox, a database, an API. The four properties that make a job worth automating first, and an exercise where you score the jobs in your own role. Most people find their best candidate is not the one they walked in thinking about.
3. Making it safe enough to leave alone
50 minA tool that quietly makes a hundred silent decisions is worse than no tool, because nobody can check it. Anything uncertain is left alone and listed for a person. You will see an agent refuse to act, and why that refusal is the feature rather than a fault.
4. Build one, together
50 minWe build a small working automation live, from nothing, on a real file the room supplies. The room picks the job. There are no slides in this part, and you leave with the working thing rather than a recording of me making it.
5. What it costs and what breaks
25 minHonest running costs at small volume. What breaks in month three, and why it is almost always a change upstream rather than the AI. When to build, when to buy, and when to leave it manual, which is a real answer and is given as one.
What each person leaves with
- •A scored list of the automatable jobs in their own role, not a generic example.
- •For their best candidate, the three things it must never be allowed to decide on its own.
- •The working automation built live in part four.
- •A one-page brief template for commissioning this work properly, so nobody gets sold a chat window.
What this workshop is not
It is not a certification, it is not a route into a developer job, and it will not make anyone an AI engineer in an afternoon. It also is not a sales pitch with an agenda attached. Nobody is sold a build in the room. If something worth building comes out of it, that is a separate decision, made later, in writing, at a fixed price agreed before it starts.
Is this for your team?
Worth being blunt about, because a half day with the wrong room helps nobody.
Yes, if
- •Your team already uses ChatGPT or Copilot daily and nobody’s workload has actually dropped
- •The repetitive work sits in operations, finance, marketing or support rather than in engineering
- •You can name one job somebody does every week that nobody enjoys
- •You want to know what to build before you pay anyone to build it
No, if
- •You need a certification or an accredited qualification. This is not one
- •You want a beginner class in what AI is. Assume the room has used these tools for a year
- •You are looking for a team-building afternoon. People work in this one
- •You have already decided what to build and just need it built. Then skip this and start a pilot
On availability, honestly: this is run alongside client build work, not instead of it, so it is roughly one workshop a month. That is a real constraint rather than a scarcity tactic, and it means dates are usually four to six weeks out.
What it costs
Three ways to book it, and the number is on the page rather than behind a discovery call.
Private team workshop
$1,900
- •Up to 10 people from one company
- •Online, or in your building
- •Your own files in the live build
- •Half a day, fixed price
Agree the date with me first, then pay.
Individual seat
$260 / person
- •Scheduled open session
- •Six to ten people, mixed companies
- •Minimum six to run
- •Same material, shared room
Refunded in full if the session does not reach six.
Monthly programme
$650 / month
- •Two hours a month, ongoing
- •One team, one recurring slot
- •Work on your real backlog
- •Cancel with a month’s notice
Cancel with one month’s notice.
On the monthly programme: you manage it yourself. Cancel, update a card or download any invoice at your billing portal, using the email you paid with. No message to me required, and cancelling takes one click.
All prices are in US dollars and exclude VAT. Romanian clients have Romanian VAT added on top. EU businesses outside Romania with a valid VAT number are normally handled under the reverse charge, and clients outside the EU pay no EU VAT. Invoiced by DEVALAND MARKETING S.R.L., VAT number RO50841395. Your exact treatment is confirmed in writing before anything is booked.
Where it goes afterwards, if it goes anywhere
Most teams stop after the workshop, and that is a perfectly good outcome. For the ones that do not, here is the whole path, so you can see it before you start rather than be walked along it. Every step is priced separately and decided on its own.
Step one · half a day
The workshop
Your team finds the jobs worth automating in their own week, and you leave with one working automation and a brief template.
Step two · one to two weeks
A paid pilot on the best candidate
One deliverable, fixed price, run on your real data, and the fee is credited toward the build if you continue. You buy proof before you buy a project. How pilots work.
Step three · only if step two earned it
The build, or a monthly slot
Either a fixed-scope build of the thing that worked, or two hours a month against your backlog so your team keeps going without me writing everything.
Nothing above is agreed by booking a workshop, and nothing is sold in the room.
Who runs it
Marius Andronie, of Devaland Marketing SRL in Romania. I write the code and I run the systems, so the workshop is taught by the person accountable for them rather than by someone reading a deck about them. Nothing belonging to another client ever appears on a screen, in any form: every demonstration runs on synthetic or consented data, and the same rule protects your material afterwards.
Frequently asked questions
Is this a course about writing better prompts?
No. Prompt technique is a half-hour topic and the internet already teaches it for free. This workshop is about the step after that: giving an AI a job it can do without you, against systems you already run. That is where the hours actually go back into your week.
Do we need developers in the room?
No, and it works better if you have a mix. The people who know which work is repetitive are usually not the developers. Operations, finance, marketing and support people get the most out of it, because they can see their own week in the examples.
What do we need to bring?
One real file and one job you already hate doing. In the fourth part we build a working automation live, and the room chooses what it is. Bringing a genuine example is what makes that part useful rather than a demonstration.
Is anything recorded or shared?
Nothing about your business leaves the session. The examples I show run on synthetic or consented data, never on another client’s material, and the same rule protects yours. If you want a recording of your own session for people who could not attend, that is your decision to make and your copy to keep.
You say you do not take sales calls. Is this not a call?
It is paid delivery, which is a different thing. I do not do discovery calls, pitch calls or exploratory chats, and intake for everything here stays written. A workshop is work you have booked and are paying for, with an agenda and a deliverable, so it does not have the problem that sales calls have.
What if we decide the answer is to leave it manual?
Then the workshop did its job. Part five includes when not to automate, and that answer is given as seriously as the others. A team that leaves knowing which three of their five candidate jobs are not worth automating has saved more money than one that automates all five.
Can you build the thing afterwards?
Yes, and it is a separate decision made later, in writing. Nobody is sold a build in the room. If a candidate from the workshop is worth building, it goes through the same paid pilot as any other project, at a fixed price agreed before it starts.
Do you run it in languages other than English?
English and Romanian. The material, the exercises and the templates are in English in both cases, because that is what your team will search for afterwards.