Where to start with AI: one evening, one task

Where to start with AI: one evening, one task

Somebody opens the subject, sees a hundred options and closes the tab. Ten-minute reviews, a trial period with every service, and the task still not done. Below is a rule that removes the choosing entirely, and three tasks any of which can be closed in one evening.

Somebody opens the subject, sees a hundred options and closes the tab. Ten-minute reviews, a trial period with every service, and the task still not done. Below is a rule that removes the choosing entirely, and three tasks any of which can be closed in one evening.


If there is no time to read: five steps

  1. Do not choose a tool. Choose one task you have right now.
  2. Describe it in ordinary words. Not "a prompt" but the way you would tell a neighbour.
  3. One evening, one task. Do not try a second until you have finished the first.
  4. Success on the first evening is one finished result. Not an understanding of how it works.
  5. Set aside a dollar or so for the attempt. No subscription is needed at the entrance; payment is on usage.

Why everybody gets stuck at the entrance

A familiar picture. Somebody decides to finally work out what all this AI is. Opens a search.

The first link is a review of the "top 30 services". The second is "how to choose AI for your task". The third is a forty-hour course. Every service has its own trial period, its own interface and its own registration.

An hour later there are eight tabs in the browser, a mush in the head, and a feeling that this is not for you. The tabs get closed and the subject postponed "until later, when there is time to work it out".

The problem is not you and not your age. The problem is the order of actions: almost everybody starts by choosing a tool when they should start by choosing a task.

The difference is like "which hammer should I buy" against "I need to put up a shelf". In the second case the hammer finds itself, and it often turns out a hammer was not what was needed.


Who I am and what my first evening with AI gave me

I have been working with AI since 2019. I started at exactly the wall I am writing about: there were fewer tools, but each needed working out.

One thing helped — I stopped studying and started doing. My first task was transcribing an hour-long recording of a conversation so I would not have to listen to it again. It took ten minutes. After that everything else went by itself, because I had a foothold: I already knew something worked for me.


What actually gets in the way

Three things, and not one of them is about the difficulty of the technology.

A catalogue at the entrance is a wall rather than an abundance. When somebody is shown a thousand possibilities, he does not rejoice at the choice, he gets lost. Choosing from a thousand never happens: it gets postponed.

Studying substitutes for doing. Reading a review gives the same sense of movement as work done: you spent an evening usefully and learned something. But there is no result and there is a sense of progress — and you can go round in circles like that for months.

People wait for understanding before a result. "First I will work out how it works, then I will try." In practice it is the reverse: understanding comes after the first result rather than before. While there is no result there is nothing for the explanations to attach to.

One participant on our courses put it precisely: "at first it was rather hard because of the sheer volume, I had to rewatch the lessons several times." Not because the material is bad — because the volume at the entrance is always more than can be taken in at once.


The rule: one task, one tool, one evening

It states in one line and removes the whole problem of choosing.

On the first evening you are not studying AI. You are solving one task of your own.

Which one hardly matters. What matters is that it is real: one you actually have rather than a practice one from an example.

What counts as success. Not "I understood how it works" and not "I looked at what the models can do". Success is one finished result you can show or use: a text written, a document analysed, a picture ready.

If the evening ended with no result, the evening was wasted. If there is a result, even an imperfect one, everything else follows by itself.


Three tasks to choose from

Any of them closes in an evening and requires nothing beyond the ability to write messages.

The first: a text that has to be written. A birthday greeting, a letter to a management company, a listing for something you are selling, a covering letter. Most people have a task like that hanging over them right now.

What to do: describe the situation in your own words, get the text, correct a couple of sentences.

The second: a document you need to make sense of. The manual for a new appliance, a contract, a long handbook. Instead of reading forty pages — three questions and three answers.

The third: an old photograph. Restore a damaged shot, remove a stranger from the frame, make a portrait as a present. The result is visible at once and can be shown to the family — the best test of whether it worked.

How to choose between the three. Take the one you have today. Not the one that sounds more interesting.


What this looks like with no tool to choose

Now the least obvious part. You do not have to decide which service to use.

There is a section where you simply write the task in ordinary words — Cerebrum. Not "find me a tool for photo processing" but "I have an old photograph of my grandmother, it is scratched, I want to restore it and print it".

What it does. Asks one or two clarifying questions if something is unclear. Selects what this is solved with. Breaks the work into steps in order and says which to start with today.

A file attaches with a paperclip, as in a messenger — nothing has to be prepared specially.

And the price is visible before you press. That matters for a first attempt: you know in advance what it will cost rather than finding out afterwards.


What is already prepared so you do not start from nothing

Worth saying separately, because many people do not know this and honestly try to compose everything themselves.

The prompt library — texts free after registering

Ready phrasings for standard tasks: texts, letters, greetings, analysing documents, working with images. You do not invent a request — you take a template and insert your own details.

More than 1 700 prompts for text models (26 categories) open with ordinary registration, no subscription. Prompts for video come with the Basic plan. If you want to try with no outlay at all, this is where to start.

Over 200 ready assistants

These are not prompts but configured roles: a copywriter, an art director, a lawyer, a sales specialist, an HR assistant. Sorted into fourteen sections.

You choose not "which model to take" but "what I need to do". Open the right one, describe the task in words, get the result.

Worth naming separately is Prompt Master: it writes the requests for you. You say in an ordinary sentence what you want and it assembles a full professional request. It is the GPT Prompt Master assistant from the assistant catalogue, Basic plan.

Ready spreadsheets for whole tasks

If the task is bigger than one question, there are spreadsheet templates with everything broken into steps: you enter the source data in the first cell and a chain processes it through dozens of requests in turn.

Eight such templates — for HR, marketplaces, online shops, audience analysis, warming up subscribers, creating lessons, copywriting and running social media.

Building a site — also from ready pieces

If your first task turns out bigger than a text or a picture — say you need a page about yourself or a simple landing page — building it from scratch will not be necessary either.

The training covers the cascading prompt technique — how to break a large task into a chain: the first step invents the structure, the second writes the headlines, the third the block texts, the fourth adds detail. The phrasing from the lesson: you lead the model along a clear plan, like a scriptwriter.

And there is a section of ready page prompts — with previews, so you choose by result rather than by description.

Why that matters for a first evening. The entry threshold here is not the ability to write requests. Most of the work is already done — you only have to insert your own details.


Honestly about money

One point better said at once, or you will run into it at the second step and decide you were misled.

There is no subscription for chat access (the catalogue assistants come with the Basic plan). There is a wallet: you put in a small sum and it is debited for what was actually done. Did not open it for a week, spent nothing.

What to set aside for the first evening. One request in the light mode is capped at sixty cents, and in practice usually comes out noticeably less. So a dollar or so covers an evening of experiments with room to spare.

Why that is better than a subscription at the entrance. One of the most common objections I hear: subscriptions are expensive and there is nothing to try several tools with. That is true — three subscriptions are debited in full whether you worked that month or not.

Here it is the reverse: you try, you pay for what was done, and a month of idleness costs nothing.

And if something did not work — money for what was not delivered is returned to your balance, and an empty answer is refunded in full.


What not to do on the first evening

Four things that will reliably turn the evening into nothing.

Do not register with five services. Every registration is twenty minutes and a new place to work out.

Do not try to understand how it is built. How it works inside is not your task. Yours is to get a result.

Do not take on something complex. "I will make a site" or "I will build a bot" is not a first task. The first must close in an evening.

Do not compare yourself with people it works faster for. One participant on our courses wrote: "I simply cannot keep up with trying everything that is given" — and that is a normal state at the entrance rather than a sign of incapacity. Keeping up with everything is not necessary.


What to do if the first task did not work

It happens. Three common causes and what to do.

The result is not right. Almost always because you described the task too briefly. Tell it more: who for, what for, what should come out. One sentence becomes a paragraph and the result changes.

It is unclear what to do next. Ask directly in the same window: "explain step by step what I should press now". That is a normal question and it will be answered.

You missed a few days and forgot it all. Do not start again with studying. Take the same task and repeat it — the second time it will take a third as long.

The general rule: if an evening gave no result, do not switch to another tool. Rephrase the task.


Where to start today

One action, fifteen minutes.

Do not open reviews and do not choose a service. Take a sheet of paper and write one line: which task do I have right now.

Not "I want to work out AI" but specifically: I need to write a greeting, I need to understand the manual for a washing machine, I need to restore a photograph.

That line is your whole first evening. From there you simply describe it in words and get a result.

Did it work? Take a second task of the same kind. Not a different one, the same type — the second time takes half as long, and it is on the second attempt that confidence appears.

Where to do it if you do not want to choose. There are ready assistants for specific everyday tasks: analysing a document, composing a letter, help with meals, getting a home in order. It opens with a link in a browser — on a phone too. No installation: you write the task in ordinary words.

Registration is free and opens three days of the Basic plan — enough to close three or four everyday tasks and understand whether you need this further.


What to take on the platform for this task

What to read next

How to write a greeting or a toast — the simplest first task of the three.

How to make sense of an appliance manual — the second task, and it is free.

Why people give up in the first week — if you have started before and stopped.

How to choose a model for a task — if you have already tried and are now choosing deliberately.


Also on AI SKILLS