<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/">
<channel><title>Articles on presentations, websites and automation — Zakharov Design</title><link>https://zakharovdesign.cc/materials/</link><description>Practical articles on how to build presentations, websites and automations with the help of neural networks, and when to bring in a professional.</description><language>en</language><atom:link href="https://zakharovdesign.cc/materials/feed.xml" rel="self" type="application/rss+xml"/><lastBuildDate>Thu, 29 Oct 2026 06:00:00 GMT</lastBuildDate>
<item><title>How to save time by building an automation for yourself</title><link>https://zakharovdesign.cc/materials/avtomatizaciya-dlya-sebya/</link><guid isPermaLink="true">https://zakharovdesign.cc/materials/avtomatizaciya-dlya-sebya/</guid><pubDate>Thu, 29 Oct 2026 06:00:00 GMT</pubDate><dc:creator>Nikita Zakharov</dc:creator><category>Automation</category><description>How to automate a repetitive task with an AI model: which tasks suit it, where to start, how to check the result and when to turn to a specialist.</description><content:encoded><![CDATA[<h2>What personal automation is and why bother</h2><p>It is a small helper for one task of yours: a script that moves data from emails into a spreadsheet, a bot that sends reminders, a formula that builds a report on its own. It has a real strength: you know your task better than any developer, and the AI turns your description into a working tool, so you do not have to write the code yourself.</p><p>It used to take a developer and a quote. Now you can build a trial tool in an evening and test it on real data straight away.</p><h2>Why “one prompt and it all works” stays a myth</h2><p>An AI does not know which of your emails turn out strange, what to do when a file has an empty row and who is to blame if the tool quietly deletes something important. All of this comes out only in practice, after long testing and patient refinement.</p><p>The better you understand your process and the more carefully you check the result, the more reliable the tool becomes. Experience matters here more than clever wording in the prompt.</p><h2>What AI gives you and what a specialist gives you</h2><p>AI wins on speed and price. You get a working version quickly, can try different approaches and do not have to explain the process to an outsider. For personal tasks this is an excellent route.</p><p>A specialist wins where reliability matters. They think through unusual cases in advance, handle data carefully, connect email, spreadsheets, payments and a CRM, set up backups and answer for it all working when you are not around.</p><h2>Which tasks can be automated</h2><p>People usually start with tasks like these:</p><ul><li>collecting data from several sources into one spreadsheet;</li><li>sorting incoming emails and files into folders;</li><li>preparing a weekly report from the same export;</li><li>sending reminders and notifications;</li><li>renaming, converting or processing a batch of documents.</li></ul><p>A simple sign of a suitable task: you catch yourself thinking “not this again”. What fits is something that repeats, is done in the same steps and has a result that is easy to check.</p><h2>Where to start: one task and a written list of steps</h2><p>Do not automate everything at once. Pick one job that repeats, is done the same way each time and gives a clear result. If you are unsure, spend a week noting what your time goes on.</p><p>Before asking for anything, do the task by hand once and write down the steps: what you receive, what you do and what comes out. This record becomes a ready description for the AI and also a checklist against which you will later check the result.</p><h3>A weak and a strong prompt</h3><p>A weak prompt: “Automate my reports for me”.</p><p>A strong prompt (an illustrative example): “Every Friday I build a weekly summary spreadsheet from clients’ emails. Each email has an Excel table attached with the same columns. Here are my steps: (list). Here is a sample input file and a sample of the final table. Write a Google Sheets script that does this and explain in simple words what it does at each step. I am not a programmer. If a file of a different kind turns up, do not guess, show an error”.</p><p>The strong prompt has a task, steps, examples of input and result and a rule for unusual cases. So the model guesses less and you have less to correct.</p><h2>What to watch for along the way</h2><p>A few habits that noticeably improve the result:</p><ul><li><strong>Pick the tool where you already work.</strong> If you live in spreadsheets, make it a spreadsheet script. If you live in email, make it an email rule. There is no need to learn a new environment.</li><li><strong>Order first, program second.</strong> If the task looks different every time, first put the process itself in order. Automating a poorly thought-out process gives the same mistakes, only faster.</li><li><strong>Test on more than one example.</strong> Take a typical case, an edge case (an empty field), a strange one (an extra column) and a case where nothing should happen. One successful run proves nothing.</li><li><strong>A copy before changing anything.</strong> If the tool changes or deletes files, it should save a copy first.</li><li><strong>No passwords in the chat.</strong> Do not paste passwords, access keys or clients’ personal data into an AI.</li><li><strong>Ask for explanations.</strong> Have the model say in plain words what each step does. That way you will notice if the tool does the wrong thing.</li></ul><p><strong>The first-version rule</strong> Automate what you already do by hand consistently. If the task looks different every time, first put the process itself in order. Automating a poorly defined process produces the same mistakes, only faster.</p><h2>The overall plan</h2><p>Put together, it looks like this:</p><ol><li>Pick one repetitive task.</li><li>Do it by hand and write down the steps.</li><li>Check whether a ready solution already exists in the services you use.</li><li>Ask the AI to build a small tool from your record.</li><li>Test it on five different examples: what you fed in, what you expected, what you got.</li><li>For the first weeks, look through the results yourself.</li><li>After a month compare two numbers: how much time it took before and how much now, counting the time spent on repairs.</li></ol><h2>Can you do it yourself</h2><p>You can, and for personal tasks it is a good path. If you understand your work well and are used to checking everything, AI will be a real helper. Anyone can learn this, just like any other craft.</p><p>But it takes time and attention, often more than it seems. A tool breaks when the format of an email changes, a service updates or an unusual file arrives. If you are a specialist in another field, you will subtract those hours from your main work, and the attempt to save money will turn into a loss.</p><p>It is worth turning to a specialist if:</p><ul><li>the tool is used not only by you but by a team or clients;</li><li>money, clients’ data or important documents pass through it;</li><li>it has to run every day without your involvement;</li><li>a mistake would be costly.</li></ul><p>In such cases you need a reliable system: thought through, tested, with backups and a clear way to repair it. I use AI as a tool for speed, but I answer for the result myself.</p><p>Tell me what you do now and how. I will reply with what can be automated and how long it takes.</p>]]></content:encoded></item>
<item><title>How to make a good presentation with Claude</title><link>https://zakharovdesign.cc/materials/kak-sdelat-prezentaciyu-s-claude/</link><guid isPermaLink="true">https://zakharovdesign.cc/materials/kak-sdelat-prezentaciyu-s-claude/</guid><pubDate>Thu, 29 Oct 2026 06:00:00 GMT</pubDate><dc:creator>Nikita Zakharov</dc:creator><category>Presentations</category><description>How to make a presentation with an AI model: what it takes off your hands, where to start, what to watch for and when to turn to a designer. With examples of slides made from a single prompt.</description><content:encoded><![CDATA[<h2>Why use AI for a presentation</h2><p>A presentation is not there for the sake of slides. It helps you get one idea across to your listeners and nudge them to act: agree, buy, remember. AI takes on what eats the most time: building a structure, finding wording, suggesting headline options and cutting long text down to a couple of lines.</p><p>Its main advantage is speed and the freedom to experiment. A draft with a plan and text for every slide appears in a few minutes, and you can ask for three different approaches and choose the best without spending a day on each. For work meetings, lectures and internal talks this is often enough.</p><h2>Why “a presentation in two hours” stays a myth</h2><p>An AI does not know what exactly you want to say to the room, cannot feel how much information a listener can take and does not see that a slide looks like a thousand others. Every answer differs a little from the last, and choosing the best one is up to you.</p><p>So a good result depends on experience, careful checking and several rounds of edits. The better you understand what a strong presentation looks like, the more precisely you ask and the less time goes on the unnecessary.</p><h2>What AI gives you and what a specialist gives you</h2><p>AI wins on speed and price. It quickly sketches a structure, helps you word an idea and suggests what can be cut. For yourself it is an excellent helper for getting to grips with the material and seeing a first version.</p><p>A designer wins where how the presentation is received matters. They build a single visual system, choose real photos and charts, make sure each slide has one main thing and take responsibility for the result. For investors, a client, a big conference or a lecture that will be shown many times, a presentation works as part of your reputation.</p><h2>Which tasks you can handle yourself</h2><p>AI does well with:</p><ul><li><strong>A working presentation</strong> for a meeting or your team.</li><li><strong>A talk or lecture</strong> for a small audience.</li><li><strong>A student project defence.</strong></li><li><strong>A draft for a designer.</strong> A plan and text for every slide that you can later hand to a specialist. It saves time for both of you.</li></ul><p>But an investor pitch, a proposal to a big client or a presentation that a deal depends on is better not built by feel.</p><h2>Where to start: the idea, the material and the first prompt</h2><p>A presentation most often turns out weak because the author did not answer three questions in advance: who will listen, what the listener should remember and what they should do afterwards. Write the answers in your own words in three to five lines and paste them into every prompt.</p><p>Give the model material: notes, figures, earlier texts. The more real material in the prompt, the less it invents on its own.</p><h3>A weak and a strong prompt</h3><p>A weak prompt: “Make a presentation about communication in a team”.</p><p>A strong prompt: “You are helping me prepare a 20-minute talk for department heads who complain that their teams do not understand each other. The main idea: communication breakdowns are more often about how processes are set up than about people’s characters. I need 12 slides, each with a headline that is a statement and no more than three lines of text. Use only the facts from my notes below and add nothing of your own. First give me a plan, then the text for each slide”.</p><p>Ask for the text first, not a finished file. Reading a plan and headlines is quicker than editing a finished design. If the headlines, read one after another, retell the whole talk, the structure works.</p><h2>What to watch for along the way</h2><p>A few habits that noticeably improve the result:</p><ul><li><strong>A statement headline instead of a topic.</strong> “Communication breaks down because of processes” says more than “Causes of problems”.</li><li><strong>One idea per slide.</strong> If a headline has an “and”, it may be two slides.</li><li><strong>A slide for the room, not for reading.</strong> What stays on screen is what the listener should remember; the rest you say aloud.</li><li><strong>Your own colours and font.</strong> One main background, one accent colour and one font that supports your language. Brand colours beat those the model picks.</li><li><strong>Rehearsing aloud.</strong> Slides that look good on a laptop screen look different on a big one.</li><li><strong>A sceptical listener.</strong> When the draft is ready, ask the model to play a strict listener and find the weak spots.</li></ul><p><strong>The three-second test</strong> Show a slide to a colleague for three seconds, then hide it. If they can say what it was about, the slide is fine. If not, it has too much or too little of the main point.</p><h2>What one request produces: a comparison</h2><p>We ran a simple experiment: we gave two models one short request to make a slide about the causes of communication breakdowns. On the left is the result untouched, on the right a slide from a real lecture. They play different roles, so we are comparing the approach, not the content.</p><p>The slide on the left is tidy: the text is correct, the idea is clear and there is a structure. But it looks like a thousand others. Four identical cards, emoji icons of different styles and sizes, and the text in the last row butts against the edge of the frame. On the right there is one thesis, a large headline and a photograph, and the design matches every other slide of the lecture.</p><p>A stronger model made a neater slide: no emoji, free space and order. But the layout stayed generic, with no style of its own and no chosen imagery.</p><p>When we attached example slides, the colour changed but the layout fell apart: the headline ran onto the cards and the lines overlapped. The reference helped with colour but not with tidiness. Such a slide cannot be shown without more work, and only someone looking carefully will notice.</p><h2>The overall plan</h2><p>Put together, it looks like this:</p><ol><li>Decide who is listening and what they should remember.</li><li>Gather material and make a plan of 8 to 12 slides.</li><li>Ask for the text of each slide with statement headlines.</li><li>Read only the headlines: they should retell the whole talk.</li><li>Only then move on to design: colours, font, photographs.</li><li>Show the slides to a friend for three seconds and run through the presentation aloud.</li><li>Before presenting, check the screen format, embed the fonts in the file and save a PDF copy.</li></ol><h2>Can you do it yourself</h2><p>You can, and that is fine. If you have a trained eye and an understanding of how a talk is built, AI will be a good helper and the presentation will be decent. Anyone can learn this, just like any other craft.</p><p>But it takes time and a lot of attention: to the idea, to details, to how everything looks on a big screen. And there is a flip side. If you are an expert in another field, say a doctor, a lawyer, an entrepreneur or a teacher, the hours spent on slides are taken from your main work. Trying to save on a designer, it is easy to lose more than they would have cost.</p><p>So for working talks and for testing ideas, build it yourself. But when a presentation affects a deal, money or reputation, it is better to hand it to someone who does this every day. I use AI myself for speed, but the decisions about the idea, composition and detail are mine.</p><p>Tell me about the task and the deadline. I will reply with how we will do it and how long it takes.</p>]]></content:encoded></item>
<item><title>How to build your own website with vibe coding and avoid slop</title><link>https://zakharovdesign.cc/materials/sajt-vajbkodingom-bez-slopa/</link><guid isPermaLink="true">https://zakharovdesign.cc/materials/sajt-vajbkodingom-bez-slopa/</guid><pubDate>Thu, 29 Oct 2026 06:00:00 GMT</pubDate><dc:creator>Nikita Zakharov</dc:creator><category>Websites</category><description>How to build a simple site with an AI model: what it suits, where to start, what to watch for along the way and when to hire a specialist. And what slop is, shown on real examples.</description><content:encoded><![CDATA[<h2>What vibe coding is and why it is useful</h2><p>Vibe coding is a way of making sites where you do not write the code yourself but explain to an AI in plain words what you want to see. You ask for “a page with a headline and an enquiry form”, look at the result and fix it step by step: a bigger button, a lighter background, the pricing block lower down.</p><p>What this approach gives you above all is speed and a low barrier to entry. It used to be that to see your idea as a working page you had to find a developer or spend months learning layout. Now the first version appears in an evening: you can show it to friends, check whether the idea is clear and only then decide what to invest in.</p><h2>Why “a site in two hours” stays a myth</h2><p>The internet is full of stories about someone building a site in an evening with one prompt. They are not made up: a draft really does appear fast. But a draft and a site you are not embarrassed to put in front of clients are different things, and the road between them takes most of the time.</p><p>The AI does not know your audience, does not see that a button has slid off the edge of the screen on a phone and will not notice that the discount promised in the headline is not something you offer at all. Every new prompt changes the result slightly differently from what you expected, and choosing the better variant is left to you. So experience and patient checking decide everything: the better you understand what a good site looks like, the more precisely you ask and the less time goes on dead-end edits.</p><h2>What AI gives you and what a specialist gives you</h2><p>Compared honestly, each path has its strengths. AI wins on speed and price: you get a draft quickly, can try as many variants as you like and do not have to explain your idea to a stranger. For yourself it is an excellent way to work out what you actually need.</p><p>A specialist wins where the result matters more than the process. They see at once which doubts of a client the page has to answer, know how it will behave on different devices, choose real photos and fonts instead of default ones and answer for the outcome. You spend your time on a conversation about the task rather than on dozens of edits.</p><h2>Which tasks you can handle yourself</h2><p>If you want to try, start with one of these:</p><ul><li><strong>A personal page or business card.</strong> Who you are, what you do and how to reach you.</li><li><strong>A page for one event.</strong> A webinar, a meeting, booking a consultation.</li><li><strong>A prototype.</strong> A quick version on which you test an idea before investing in a real site.</li><li><strong>A small portfolio,</strong> if you already have something to show.</li></ul><p>An online shop, a personal account, payments, or a site that brings the business its main enquiries is better not built by feel. A page can look good and still lose enquiries, open slowly on a phone or keep other people’s data where everyone can see it.</p><h2>Where to start: the goal and the first prompt</h2><p>The most common beginner mistake is the prompt “make a beautiful site”. Beautiful for whom, and why? Until you have answered that yourself, the AI will give an averaged result: it is built to answer with the most likely thing, and the most likely thing is exactly what is called slop.</p><p>So before the first prompt it is worth answering a few questions for yourself. What one action should a visitor take: write, book or call? Who is this person and what worries them? Why should they believe you? The answers will form the basis of your prompt.</p><h3>A weak and a strong prompt</h3><p>A weak prompt: “Make a beautiful first screen for a psychologist”.</p><p>A strong prompt: “Make the first screen of a personal page for a psychologist who works with adults who are tired and do not know where to start. The tone is calm, with no promises of results. I need one headline, one sentence under it and a “Book a consultation” button. Leave empty space on the right for my photo. Do not add reviews, prices or numbers that I did not state”.</p><p>The strong prompt has an audience, a tone, a structure and a ban on invention. So the model makes fewer random decisions and you have fewer edits to make.</p><h2>What to watch for along the way</h2><p>A few habits that noticeably improve the result:</p><ul><li><strong>Copy first, design second.</strong> A site stands on its words. If you first make a beautiful empty page and then drop text into it, it will stay faceless.</li><li><strong>One block at a time.</strong> Bring the first screen to a good result and only then move on to the next block. A request for the whole page at once gives a mountain of code in which it is hard to fix anything.</li><li><strong>Your own material instead of invented material.</strong> Give the model your notes, real phrases from clients and your own photos. If you have no reviews, do not ask it to make them up.</li><li><strong>A reference.</strong> Attach a screenshot of a site you like or describe it in words: “lots of free space, one accent colour, large photos”. The result changes noticeably.</li><li><strong>Design rules.</strong> Write down two or three colours, two fonts and a set of spacings once and ask the model to stick to them. Then the site looks coherent.</li><li><strong>A phone.</strong> Open the page on a real phone: most visitors will come from there.</li><li><strong>Versions.</strong> Save a working version before every big edit so there is somewhere to go back to.</li></ul><p><strong>Empty space beats invention</strong> If you have no testimonials, do not ask the model to write them. If you have no exact price, do not write &quot;from $1&quot;. A blank space or an honest line such as &quot;the price depends on scope&quot; is better than a plausible lie. Visitors notice such a lie, and sometimes you have to answer for it.</p><h2>What slop looks like: a comparison</h2><p>We gave two models one short request with no reference and no material: make the first screen of an online dog-training course. On the left is the result untouched, on the right the first screen of a real course landing page made by hand. Drag the slider to compare.</p><p>On the left the signs of slop are obvious. The AI came up with “42,000+ dogs trained”, “4.9/5” and “15 minutes a day” on its own, though we never asked for that. There are no real photos, the dog is a cartoon in a circle, and the headline would suit any course. There is a simple test: cover the name and the logo and ask yourself whose site this is. If the pictures and promises do not tell you, you are looking at slop.</p><p>A smaller model, given the same request, drew an interface mock-up instead of a dog and added figures of its own. The screen is tidy, but it says nothing about your course: the model does not know what you want to say.</p><p>When the prompt includes real photos and a reference, the result is noticeably better. But it is still not ready: elements overlap, spacing is uneven and the model again added figures nobody gave it. All of this can be fixed if your eye is trained, and that is exactly where the time goes that “two hours” does not cover.</p><h2>The overall plan</h2><p>Put together, it looks like this:</p><ol><li>Define the purpose of the page and the action you expect from a visitor.</li><li>Gather material: copy, photos and three to five sites you like.</li><li>Write down the design rules: colours, fonts, spacing.</li><li>Ask for one block at a time, starting with the first screen, and check every step.</li><li>Open the result on a phone, press every button and send yourself a test enquiry.</li><li>Read the copy aloud and remove everything that is not true.</li></ol><p>That is the working minimum. After it come the finer points: loading speed, image sizes, readability, settings for search engines. These can be learned too, but with every item the time grows as well.</p><h2>Can you do it yourself</h2><p>You can, and that is perfectly fine. If you have a trained eye and an understanding of how a site should look and work, AI will be a good helper and the result will be decent. Anyone can learn this, just like any other craft.</p><p>But it takes time and a lot of attention: to the text, to details, to how everything looks on different screens. And there is a flip side. If you are an expert in another field, say a lawyer, a doctor, a coach or a business owner, every evening spent on the site is taken from your main work. Trying to save on a specialist, it is easy to lose more than they would have cost.</p><p>So for interest and for testing ideas, build it yourself and enjoy it. But when enquiries and reputation depend on the site, it is better to hand it to someone who does this every day. I use AI myself: it speeds up the work, but the decisions about text, composition and detail are mine.</p><p>Tell me about the task. I will reply with how we will solve it and how long it takes.</p>]]></content:encoded></item>
</channel>
</rss>
