How to Get Started with Vibecoding? Artificial intelligence will write the program and be very helpful with spreadsheets.
Vibecoding, or creating applications and automating tasks using artificial intelligence, is gaining popularity. It is a tool that speeds up the work of experienced programmers while also opening the door to creating functional applications for people with no knowledge of programming languages. However, it is worth noting that without programming experience, creating a commercial and secure solution from scratch using AI alone remains a significant challenge.
The true power of vibecoding lies elsewhere—in simplifying everyday, repetitive tasks, especially those performed in programs such as Excel or Google Sheets. Far more people use spreadsheets than create complex programs. Formulas and macros have long made it possible to significantly speed up work, but it is often not emphasized enough how AI can revolutionize these simple activities.
Creating a “program” sounds more appealing than saving a dozen or so hours of work each month, even though it is precisely these smaller improvements that accumulate into enormous savings in time and resources. After all, simple “programs” can be created in spreadsheets using advanced formulas.
Currently, it is possible to create fully fledged commercial programs without manually writing a single line of code, but this already requires some knowledge of software architecture, the fundamentals of security, and potential threats.
Table of contents
-
How to learn vibecoding most effectively from scratch? A list of stages.
-
How to start programming with the help of artificial intelligence
-
Next steps: Bolt, Figma, app prototypes, artificial intelligence writing in Python.
-
Basic tips for creating applications using artificial intelligence.
How to Learn Vibecoding Most Effectively from Scratch: A List of Stages.
Below is a universal learning path designed for non-technical people who have never programmed before and want to learn after work. It takes into account the gradual development of motivation through completing simple tasks that genuinely speed up work and are applicable in almost any office job.
Of course, there are many different paths for development. If someone is highly motivated and has the time and opportunity, the fastest way to master vibecoding is to learn the basics of programming (for example, in Python) and understand program architecture and technology stacks, using AI as support in the learning process. It is also useful to set up a coding tool (e.g. Copilot or Codex), which is practically essential when working with longer code and a larger number of files.
However, for most people, a useful intermediate step is creating advanced formulas or simple programs in Excel. This way, even if you stop learning at any point, the time you have invested will not be wasted.
Stage 1: Playing with ChatGPT
At first, it is best to simply experiment with generating simple programs and games in ChatGPT (Canvas), Claude AI, or Gemini. Based on a single command, you can obtain a working program within a few minutes that makes an impression; if you have an account with the aforementioned applications, you do not need to install anything. This way, you will learn about the capabilities of artificial intelligence. It is a good idea to improve such applications afterward, but with a larger number of modifications and specific requirements, you may hit a wall where AI is unable to write a functioning program.
Stage 2. Creating Macros and Mini-Programs in Spreadsheets
Spreadsheets are a versatile tool that can be useful at work, in education, and in everyday life—from personal finances to trip planning. Artificial intelligence makes it easier to create formulas, while a clear cell layout helps you better understand how simple programs work. It is worth exploring their full potential because, beyond summing data in Excel, you can build complex solutions and also use macros based on programming languages.
Stage 3 (optional): creating simple websites (business card sites), writing CSS code for WordPress websites.
Artificial intelligence makes it possible to quickly create a website from scratch and often publish it immediately or download it to your own server. Although such websites have limitations and are more difficult to expand, they work very well for simple projects or as an alternative to presentations. In an existing WordPress site, AI can help create additional CSS code or simple plugins. Websites are needed for many projects today, although their use is not as widespread as Excel.
Step 4 Installing Python, simple programs run from disk
Installing Python is simple, although it can sometimes be problematic. It enables you to create simple programs and websites without additional installations, albeit with limitations. Running code locally gives you full control and independence from online environments, while working with files teaches us about their structure and the location of resources. With Python, you can create desktop applications, for example, for taking notes, organizing folders, or collecting data from the internet. ChatGPT can generate an entire zip folder containing multiple program files.
Stage 4.5 – coding tool
Language models handle longer texts less effectively, while many programs consist precisely of extensive files or sets of files. That is why dedicated programming tools, such as Codex or Copilot, operating within code editors (e.g., Visual Studio), significantly facilitate the work.
Stage 5: A polished look for the application
Bolt.new and Figma Make enable the creation of consistent, comprehensive application visualizations. Figma Make stands out for its high editing precision. Both tools are available for free with daily and monthly limits. A back end—i.e., databases and program logic—can later be added to the front end created in this way.
Stage 6 Application structure, logic, and code execution.
At this stage, planning the application, writing descriptions, mapping out the logic, and organizing tasks becomes essential. We need to have a complete plan for how it will work and look, both in our heads and in our notes; know the basic “building blocks” that make up programs; and understand how the code works. Even without programming skills, it is worth knowing what is responsible for each element. This way, instead of writing “fix this function,” we can precisely indicate “improve this code.” This means that, in addition to working with a language model that generates code, we need to spend time preparing the application design.
How to Start Programming with Artificial Intelligence
Step one: create your own game or another simple program in ChatGPT, Claude, or Gemini. You can run such a project directly in the app, without installing any additional software. It’s a good way to discover AI’s capabilities in programming and find motivation to continue learning. Although creating complex formulas and commands in Excel may be more practical for most people, designing games is simply more enjoyable. In a similar way, you can also prepare practical solutions, such as dashboards or preliminary visualizations of applications.
.
This program should be launched in Canva,
."""
Then click “Run code” (top right).
.
In this way, we obtained a very interesting simulator of a game known as life.
."""
In this way, you can create many simple games and programs, such as: Snake, Pong, mazes, etc. More practical solutions include notebooks, to-do lists, calculators, and converters. We can also ask for data visualizations in the form of an aesthetically pleasing table, timeline, etc.
What can’t be done this way? All more complex programs and 3D games require a more advanced approach, and cannot be created quickly in ChatGPT, Claude, or Gemini. Programs run directly in ChatGPT do not have their own memory, meaning that after you close the program, everything you did there disappears. This can be worked around to some extent by creating a file upload and download module.
Vibecoding in Practice: Start with Spreadsheets
The Simplest Scenario
The practical application of Veicoding is primarily creating formulas in spreadsheets. More complex spreadsheets function like simple programs. With the right formulas, a task that previously took hours can be completed in just a few minutes. Often, simply knowing that something can be done is enough.
Example: You have a column containing first and last names, and you want to sort all the data alphabetically by last name, so you need to split this column into two, with one containing the first name and the other the last name. If the list has, for example, 1,000 entries, doing this manually makes no sense. You can therefore look for formulas. If the problem is simple, you may be able to find a suitable solution on Google.
.
However, in the case of more complex commands, finding an answer online is more difficult. Sometimes, a command copied exactly as it is does not work either. The language used is often the reason. Formulas in Excel or spreadsheets look slightly different in Polish and other languages; sometimes a dot needs to be replaced with a comma, or a semicolon with a comma.
In such cases, ChatGPT is invaluable. Language models handle spreadsheet problems very well. We can describe our problem to them or, for example, paste the entire table or a screenshot (being mindful of sensitive data if the model is not deployed locally).
.
Thanks to AI, it is easy to find out that the problem can be solved using the SPLIT function. Similarly, two columns can be combined into one. The key is knowing that such simple solutions exist.
The Snowball Effect When Using AI
However, it is not always that simple. Before the era of widespread artificial intelligence, working with advanced formulas could be time-consuming and frustrating, especially when adapting formulas from other languages (e.g., changing semicolons to commas).
It was often jokingly said that “it is always worth spending 3 hours automating a task that would take 2 hours to do manually.” Of course, if the task was to be repeated many times, such an investment paid off. And this is precisely where artificial intelligence offers its greatest advantage: thanks to it, an automation process that would previously have taken 3 hours may take as little as 15 minutes.
Many people are unaware of the huge savings this generates, especially in terms of the snowball effect. Creating useful tables faster (and often making them possible at all) enables better analyses and more effective strategies.
AI helps right from the very beginning.
If you’re not sure whether a given task is easy to automate, ask AI. You can describe the problem to ChatGPT, paste in an excerpt from a document (remembering to protect sensitive data), or even share a screenshot. It’s also helpful to ask how an existing table could be improved.
If you have paid Gemini subscriptions (e.g., in Google Workspace), or use a paid Office package with Microsoft Copilot, you can use AI directly in spreadsheets.
.
You can also use various free extensions, such as ChatGPT for Excel or Gemini for Google Sheets. It should be remembered that these are third-party solutions. However, simply having a conversation with ChatGPT and pasting formulas is still a huge improvement in quality compared with searching for advice on Google.
AI in Spreadsheets – Practical Tips.
Simple Functions
When merging/splitting columns, a simple command is all we need.
.
A simple query in ChatGPT allows us to achieve a satisfactory result.
. .
In this case, artificial intelligence was not essential; anyone can search for this function on Google.
Gantt Chart
The true power of AI is revealed in more complex commands, where we cannot find an exact function to paste.
For example, let’s assume that we want to create a Gantt chart in a spreadsheet, but we don’t want to use the Insert Chart function; we want the chart to be directly in the table.
.
Chat indicated a simple solution. First, we need to create a formula and paste it into every empty chart field. The formula’s task is to insert a character in the appropriate place, and then conditional formatting completes the effect.
Creating Charts with AI
As a general rule, for more complex charts, it is better to use dedicated solutions. You can also generate chart code in ChatGPT, which can then be run using Python. On the other hand, an Excel chart can often be more convenient. You can also use Canva in ChatGPT, Claude, or Gemini, in which case you immediately get a convenient preview. If you want, you can achieve an interactive effect, whereby the program displays additional information when you click on a given element.
."""
Accessibility Planning
Planning availability can often be challenging. When the operating model is fixed and predictable, it is best to use a dedicated solution. However, implementing a new program takes time, and it is necessary to check whether it will be suitable in a particular case. For newly forming, small teams, spreadsheets are often more convenient, although they have some drawbacks—anyone can modify anything, so one employee may accidentally mess up the entire spreadsheet or sign someone else up, creating chaos. There are ways to secure spreadsheets, but in such cases it is more convenient to use dedicated software.
Attendance planning is a good example of a simple program that can be created in spreadsheets. This particular example probably won’t be useful to you in practice anyway. But it shows that you can create various programs in spreadsheets, such as simple task systems, inventory tracking, and tables for managing finances. And thanks to artificial intelligence, you can do it fairly quickly, even if you don’t know Excel.
I created the table below entirely using ChatGPT. It first generated a file for me, but the functions in it did not work, so I had to ask it to fix them in a separate prompt. When creating the table, I already had the entire plan in mind: the program was supposed to create a consolidated schedule based on the availability marked by each user. By entering “1” in the appropriate cell, the user indicates when they are available.
. .
This way, we obtained a convenient table where each employee can enter their availability, and we have a consolidated list in one place.
ChatGPT can create a functioning table, albeit a little less convenient, but much simpler to create.
Command: design a table for planning availability in the schedule. Employees have flexible hours and enter when they are available. I need a single table where I can see everything at once, namely how many people have signed up for a given hour.
After additional general commands (I want to see this in one place; I want to see who is available on a given day at a given time), Chat came up with a useful table.
.
Spreadsheet Capabilities
When it comes to artificial intelligence, our main limitation is imagination (or, more precisely, knowledge of what is possible). Most people stop at the stage of a table that sums up costs. But Excel’s capabilities are much greater.
In spreadsheets, you can create fully-fledged programs or even games. Although beyond a certain level of complexity this no longer makes much sense, it demonstrates the range of possibilities. Many large companies still rely on Excel (which leads to certain problems, for example when scaling).
General rule:
Spreadsheets are useful for creating simple applications in which every user can see the entire code, all functions, and all data. It is sometimes possible to restrict visibility and access to elements, but this is not the most convenient solution.
Advantages of sheets:
- We have a great deal of control over the program’s appearance and functionality. It is often easier to create a simple table with a few functions than to test a dozen solutions. Nowadays, virtually everyone has access to spreadsheets, so we can be sure that many people can use them. In the case of dedicated solutions, even if they have significant features, they may not be available to a larger number of people (or may be paid solutions).
- We can easily add new features.
- You do not have to be a programmer to modify the program; a minimally technical person can even easily teach a completely non-technical person how to modify it.
Examples of applications.
Learning – In cases such as a list of characters, a vocabulary list, or a list of events, tables are invaluable.
Notes – Thanks to the cell merging and splitting functions, as well as “tabs”—that is, multiple worksheets in a workbook—Excel is convenient when we want to create simple dashboards or, for example, product cards. It therefore serves as a note-taking tool.
Planning – checklists, charts, or time-based tables make it easier to plan and coordinate tasks. This is often more convenient than, for example, Google Calendar.
Finance and analysis – that is, the original, primary use of spreadsheets. Using spreadsheets, we can track and plan a household budget. In tables, we can analyze a company’s sales, control expenses, and determine strategies (e.g., by analyzing target groups). With a few tables, we can create a simple CRM.
Division by functions:
- Sorting. One of the more important advanced functions, especially when we take multi-level sorting into account. This gives us complete control over the data, and we also have control over the tags. A major problem when creating traditional lists, as well as folders, is when a given person belongs to several categories. Thanks to tables with sorting, we can easily obtain, for example, a list of people who have a driving licence and know how to weld, a list of people with a driving licence who know how to install electrical systems, and people with welding and electrical skills. At the same time, we retain the main list, so that adding or removing a person from a given set, or modifying data, is done in one place.
- Counting – together with sorting, this allows us to perform quite complex operations.
Macros, scripts
Spreadsheets allow us to create macros. Macros are simple mini-programs that speed up working with the program. We can write them from scratch, or record a list of steps when performing repetitive tasks.
.
If something cannot be done using formulas, simple programs can help. For example, it is not possible to link cells together using formulas. If we want two cells to have the same content, we can use the formula =(A1) in cell A2; then, when we edit cell A1, we also edit cell A2. However, making this work in both directions is impossible. There are various third-party solutions and paid extensions, but if we only need this function, we can use a simple program written entirely by ChatGPT.
.
Practical application:
We have a large table with several dozen columns. Therefore, we are unable to display all the fields at once. This can be addressed using the Transpose function:
.
But if we want a more convenient solution, we can use a simple script. It was written entirely by ChatGPT.
.
This way, I created a simple form for managing the table. In cell C4, I enter the row I want to modify, and the changes are saved automatically. This is, of course, an initial version; I can easily add searching by surname, moving right/left, etc. Ultimately, each row will have a corresponding entry in a separate sheet.
This functionality is offered by a number of database programs, such as Microsoft Access, but learning a new program always takes more time, and such spreadsheets are also easier to share.
.
Further stages: Bolt, Figma, app prototypes, artificial intelligence writing in Python.
The next step in vibecoding is creating standalone applications. However, this is more difficult than creating formulas/simple macros that work in Google Sheets. New problems arise, related to the context window. AI can create a nice, functioning application with a single prompt, but then changing a small element may require many prompts, and sometimes fixing them with AI is virtually impossible, making manual code corrections necessary.
.
At this stage, you need to start taking security into account, but as long as we are creating applications for our own use, we do not need to worry about it too much, although it is a good idea to be familiar with the basic threats from an early stage. Particular care is needed if we use API keys for certain solutions or create applications in which users are required to log in and provide a password. In such cases, it is best to consult an actual programmer about how the application works, while the absolute minimum is a general understanding of how security measures work.
It is also essential to understand how the program works, know which code is responsible for what, what specific files do, etc. In the case of larger problems, it is often more useful to write: fix this piece of code (and other pieces that refer to it), rather than: fix this function that the user sees.
At this stage, it is also important to create a programming environment by connecting a language model to a code editor, such as Visual Studio. This allows artificial intelligence to modify multiple files simultaneously.
Basic tips for creating applications using artificial intelligence.
Save good versions
In theory, everything is saved automatically. In practice, if we have, for example, 100 versions, finding the working one is not very convenient. Therefore, good versions should be labeled appropriately. We can copy the code or record the numbers of the good versions.
If something cannot be fixed in 3 commands, you should roll back.
If the problem persists despite a dozen or so commands requesting a fix, it means we are wasting time. Sometimes the problem can be fixed with the fifth command, but generally, after 2–3 attempts, it is best to revert to a working version.
One command – one change
If we want to do many things at once, artificial intelligence gets confused, and the program does not work.
Different Models
Good results can be achieved by using different models when we encounter problems. Sometimes GPT 5 works better, and sometimes Claude 4.0 Opus.
If something cannot be fixed, try modifying something else
If, after several rounds of fixes involving commands that change nothing and reverting to a working version, there are still no results, it is often effective to ask for something else to be changed. If the artificial intelligence manages to do that, you can return to fixing the main problem, and this is often more efficient. Of course, this only makes sense when the program can run.
Proper Documentation
The foundation of writing code with artificial intelligence is creating good documentation. A thorough description helps the language model write better code. Additionally, it makes it easier to make manual corrections or rewrite everything properly from scratch using a ready-made prototype.
Summary
Vibecoding opens up the world of programming and automation to everyone, even without technical experience. Although creating a fully commercial application from scratch remains a challenge, the true power of this method lies in something much more accessible: streamlining everyday, repetitive tasks.
It is precisely in spreadsheets, where advanced formulas and simple macros created with the help of AI can save dozens of hours, that the greatest and fastest benefits can be seen. Small automations accumulate, leading to enormous savings in time and resources.
The learning path presented—from simple experiments, through practical applications in Excel, to creating websites and programs in Python—allows you to gradually build skills and motivation. With this approach, every hour invested brings real and immediate benefits, making learning practical and effective from the very beginning.