🌐 🇵🇱 Polski · 🇬🇧 EN
As a specialist dealing with automation and artificial intelligence on a daily basis, I constantly monitor the boundaries of modern language models' capabilities. Instead of theorizing about AI's potential, I prefer to test tools in extreme production scenario conditions.
This time, I set a clear task for myself: to check if the software development process (from concept to deployment) can be fully automated with minimal human involvement. I set a rigorous criterion – the application was to be created using a maximum of 5-6 precise prompts.I used the Codex model for the test, providing it with access to a local development environment. The result of this experiment is Master of Keyboard – an advanced tool for learning touch typing.
The operation of the finished system can be seen in the video below:
Experiment Architecture and Initial Prompt
I began the experiment by defining the architecture and functional requirements in one condensed prompt:
Build an application that supports touch typing learning. I want a beautiful interface, exercises, games, a highlighted keyboard, hand positioning hints, and typing speed measurement.
The first iteration defined the basic architecture and data structure. The key to success was not the one-time code generation, but the model's ability to work in a feedback loop. In the next 4-5 optimization steps, my role was limited to the roles of Product Owner and system architect – I defined goals, and AI worked as an autonomous engineer.
Codex took full responsibility for the development pipeline:
The final result is a fully scalable web application that in many ways surpasses commercial subscription-based solutions available on the market.
1. Telemetry Module and Live Analytics
The system processes keyboard input events in real-time (classic event listeners optimized for latency) and calculates key efficiency metrics:
Instead of monotonous text repetition, a game salon was implemented (including catching falling letters, destroying objects marked with characters, and a rhythmic marathon).
From a technical point of view, the most interesting is the adaptive difficulty level algorithm. AI implemented logic that analyzes the user's current error rate – if accuracy drops, the game loop slows down, minimizing frustration and consolidation of bad motor habits.
3. Interface and Micro-Course System
The application includes full support for muscle memory mechanics. The visual interface maps base positions (bumps on F and J keys) and dynamically highlights the corresponding keyboard areas and fingers to use for the movement.
Additionally, in just a few iterations, the model implemented 6 complete graphic themes (including Dark Mode, Neon, and Twilight), taking care of the proper contrast ratio in accordance with WCAG guidelines.
DevOps Automation and Infrastructure
What differentiates an advanced AI agent from a regular code generator is the ability to operate on infrastructure. In this experiment, the model independently managed the runtime environment:
Codex had no problem with key configuration, deployment to a remote device on the local network (Raspberry Pi), and proper preparation of the structure for static hosting on GitHub.
The project is fully public and available for testing at:
Conclusions from the Experiment: The Dawn of Intentional Programming
This test proves that we are entering the era of intentional programming (Intent-Driven Development). So far, building dedicated educational tools has required purchasing expensive SaaS licenses or engaging human resources for many working hours.
Today, as engineers, we receive technology that allows for the radical democratization of software creation. If any application on the market does not meet your expectations – whether in terms of aesthetics, ergonomics, or missing functionality – you can prompt, test, and deploy your own alternative in one afternoon. Such solutions are also an ideal way to draft a project, test its operation, show it to a test group, and collect feedback.
This was just the first in a series of automation tests I have planned for the near future. The next projects will cover even more complex system architectures.
I encourage you to follow my blog, where I will regularly publish technical case studies from the intersection of process automation and AI engineering. A broader perspective is coming – stay up to date!
Codex took full responsibility for the development pipeline:
- Static and semantic analysis of the existing code base,
- Implementation of business logic and computational algorithms,
- Automated integration tests directly in the browser environment,
- UI/UX debugging, including responsiveness optimization (RWD) and cross-browser compatibility,
- Deployment: managing the Git repository, building the production version, and configuring the CI/CD pipeline for GitHub Pages.
The final result is a fully scalable web application that in many ways surpasses commercial subscription-based solutions available on the market.
1. Telemetry Module and Live Analytics
The system processes keyboard input events in real-time (classic event listeners optimized for latency) and calculates key efficiency metrics:
- WPM (Words Per Minute): smooth typing speed,
- Accuracy: expressed as a percentage of correctness,
- Volume Counter: the total number of characters entered,
- Streak: an algorithm that counts the maximum series of error-free keystrokes.
Instead of monotonous text repetition, a game salon was implemented (including catching falling letters, destroying objects marked with characters, and a rhythmic marathon).
From a technical point of view, the most interesting is the adaptive difficulty level algorithm. AI implemented logic that analyzes the user's current error rate – if accuracy drops, the game loop slows down, minimizing frustration and consolidation of bad motor habits.
3. Interface and Micro-Course System
The application includes full support for muscle memory mechanics. The visual interface maps base positions (bumps on F and J keys) and dynamically highlights the corresponding keyboard areas and fingers to use for the movement.
Additionally, in just a few iterations, the model implemented 6 complete graphic themes (including Dark Mode, Neon, and Twilight), taking care of the proper contrast ratio in accordance with WCAG guidelines.
DevOps Automation and Infrastructure
What differentiates an advanced AI agent from a regular code generator is the ability to operate on infrastructure. In this experiment, the model independently managed the runtime environment:
[Lokalna maszyna deweloperska]
│
├─▶ Auto-sync przez SSH
│ → [Raspberry Pi — serwer testowy]
│
└─▶ Git + CI/CD
→ [GitHub Pages — produkcja]Codex had no problem with key configuration, deployment to a remote device on the local network (Raspberry Pi), and proper preparation of the structure for static hosting on GitHub.
The project is fully public and available for testing at:
🎮 Try Master of Keyboard: margib.github.io/MasterOfKeyboard
💻 Source code of the project: github.com/MarGib/MasterOfKeyboard
Conclusions from the Experiment: The Dawn of Intentional Programming
This test proves that we are entering the era of intentional programming (Intent-Driven Development). So far, building dedicated educational tools has required purchasing expensive SaaS licenses or engaging human resources for many working hours.
Today, as engineers, we receive technology that allows for the radical democratization of software creation. If any application on the market does not meet your expectations – whether in terms of aesthetics, ergonomics, or missing functionality – you can prompt, test, and deploy your own alternative in one afternoon. Such solutions are also an ideal way to draft a project, test its operation, show it to a test group, and collect feedback.
This was just the first in a series of automation tests I have planned for the near future. The next projects will cover even more complex system architectures.
I encourage you to follow my blog, where I will regularly publish technical case studies from the intersection of process automation and AI engineering. A broader perspective is coming – stay up to date!
Comments