How AI Unlocked My Coding in two years | karotammela.fi

My journey with coding began traditionally: I took a few Python courses and learned the basics of programming. To be honest, however, coding and the process of learning it felt quite tedious back then. Theory-heavy practice didn't really take off, and my progress often stalled as my motivation vanished into the depths of syntax and exercise drills.

Everything changed when Gemini 2.0 hit the market. It provided a way to skip the endless repetition of basics and move directly into practical building. Coding was no longer just about learning rules; I could start solving real technical challenges at a project level. This shift made learning goal-oriented and significantly more rewarding.

Copy-paste koodauksella alkuun

My first concrete project was a PHP-based Green Fee sales platform. The implementation was straightforward, but slow and error-prone: I discussed the necessary logic with Gemini, took the PHP snippets it suggested, and moved them directly into the cPanel file manager. The project involved integrating a MariaDB database and Stripe payment processing. I developed a workflow to manage context limits while coding. When Gemini’s context window filled up, I used a text-based 'handover file' to port the progress from the previous session into a new chat, attaching only the most essential code files. I quickly realized that the AI would start to lose its grip and become erratic once the context window became too cluttered.

Even though I started by heavily copying and testing, the process turned out to be a surprisingly good teacher. Seeing the application come to life and fixing errors through conversation grew my understanding of code structure much faster than online courses ever did. I began to grasp how database queries, APIs, and the UI work together, even if I hadn't written every single line from scratch yet. My understanding grew exponentially during that phase.

Two years in – from PHP to a modern stack

Almost two years have passed since that first experiment—maybe about 1.5 years of active development. Today, coding is a staple of my daily life, and I dedicate about 20–30 hours of my free time to it every week. Along the way, I’ve found a tech stack that works for me, allowing me to manage multiple projects simultaneously without the whole package falling apart.

Currently, I have several active projects on my desk:

• Next.js-based sales platform: I’m rebuilding the original Green Fee concept with Next.js, which has taught me a lot about modern web development and performance.
• Cabin booking site: This project is already finished and in use. It taught me a massive amount about booking logic and reactive user interfaces.
• Wholesale e-commerce store: I’m building a custom B2B webstore for a wholesaler, focusing particularly on a smooth ordering process and complex pricing logic.

Developing has shifted from random experimentation to goal-oriented engineering, where every new project teaches me something new about architecture and scalability.

Agentic AI – conversing with the team

My workflow has evolved significantly since the early days. Nowadays, I prefer to speak of agentic AI coding. When I’m coding, it feels more like I’m conversing with skilled colleagues rather than just giving mechanical commands to a machine.

I’ve learned to manage the context window of language models to keep project implementation efficient. My prompting skills have reached a level where I rarely need to spend time double-checking if the agents understand my intent. It also helps that LLMs have taken massive leaps forward during this time. I use different models for different purposes based on their strengths; currently, my primary tool is GPT Codex 5.3. It serves as a reliable partner for building complex systems and refining application logic. For lighter tasks, I actively utilize more affordable models.

Learning has never been this fast or this much fun

Even with a lot of hours behind me, the journey is just beginning. The learning curve has remained gentle, and absorbing new information is faster than ever with current tools. What felt like a chore in the beginning has turned into addictive problem-solving. As a kid, I loved playing world-building and strategy games like Civilization, Command and Conquer, Theme Park, and A-train. This modern way of coding feels like those games brought to life. Instead of building a fictional game world, I’m building real-world applications. It’s incredibly rewarding.

I hear this same sentiment in almost every podcast or blog I consume regarding agentic AI or language models. Many developers have rediscovered their passion for building software now that AI handles the tedious work, allowing them to focus on the big picture and challenging problems.

Staying current

I keep my skills sharp by reading tech blogs, listening to podcasts, and using Gemini as a tutor to study new technologies. I am constantly expanding my understanding, and every new library or tool is easier to master than the last. Current tools offer immense opportunities, and it’s exciting to see how much my professional expertise will continue to evolve now that my work is daily and goal-driven. The journey is truly just getting started.