My Journey with AI Models and Coding Environments
Starting with Gemini 2.5 Pro
I started off with Gemini 2.5 Pro. It was great for a start because it was cheap and had a large context window. However, it tended to write a lot of unnecessary code and would change things that I never asked it to change. This would get crazy when you ran multiple instances to solve different problems and the results ended up conflicting and overwriting one another.
Moving to Claude 4
I then switched to Claude 4, and it was a needed change. Claude was more precise when it came to solving problems. It kept things concise and to the point, and it was very easy to follow what it was doing. However, the context window was small and it would struggle with long conversations. This meant that you had to have one chat to solve one problem and then create another chat.
GPT-5: Golden but Heavy
I then moved to GPT-5, which was golden but slow and expensive. However, the results were always worth the wait. It was better at planning than Gemini and had a better understanding of the context of your code than Claude. It could write more code under one chat. However, it is always best practice to keep one chat to solve one problem.
Claude 4.5: Surgical Precision
I have now moved to Claude 4.5, and the best part about it is that it is surgical, understands context, and is fast as well. It is like having the context understanding of GPT-5 but faster. Claude also does a lot of planning and documentation using temporary markdown files, which is helpful for its memory and execution.
Coding Environments
Cline on VS Code
I started by using Cline on VS Code, but it gets expensive very fast. Because you are an individual, you don’t benefit from the enterprise discounts for using the APIs at scale. I would connect to Claude using my personal API key, but it was just not worth it. In fact, it was more expensive to fix bugs within your code than it was to write code. As any developer will know, 80% of your time is spent writing code and the other 80% of your time is making sure that the code is doing what you want it to do, which is a lot of debugging and restructuring. It would cost $0.40 for one prompt to fix code on Cline. It was insane.
Windsurf
I then switched to Windsurf. I liked that they had free models and that they were offering Gemini 2.5 Pro for free at some point. It really helped me build out a lot of my projects. However, Windsurf’s connection to the OS environments is weak, especially when you are on Windows and use WSL. It would slow me down when I had to run the commands myself because it would always fail to run them on WSL. For some reason, it was sticking to CMD instead of using Ubuntu WSL. I couldn’t change it even from the config files under settings.
Cursor: A Game-Changer
Everything changed when I moved to Cursor. I chose Windsurf because it was $5 cheaper, but it was a mistake. I should have gone with Cursor from day 1 and paid the $20. Right now, I have spent more than that. Cursor has become as indispensable as the internet. I believe I have spent over $250 on Cursor to build my projects, and I am happy to pay even more because of the amount of value that it offers.
Other Coding Tools
I have heard people talking about Claude Code, but I am an OpenAI junky. OpenAI has a lot of context on my work and I am not willing to just pay for Claude for coding when I already have Cursor that I am paying for. ChatGPT is good for everyday tasks like research, drafting documents, making plans, therapy, and so on. Yes, therapy. ChatGPT probably knows too much about me at this point.
I am also looking into a Claude Code competitor called Warp, but not seriously. I am still happy with Cursor.
Productivity Tools
Gamma
I have subscribed to Gamma, which is a text-to-slides SaaS for creating presentations. It has been a game-changer when it comes to creating slides out of text for a quick meeting. I usually know that if I have an agenda and a report on what has been done and next steps, I can have ChatGPT structure the content for me as a Google Doc and then paste the text on Gamma to get slides in less than 5 minutes. No more wasting time on slides unless you are building an investor deck. In that case, you can take more time to tweak the Gamma slides with pictures, edit text to make it shorter, and so on. Gamma has been a life-saver.
Bolt and Lovable
I have also used and paid for other tools like Bolt and Lovable. I like these tools because they are good at creating frontends which I use as prototypes. Instead of going onto Figma and trying my hand at design, I can just get Lovable to develop a frontend for me and I can use it as a prototype to demo a solution or a website. No more moving rectangles and icons, just let the AI build it out and get code as the result that you can deploy on Netlify.
However, I stopped paying for them because I don’t see the need to pay $20 a month when I am not prototyping every day or working on designing. I have also found that once you have a good frontend on Lovable or Bolt, it is best to push that code to GitHub and continue working on it with Cursor. Cursor has a large context window and it has been battle-tested to build production-level applications. Lovable or Bolt could do it, but they are not as malleable as a code editor based on VS Code.
Transcription: Turboscribe
Additionally, I have been using a tool like Turboscribe to transcribe audio for me. One thing that I would like to point out is that transcription is not valuable enough in my life to pay $20 a month. I have even built a transcription tool for myself, but I haven’t hosted it because Turboscribe offers 3 free transcriptions a day which are 30 minutes each. This is more than enough and I barely use it.
It would be interesting if this could simply be made free or as affordable as Spotify at $3 to $5 a month. However, I doubt I would pay for it anyway. The only problem with Turboscribe is that it is a web app, and therefore if I record audio on my phone, I have to share it to the laptop via Google Drive or WhatsApp, download it, and then upload it to Turboscribe. I then have to download the txt file from Turboscribe and put the transcription in ChatGPT to get insights and come up with a report. Not the most efficient thing in the world, but it works. Having something more integrated where I am not paying $20 a month would be a plus. It would even be better if we had something that just runs these models locally and is self-contained but also syncs to multiple devices.
My AI Budget
- Cursor – $20 minimum (can get to $150)
- Gamma – $20
- ChatGPT – $20
Final Reflections
An interesting observation is that I have never used my debit card to make this many subscriptions. AI has opened a new resurgence for SaaS where building the right AI application can open up a new market of people who would normally not pay for software.
Most people don’t believe they should pay for software unless it is absolutely necessary. However, with AI and the productivity gains, many people are opening up to the idea of paying for software more and more. Before AI, I only paid for Spotify, and it was just $3 a month—and I had to be forced because of the annoying ads. Now I am paying crazy amounts because I need to build a project, make some slides, or talk to ChatGPT for planning and motivation.
There is real opportunity in this space. Now it finally makes sense to pay for software when it can get things done for you in a productive and flexible manner.
It goes back to the article on the Horseless carriage. The AI revolution will be just as significant—if not more significant—than the invention of the internal combustion engine. We cannot imagine a world without cars. We also will not be able to imagine a world without AI.
The best part about AI is that it allows us to not just consume but co-create with it. I believe this is the most important differentiating factor of AI from other technologies. It is your second brain, the boost that your brain needs to get cognitive work done. Those who use it will win the future over those who do not.