AI Tips and Tricks for Programming Anything
How AI Tools Can Help You Get Things Done
Many people think that AI tools can be used with simple writing prompts, but that is not the case.
There are many ways that I have found to help LLMs deliver better results, and I would like to share them with you.
This is not limited to any specific model or tool. I use Claude Code, Codex and also GitHub Copilot with regular VS Code, and the tips and tricks I am sharing are applicable to all of them.
Planning
Before you start asking the model to generate code, it is important to have a clear plan of what you want to achieve. The more clarity you have about the problem you are trying to solve, the better the results will be.
Feel free to ask the model to search and argue about ideas.
Moreover, all of them have planning mode in some manner, so they can create a plan for you, which you can then review and adjust.
Be specific, be clear and consult anything which is ambiguous with the model. It will help you to get better results and also to understand the problem better.
Try to ask only one thing at a time, because some models will just forget the previous context if you ask for too many things at once.
Context Rot
Even if models have large context windows, they can still forget important information if you ask for too many things at once.
Feel free to read whole research about how models lose their sharpness when they have too much context, and how to mitigate it: https://research.trychroma.com/context-rot
Long story short, you can mitigate it by doing one thing at a time. Instead of asking for a full implementation, guide it one by one.
If you want to start a completely different topic, just clear your chat and start fresh.
Do not keep long conversations with the model, because it will just forget important information and deliver worse results.
I rather go over multiple iterations with multiple conversations, and when I have a good picture, then I specify the final implementation and ask for it.
Give it a specific example, give it a specific context, and it will deliver better results. Do not add files, just because they might be relevant.
This is more of a personal note, with recent models like Codex 5.3 and Opus 4.6, they both have gotten better at keeping the thread and focusing on the task, but it is still better to do one thing at a time.
Do not ask model, what you do not know
Some people, not everyone, still tend to work without AI assistance and then when they do not have a clue, they ask the model to do it for them.
How do you expect the model to know, if you do not know? Under precondition that you are considered an experienced programmer. There is a high chance that the model will not know either, because it is trained on the same data as you are.
It will try, it will do its best and maybe it will even deliver something usable, but the chances are not high.
Switch the mindset. Let model do the work, which you know how to do already. You can guide it and it will do it for you. If you encounter a complex problem, that is the opportunity for you to explore, research and deliver a solution.
Try to understand the problem, use AI to research and find a solution. It is an experience for you.
Do not overdo it with configs
On the internet, you can find many prompts, instructions, skills and other plugins to make your AI assistant better. Some of them are good, some of them are not, but the point is that you do not need to use all of them.
Remember the context rot? You are bloating the context with all these plugins, and you are asking the model to do too many things at once.
Keep it simple, add one custom instruction, point it to the correct skill via prompt and it will do the work for you. You do not need to add 10 custom instructions, 5 plugins and 3 skills, because it will just confuse the model and it will not deliver good results.
Your Project
If your project is a mess, the model will never deliver good results.
What your project ideally should look like:
- it has type checking and linting active - so model can see syntax errors
- a simple launch command - so model can run the code and see the results
- tests are in place - so model can run them and see if the code is working as expected
The coding standards do not apply only to human developers, but also to AI tools. If you want to get good results, you have to provide good context and environment for the model to work with.
Improve the AI tooling and play around
The best suggestion I can give you is just:
Work with the model, play around with it, try different prompts, different approaches and see what works best for you.
The more you work with the model, the better you will understand its capabilities and limitations, and the better results you will get.
If you find a limitation, try to find a way how to mitigate it. Improve instructions, create a skill for it or use different prompt.
I think everyone to who I talk to, who is using AI tools, has a different approach to it, and that is the beauty of it. Do not search for manual or best prompt. Practice and experience will teach you.
Conclusion
AI tools are powerful, but they are not magic. They require guidance, context and understanding of the problem you are trying to solve. You are the one who has to guide the model, review the results and ensure that it meets your requirements.
In simple rundown:
- Plan your work, do not ask for too many things at once.
- Do not ask for what you do not know, let the model do the work you already know how to do, and use it to research and find solutions for complex problems.
- Do not overdo it with configs, keep it simple and add only what you need.
- Keep your project clean and organized, so the model can work with it effectively.
- Play around with the model, try different approaches and see what works best for you.
Socials
Thanks for reading this article!
For more content like this, follow me here or on X or LinkedIn.
