AI for Mobile Development: Tips and Tricks for Making Your AI Assistant More Efficient and Smart

Today, there is no turning back – everyone should use AI in IT, as it is required by most clients and businesses. Everybody uses it to some extent, but if you are interested in doing this more effectively and elegantly, I have collected a few essential tips and tricks on how to make AI better for assisted development, with no additional cost (well, mostly).

A rolling stone gathers no moss: try new agents and models frequently

AI changes extremely rapidly at the moment. For a few months in a row, no week passes without some major update for some AI agent or model. 

So you need to try new AI models and new AI products, even if the current AI setup is working fine. It is ok for most of your work to use your favorite model, but try new products for routine and simpler tasks, or otherwise, for challenging tasks that your current favorite AI can’t handle. This way, you will create a stack of AI assistants or models that you like and know how to use efficiently. 

Good Code comes from restriction: add guidelines for your architecture and code style

You can try to add loads of files to the AI context alongside the prompt, but your token will run out in no time. So if you want AI to generate beautiful, architecture-right code, you need to provide guidelines. For example, if you use Copilot, you need to add to your .github folder copilot-instructions.md file and add as many guidelines as you can think of. Add architectural tips, naming style, codestyle preferences, UI design guidelines, and all other things that you painfully add to every other prompt. Believe me, this will make a difference that you will notice fast.

Provide context: Add more documentation to your project

Every time you press the “New Agent Session” button in your IDE, your AI assistant loses any context of the feature you are working on. This is a problem because AI would not inform you of that and will try to perform your request using only the info from the prompt and your codebase. Well, in this case, documentation is key.

To make things easier for AI, try to add documentation comments to the head of your class files and to each field in your class. It can be a few words, but if you can do this, this will make AI more in context of how your existing code works. Also, these comments will help you in the future. When you go back to “that part of the app” that nobody touched for years and now need to refactor old code, you will be so glad that you left those documentation comments. Also, you can add rules to your AI guidelines (that we covered in the previous tip) so it will generate code with documentation from the beginning.

 Add augmentations to your AI: Use MCP Servers

From my personal experience, most developers do not use any MCP Servers or integrations. And they are missing out, because MCP can really enrich the AI experience. If you use Figma, Jira, Notion, GitHub, or write your code with something except JS, you need to at least try MCP integration with your AI Agent. Especially if you are using Figma and working with the Frontend/Mobile part. MCP integration is quite easy, and you can ask AI to help with MCP anyway.

In conclusion, I have to say that you need to think about AI as a very smart, but junior developer with amazing googling skills. You need to teach, restrict, and check everything to have good results. But when it clicks, you will see how many routine and soul-sucking tasks you can automate and then you can focus on what really matters – developing a good, consistent codebase that will cover business needs with ease.

To find out more about how our company can help your company with its technology needs, schedule a free assessment.