Software development is one of the industries that is currently seeing the highest ROI from Large Language Models, and while it can now allow non-tech people to enhance their coding abilities, I believe that professional developers will benefit the most by becoming better equipped and more efficient.
From personal experience, as a non-developer, I can say that the ways in which I have been using AI in my projects have dramatically improved my development capabilities, allowing me to make progress in hours on tasks that would have previously taken me months.
My Limited Technical Experience
Disclaimer: I am in no way a professional developer. Having spent almost a decade in the startup and tech world, I've gotten my feet wet with various coding languages - Java, JavaScript, HTML, Cordova, PhoneGap, Swift, Objective-C, Python, Bash, SQL, and more. In having this limited experience, I gained some confidence with code by picking up bits and pieces along the way, but I'm far from proficient.
Before the recent AI wave began, I would spend hours Googling, reading Stack Overflow, Reddit, and watching YouTube videos to understand coding concepts, and how to fix things that wouldn't work how I wanted. In the end, I would have to ask a friend or colleague for help. Now I use Perplexity for understanding my problems, and I utilize various coding LLMs to solve them, allowing me to accomplish my goals much faster.
Five AI Tools That I've Used to Improve My Code
I must mention that every time I'm about to post on this subject, a new tool either comes out or an older tool is updated. Things are moving very fast, and it's exciting (and overwhelming). Like Pokémon, I gotta try* them all. Here are my five favorites and tools I think you should try.
1. Continue.dev (Free)
Continue is a powerful VS Code extension that connects various LLMs directly to your IDE (integrated development environment). The tool adds the power of AI to be able to see your entire codebase, terminal outputs, and more. It can help complete code, offer autocomplete suggestions, fix errors, answer questions in natural language, and understand your project's context without constant copying and pasting.
One of the great things about Continue is its ability to connect to both frontier models like ChatGPT (via APIs) or locally hosted open-source models via Ollama like Llama3.1 (how I use it). This means you can get AI assistance without relying on internet-connected services or paying for API calls, (making it a free solution).
For the open-source models, I'm currently using Deepseek Coder v2 and CodeQwen.
2. Cursor (Free with Pro Plan)
Cursor is an AI-powered IDE that's very similar to Continue. It's actually a fork (enhanced copy) of VS Code but with all of the AI features built in. You can pay Cursor for its AI assistant, or connect the frontier models via API.
What I love about Cursor is its intuitive interface and how seamlessly it integrates AI into the coding process. It's particularly helpful for getting quick explanations of code snippets or generating boilerplate code (similar to Continue).
A few things I like about Cursor are that it can search the internet (documentation), is multimodal (so you can share screenshots), and how fast it's improving (it's more feature-rich than Continue).
Everyone is going crazy about Cursor right now, but I still prefer using the native VS Code with Continue. It's almost the same experience; I haven't really seen anything different in Cursor.
3. Claude Dev (Free Extension - need Claude API credits)
Claude Dev is another VS Code extension (similar to Continue), but it's specifically designed to work with Claude. It leverages Claude's API and features like context caching to provide powerful coding assistance directly in your IDE.
Claude 3.5 Sonnet is currently hands down the best frontier coding model and is particularly strong in understanding complex coding concepts and providing detailed explanations.
While I can connect Claude directly to VS Code with the Continue extension, I prefer using Claude Dev as its own extension because it utilizes Anthropic's recently added content caching, and it tells me how much it's costing me (token-wise), which helps me keep track of my usage.
4. Claude "Projects" (Paid)
This is really the key to enhancing my product and development workflow, even when I don't use it for writing code (when I'm using Continue or Claude Dev).
Claude Projects is a paid feature that comes with a Claude Professional account, that allows you to create customized AI assistants for various tasks (without the need for fine-tuning). It's a very powerful feature that is not limited to just coding - you can use it for any type of project (I recently made a post on how to use Claude Projects or Custom GPTs for better document handling).
Its power comes from its custom instructions and knowledge base features. A very brief explanation of how I use it:
I start by creating a new Claude project for each actual project I am set to begin.
Then I give it custom instructions, telling it that it is a coding assistant, explain my coding level, my hardware setup, and any other relevant context.
I then upload all relevant documents about what I'm trying to build, including mockups, Product Requirement Documents, user stories, and project roadmap.
Most of these documents I also create with Claude.
I then have Claude summarize what I shared with it, and have it create an even more powerful custom instruction (which I replace the original custom instruction with).
Now that I am equipped with all supporting documents and files, I start coding with Claude, asking it for advice, I can share my code with it, or vice versa, and even share terminal outputs (all via copy and paste).
After each coding session, I have Claude create a summary of what we did that session, which I then upload directly to the knowledge base, so it now has memory and knows where we stand next time we start a new session.
Lastly, I upload the entire codebase to the project's knowledge base whenever I make significant changes. This way, when I ask for help, Claude has a comprehensive view of my project without me having to paste all the code each time. It's a huge time-saver and really improves the accuracy of Claude's guidance. I do this via a custom script that takes my whole codebase, turns it into a markdown file and then I upload that to the project knowledge base.
5. ChatGPTs: "Custom GPTs"
Custom GPTs, a feature of OpenAI's ChatGPT, are very similar to Claude Projects. They also allow you to create customized AI assistants. To be honest, Custom GPTs came out a long time before Claude Projects, and I used to use the entire Claude workflow I mentioned above, so if you prefer ChatGPT, you can do that too.
My AI Coding Workflow
Currently i am using three of these tools simultaneously, and to be honest, they are getting to a point that they are easily drag and drop (Cursor vs Continue vs Claude Dev), it really comes down to budget, feature set, and ease of use.
Claude Projects is my second set of eyes for both product and development by understanding the broader context of what I'm trying to build by having access to both the codebase and relevant documents. Additionally, Claude has this great feature called Artifacts which makes taking code snippets much easier than other LLMs, and its being multimodal allows me to give it screenshots to show it UI issues.
Continue lives within VS Code, helping with code directly in my coding environment, without having to go back and forth to Claude. I often use a local LLM via Ollama to help me avoid the need to pay for API calls. Lastly, it's autocomplete is very helpful, and its ability to read context of the terminal saves me from many headaches.
The Claude Dev extension is newer to my workflow, but I like using it for more complex coding tasks, and I really like its built-in context caching, ability to run code and terminal commands, and token cost tracking.
Words of Wisdom for the Non-Coders Eager to Dive In
While these tools can really enhance our coding abilities, they're not a magic solution. Here are some key points to consider:
Focus on Fundamentals: Begin with simple, functional projects that solve specific problems. Build a strong foundation before tackling complex apps or polished UIs.
Prepare to Work and Learn: Even with AI, you'll spend time refining prompts, reviewing code, and troubleshooting. Be ready to break things, and learn how to fix them. Tools like Perplexity can help when searching for solutions.
Set Up an Environment: Learn to set up development environments and manage version control. I have found that these are things that AI usually skip over, and it,s crucial before starting project.
Be Specific: Provide clear, detailed instructions, and break down your project into smaller tasks. Remember, AI still has many limitations and will make mistakes, so always review and test code thoroughly.
Be wary: Proper practice is to never run code you don't fully understand, that being said we are talking about having AI write code for you, but be warned, some issues can't be fixed with a simple UNDO command.
I really believe it's still worth it to learn coding, even just for the basics, to understand concepts, syntax, and frameworks. I suggest even taking a Python course since its what AI is often coded with and its a relatively straightforward language to understand.
Conclusion
While these tools significantly enhance coding abilities, I don't believe they will replace software developers anytime soon. The product lifecycle involves much more than coding – from architecture design and business logic to QA and UI/UX. These assistants are powerful allies, complementing us rather than replacing human creativity and problem-solving.
For non-developers, these tools offer a foot in the door to coding, but professional developers can leverage them to boost productivity. My point being is that it's crucial to remember that AI doesn't eliminate the need for fundamental coding knowledge and problem-solving skills. Not yet at least.
In future posts, I'll delve deeper into specific use cases and discuss the limitations of AI in coding.
Whether you're just starting out or are an experienced developer, I hope this overview helps you understand how you can use AI right now to enhance your development experience.
Yorumlar