Harnessing AI Effectively in Coding: Modern Strategies

Staying ahead in programming means rethinking how we use AI. Since the debut of ChatGPT in 2022, developers have largely divided into two factions: those who reject AI in their coding practices and those who leverage it. Many experienced developers discourage the use of AI-generated code, citing concerns about its reliability.

Regardless of your stance on AI, it’s likely you’ve encountered obstacles when integrating it into your coding routine. The essential task is to develop effective strategies to harness AI to your benefit in a practical manner.

Many programmers still rely on outdated methods that were once effective. Using rudimentary tools is akin to attempting to chop down a tree with kitchen utensils.

In this piece, I will outline two significant pitfalls in AI usage for developers and introduce four dependable AI tools that can elevate your coding practices, regardless of your expertise level.

The Ineffective Ways You Might Be Using AI
- Using the Wrong Interface
- Holding Unrealistic Expectations of AI
Cursor: An AI-Driven IDE
Micro Agent: Code and Test Case Integration
SWE-agent: Resolving GitHub Issues with AI
AI Commits: Streamlining Commit Messages
Let’s Connect!
Additional Reading

The Ineffective Ways You Might Be Using AI

Identifying ineffective AI usage can be categorized into two main issues:

Using the Wrong Interface

Upon ChatGPT's release, the predominant approach for interacting with AI involved visiting a website and conversing with gpt-3.5 directly in a browser. At that time, the AI ecosystem was quite straightforward, with limited tools or solutions.

Many, including myself, were intrigued by ChatGPT's coding capabilities. While not revolutionary, it could generate simple functions rapidly.

Wanted to have ChatGPT document your code? 1. Copy the code from your IDE. 2. Provide it to ChatGPT with a basic prompt, such as "add comments." 3. Receive the revised code. 4. Verify its accuracy. 5. If correct, paste it back into your IDE.

This process could become tedious, especially when repeated for multiple functions. Yet, many still rely on this outdated method, particularly among students and novice programmers.

As the AI landscape has evolved over the past couple of years, our approaches have lagged behind, failing to leverage the full potential of these technologies.

Holding Unrealistic Expectations of AI

Another common misstep with AI usage involves how we engage with it. Typically, a developer might request code from an LLM, test its functionality, and engage in a back-and-forth process to troubleshoot issues. Often, this leads to frustrating cycles of hallucinations where the model struggles to grasp the problem.

While it’s clear that the code doesn't work, we may continue to push for solutions without a clearer path, leading to frustration.

With that context in mind, let’s explore four tools designed to enhance your programming efficiency using AI.

Cursor: An AI-Driven IDE

Best For: Everyone

When I was first introduced to Cursor, I had little understanding of its capabilities. However, after experimenting with it, I recognized it as a transformative tool. Cursor is an AI-enhanced IDE, essentially an upgraded version of VScode.

Cursor integrates your chatbot API to assist with code writing. Here are some features that significantly improve coding efficiency:

Ctrl K: This key feature allows for multi-line selection, prompting the chatbot for assistance. You can request the LLM to draft entire functions, document your code, or modify existing segments. It’s user-friendly and clearly indicates code changes.
Tab: An advanced auto-complete function that aids in writing both single lines and complete functions.
Chat: Engage with your AI regarding your entire codebase, providing context through documentation, images, or web links (Ask the Web feature).

The beauty of Cursor lies in its familiarity as an enhanced version of VScode, making adaptation seamless for current users.

You can integrate your preferred API model, be it GPT-4, GPT-4o, or Claude 3.5 Sonnet, alongside the built-in free option, cursor-small. Many users appreciate the combination of Cursor + Sonnet 3.5 for its coding proficiency.

Cursor represents a significant upgrade from the tedious copy-pasting between ChatGPT’s website, which can feel unprofessional at this stage.

Micro Agent: Code and Test Case Integration

Best For: Reliable Code Generation

Source: micro-agent (under MIT License).

When using AI for code generation, reliability can often be a concern. As previously mentioned, LLMs can get caught in frustrating loops of hallucination.

Micro Agent offers a dependable alternative. Instead of delivering large code segments to be vetted by users, it begins by generating test cases for your prompt and subsequently produces code that meets those criteria.

The rationale is that LLMs excel at creating test cases rather than directly generating solutions. While specific research is limited, it’s evident that tested code tends to be more reliable than untested code—something Micro Agent ensures.

If you’re aiming to develop functions that are moderately complex yet challenging to perfect, Micro Agent yields more reliable outputs compared to standard LLM results.

Be sure to explore their blog and GitHub page for more insights.

SWE-agent: Resolving GitHub Issues with AI

Best For: GitHub Enthusiasts

Among the effective agent tools is SWE-agent, developed by Princeton Language and Intelligence. This tool specializes in addressing real-world issues on GitHub repositories and generates pull requests for review.

It’s particularly beneficial for larger repositories, organizations, or smaller projects. Its effectiveness stems from specialized tools that analyze codebases, pinpoint issues, and implement necessary corrections.

SWE-agent is an open-source initiative with over 13.1K stars on GitHub.

AI Commits: Streamlining Commit Messages

Best For: Committed Developers

Struggling to craft meaningful commit messages? You’re not alone. There must be a more effective way to convey the essence of your changes than simply adding timestamps or vague notes like "minor changes."

AI Commits provides a straightforward solution. It’s user-friendly and installs easily. By analyzing your git diff, aicommits generates an AI-crafted commit message, utilizing gpt-3.5—a cost-effective and efficient choice for this task.

Ultimately, what’s crucial is employing tools that enhance your productivity and help you achieve your objectives. The AI landscape evolves rapidly, and so do the tools within it. If you’re interested in staying updated on AI advancements, I’ve written an article outlining how to do so in just a few minutes each day.

While we can’t completely eliminate AI hallucinations or limitations, we can take steps to mitigate them. Engaging in simple, manual interactions with AI—like using ChatGPT to generate code or comments—can prove frustrating and inefficient. By selecting the right tools, you can avoid common pitfalls and work more confidently.

Thank you for reading! I aim to provide accurate and insightful content, so please share your thoughts and suggestions for improvements.

Let’s Connect!

Subscribe for FREE to receive notifications about new articles! Connect with me on **LinkedIn* and Twitter.*

newcohospitality.com