AI for Coding: A Revolution or Just a Buzzword?

Hello everyone, Dimitri Bellini here, and welcome back to my channel, Quadrata! It’s always a pleasure to share my thoughts on the open-source world and IT. If you haven’t already, please give this a like and subscribe to the channel. In this episode, I’m diving into a hot topic: artificial intelligence for coding. Is it truly the game-changer many claim it to be, or is it just another overhyped buzzword? Let’s find out.

The Promise of AI in Coding

The idea that AI can help us write code is incredibly appealing. Coding, whether it’s in Python or any other language, isn’t always straightforward. It involves working with libraries, understanding complex concepts, and debugging. So, the prospect of AI assistance to generate scripts or entire software is definitely something that excites many people, including me!

However, there’s a catch. Accessing these AI coding tools often comes at a cost. Many platforms require a subscription, or you need to pay for API usage, like with OpenAI’s ChatGPT. And, of course, you’ll need a computer, but the bulk of the processing is increasingly cloud-based.

Personally, I’ve experimented with AI for tasks like creating widgets in Zabbix and tuning parameters in Python scripts. The results? Mixed. Sometimes AI does a decent job, but other times, it falls short.

Popular AI Coding Tools

Let’s look at some of the popular tools in the AI coding space:

• Cursor: One of the most well-known, Cursor is essentially a fork of Visual Studio Code. It provides a suite of AI models (OpenAI, Anthropic, Google) for a subscription fee, starting at around $20 per month. The pricing model, based on tokens, can be a bit complex. Initially focused on code creation, Cursor now seems to emphasize code suggestion and autocompletion.
• Windsurf Editor: Another VS Code fork, Windsurf also integrates API calls to major AI models. It’s priced slightly lower, around $15 per month. Like Cursor, the actual cost can vary based on token usage.
• Cline and Roocode: These are open-source VS Code extensions. Roocode is actually a fork of Cline. While they offer the advantage of being free, you’ll need to manage your subscriptions with AI providers separately. This approach can be cost-effective, especially if you want to use local AI engines.
• Bolt DIY: Similar to Bolt.new, Bolt DIY is an open-source platform focused on code generation. While it can be useful for small tasks, I have doubts about its effectiveness for more complex projects. It also comes with a subscription fee of around $20 per month, but the token allocation for AI models isn’t very clear.

In my own testing, I used the trial version of Windsurf. I attempted to create a widget for Zabbix and modify a Python script. In just two days, I exhausted the available credits. This highlights the importance of carefully evaluating the cost-effectiveness of these tools.

The Concept of AI Agents and Tools

To improve the output from AI, the concept of using specialized AI agents has emerged. Instead of giving an AI model a broad task, breaking it down into smaller, specialized tasks can lead to more efficient and sensible results.

This is where “tools” or “function calling” comes in. These techniques allow AI engines to use external tools. For example, if an AI model’s dataset is limited to 2023, it won’t be able to provide real-time information like today’s flight details. However, with tools, the AI can be instructed to use an external script (e.g., in Python) to fetch the information from the internet and then process the output.

This capability extends the functionality of AI models, enabling them to, for example, pull code snippets from documentation or connect to APIs.

Challenges and the Model Context Protocol (MCP)

Despite the promise, there are challenges. Not all AI models support tools or function calling, and even those that do may have different formats. This is where the Model Context Protocol (MCP) comes in.

Introduced by Anthropic, the company behind Cloud, MCP aims to standardize communication between different tools and AI models. Think of it like a USB hub for AI. It provides a standard way for AI to discover available tools, understand their functions, and invoke them. This standardization could simplify development and reduce the complexity of integrating various services.

The MCP server, which can be hosted in your private cloud, exposes an API to allow AI or MCP clients to discover available tools and their capabilities. It also provides a standardized method for invoking these tools, addressing the current inconsistencies between AI models.

The Road Ahead

Despite these advancements, AI for coding still faces challenges. AI models often struggle to interpret the output from tools and use them effectively to produce satisfactory results. We are still in the early stages of this technology.

There are also concerns about the complexity introduced by MCP, such as the need for a server component and potential security issues like the lack of encryption. It’s a balancing act between the benefits and the added complexities.

Personally, I don’t believe AI is ready to handle serious coding tasks independently. However, it can be incredibly useful for simplifying repetitive tasks, like translations, text improvements, and reformatting. AI is excellent at repetitive tasks. While I may not be using it to its fullest potential, it certainly makes my daily tasks easier.

The future of AI in coding is promising, especially with the development of smaller, more efficient models that can run locally. Models like the one with 24 billion parameters, having the same capacity as DeepSeq R1 and requiring 20GB of RAM, are a step in the right direction. If we can continue to refine these models, AI could become an even more integral part of our coding workflow.

Let’s Discuss!

I’m eager to hear your thoughts on AI for coding. Please share your experiences and opinions in the comments below. Let’s learn from each other! You can also join the conversation on the ZabbixItalia Telegram Channel.

Thank you for joining me today. This is Dimitri Bellini, and I’ll see you next week. Bye everyone!

Visit my channel: Quadrata

Join the Telegram Channel: ZabbixItalia