AI tools for programmers

Artificial intelligence is revolutionizing the world of software development, offering increasingly sophisticated tools that increase programmers' productivity and efficiency. In this article, we will explore the main categories of AI tools currently available for programming.

🧠 Large Language Models (LLM).

This is the backbone of AI tools for everyone and, of course, for programming as well. The least you need to know about an LLM is that it is a gigantic database containing information on any topic that can be queried using natural language. It will answer you equally with text, and for that, it uses super complex statistical techniques that it refines during learning processes.

And the work of a programmer is based on writing code, to express complex processes in detail in a concise and unambiguous language. For an LLM working with code instead of human text is a breeze.

It is therefore not surprising that LLMs have become a fundamental part of AI tools for programmers. These models can generate, explain, document and correct code. But increasingly, it is helping in all phases of development. It pays to know a little more about them.

🔒 Closed LLM

For starters, this is software, data and computation. So, developing, training and running all of this costs money. A lot of money. It is normal that the first and let's say the best LLM, are closed products, developed and maintained by large private companies. Surely you have heard of OpenAI, Anthropic, or Mistral, in addition to the usual giants like Google, Meta, Microsoft, etc.

Claude de Anthropic

• Although less well-known than GPT-4, Claude offers the best programming assistance and code generation capabilities.

GPT de OpenAI

• Used in tools like GitHub Copilot, it offers advanced code generation capabilities and programming assistance.

🔓 Open LLMs

Open LLMs are more or less open-source models that can be used, modified, and distributed more or less freely. All a little bit more or less, you know....

Gemini from Google

• Also offers advanced code generation capabilities and programming assistance.

LLaMA from Meta

• A large-scale model that can be tuned for specific programming tasks.

Codestral from Mistral

• Very good models for code, and of smaller scale than the previous ones.

Qwen from Alibaba

• Another model that stands out for its small size and large capacity for code.

As you might guess, software is one thing, but the hardware required to run these models is quite another. This is where GPUs, gigabytes of RAM, and terabytes of SSD come into play. The major manufacturers offer you their own cloud services or use the most powerful infrastructures like Google, Azure or AWS. From there, you can use them and pay for the use you make of them.

📦 API services

But there is more. Companies have proliferated and specialized in offering accelerators, hubs, and obfuscators of calls to those APIs. For example:

Groq

• Offers a call accelerator to LLM APIs from several vendors.

OpenRouter

• Multi-vendor API integrator, with a catalog of more than 150 models that allows you to call them transparently.

Hugging Face

• You can use any model on their servers, and the best thing, you can fine tune it to your needs.

And yes, you can also use local AI models on your own machine.

🏠 Local installation of LLM

Local LLM installation offers advantages in terms of privacy (that's for sure because you don't use external services) and speed of response (you'd have to see it because running an LLM locally is crazy for most machines). But, if you want to experiment, some popular options include:

LM Studio

• An application that allows you to run language models locally on your computer.

Ollama

• A tool for running LLM locally, with support for various models such as Llama 2 and CodeLlama.

These local LLM runners need to download the models, usually from Hugging Face. And, as said, to run them, they need quite a lot of computing power and RAM.

🧑‍💻 AI-enhanced code editors.

The impact of AI on code editing has also given rise to a diverse ecosystem of tools ranging from classic editors with AI-enhanced extensions to next-generation editors with deep AI integration.

🧩 AI extensions in classic editors

Traditional editors such as VsCode or JetBrains have quickly adapted to the rise of AI with performance-enhancing extensions.

Aider

• It almost deserves a separate category because it's really a CLI that integrates into your editor terminal to help you write code.

Cline

• More than an editor, it uses agent capabilities to access the browser and enrich the context of the development session. It can use any model provider, remote or local, via OpenRouter.

Cody

• Provides intelligent code completion and online documentation for several programming languages (with free overlay, which is not bad).

Continue

• Works with VsCode, Codeanywhere and with JetBrains, in remote and local mode, with code, repository and documentation contexts.

GitHub Copilot

• Although not an editor, GitHub has integrated its own LLM into its code editor.

Pythagora

• One of the newest, it's being presented while I'm writing this...

Qodo

• A tool that focuses on code quality and testing.

Tabnine

• An AI assistant that integrates with several editors and offers AI-based code auto-completion.

Anyway, I'll stop now. There are many more, and they are popping up like mushrooms. So this list is just to give you an idea. All these tools integrate AI as an additional layer on top of robust and highly customizable editors, allowing you to combine your usual work preferences with the power of AI. But others have already been born as AI editors.

🤖 AI-integrated editors

It couldn't take long for new editors to appear to integrate AI natively without the need for extensions. Some are specialized in certain technology stacks and seem more focused on the entry-level audience, such as SaaS developers.

Bolt

• The multi-framework response of StackBlitz and the closest thing to "Computer, make me an app"

Cursor

• A fork of VSCode with outstanding AI and spectacular usability.

Repplit

• Specialized in Python, capable of generating and deploying complete applications.

Spark

• The GitHub editor with integrated AI to create micro-apps directly usable from the browser.

V0

• The same but with Next.js

Windsurf

• Super powerful and well thought out for enterprises. More than an auto-completer; an agent that helps you in the coding process.

Zed

• Its differentiating mark is speed and that influences how AI is integrated into the development process.

📦 Conclusion.

AI is transforming the way programmers write and maintain code. Whether through powerful closed LLMs or privacy-enhancing open options, or through editors that integrate AI, the future of development is looking more efficient and automated. Developers have more options than ever to leverage these tools and improve productivity without sacrificing control over code.

The landscape of AI tools for programmers is constantly evolving, with new solutions emerging regularly. I'll be describing some of them in subsequent posts. For now, if you're a bit lost, I suggest you start by installing Cursor and/or adding GitHub Copilot to your VSCode. It has a free tier, and they are among the ones that will surprise you the most. Then you can try something more complex like Aider... or, who knows, install your own LLM on your machine.

If you represent a company or a development department, you will have already been offered some of these tools. So I don't want to bother you any more.

In any case, these tools not only increase productivity but also help to improve code quality and make it easier to learn new languages and technologies. Start incorporating them into your day-to-day work.

Code smarter, not harder.