Choosing the Right AI Model for Developer Advocates

Over the past few months, I’ve spent half of each day building content—blogs, sample apps, hackathon materials, solution blueprints—and that’s not even counting personal projects. AI assistants and agents have been invaluable in streamlining my productivity. This got me thinking about a post I wrote six months ago on my go-to AI tools for developers. The landscape of AI tooling, particularly around large language models (LLMs), has evolved significantly since then, prompting me to write this follow-up.

The AI Journey So Far…

Taking a step back to consider the macro trends in AI helps set the stage. Last year, when I wrote that post, we were in the middle of the generative AI (genAI) boom. Vendors scrambled to integrate productivity enhancements into their tools. These tools were well-defined in their scope—either specifying the task themselves or requiring a specific prompt to deliver optimal results. In essence, AI was an assistant: you gave it a task, and it did its best to complete it. Many tools remain at this stage.

Fast-forward to today, and agentic AI is gaining traction. Think of agentic AI as a collaborator rather than a mere assistant. It requires less direction, understands the broader context of your work, and supports your entire workflow instead of just isolated tasks. A prime example is Cursor, an AI-powered IDE. Cursor’s latest release not only generates code but also provides refactoring support, outlines requirements for your code/project, and handles configuration tasks for you.

What’s next? The ultimate goal is artificial general intelligence (AGI), which would reason, learn, and adapt based on broad, unstructured prompts. While AGI isn’t here yet, it’s inching closer. The choice of the right AI tool is becoming less about the tool itself and more about selecting the appropriate model for the task at hand.

Choosing the Right Model for the Right Job

As AI advances, so do the LLMs offered by companies like OpenAI, Anthropic, and Google. The list of available models grows daily. Many tools now allow users to choose their preferred model, enabling vendors to stay current, offer choice, and let users leverage existing subscriptions.

As a developer advocate, my work spans coding, long-form content creation, and side projects. After experimenting with nearly all the available models—including OpenAI’s premium (and pricey) ChatGPT Pro subscription—I’ve realized that certain models excel at specific tasks. For developer advocates, these tasks typically include:

1. General AI assistance

2. Coding (from scratch or inline pair programming)

3. Research and summarization

4. Creative writing

5. Video/image generation

Let’s break down the best models for these tasks.

GPT-4 Turbo (4o), GPT-4 Turbo Mini (4o Mini), and GPT-4 o1

OpenAI models remain my everyday go-to tools, particularly GPT-4 Turbo and its Mini variant. They are exceptional generalists. Think of them as the new Google: instead of delivering ranked search results that I sift through manually, these models summarize information and draw conclusions for me. Whether I’m looking for new app development platforms or the latest frameworks, I start with these models.

The ChatGPT Mac desktop app is my new Google.

The recently introduced GPT-4 o1 model is even more intriguing. While still experimental, o1 excels at handling vague, less-structured prompts. For example, I’ve used it to create detailed product requirement documents (PRDs) from broad descriptions of sample apps I want to build. These PRDs can then be input into other models to generate code.

However, neither GPT-4 Turbo nor GPT-4 o1 is ideal for coding. That’s where Claude shines.

Claude 3.5

Claude, from Anthropic is unmatched for coding tasks, whether starting a project from scratch or serving as an AI coding assistant. I primarily use Claude in tools like Cursor or Windsurf to generate code for sample apps and blog posts. While occasional hallucinations occur, and prompts sometimes need fine-tuning, Claude consistently outperforms other models for programming.

In addition to coding, Claude excels at creative writing. When drafting blog posts or long-form pieces, I often ask Claude to write 500 words on a specific topic. While I ultimately rewrite and refine the content myself, Claude’s natural writing style provides an excellent starting point.

If there’s one takeaway from this article, it’s this: Claude is the best model for coding tasks.

Gemini 1.5

Gemini 1.5 is another versatile model, suitable for general knowledge, writing, and image/video tasks. While not a standout in any single area, Gemini does a better job at coding than ChatGPT. If you’re on a budget and need an all-in-one model, Gemini is hard to beat.

One hidden gem in Google’s lineup is the Gemini Flash model, which excels at complex summarization and code documentation. In my tests, it outperformed competitors at a fraction of the cost. Despite its capabilities, Gemini Flash hasn’t received the attention it deserves.

I haven’t had the chance to test the new Gemini 2.0 models yet, but their ability to use screen context and reasoning sounds promising. Imagine showing Gemini your iPhone app in the Xcode simulator alongside screenshots of a desired style, and it generates the necessary code—huge potential!

MidJourney

MidJourney stands out as the best tool for generating images. While DALL-E 3 is easier to use, MidJourney delivers more lifelike results, especially for people and animals. Its natural color palette also sets it apart. For text-based image generation, I’d recommend Ideogram, but MidJourney remains my top choice for visuals.

For video creation, I’ve been experimenting with HeyGen for generating explainer videos featuring talking-head avatars. While the avatars’ movements can feel a bit exaggerated, the tool is impressive overall. With video becoming an increasingly vital medium for developer relations, this is a space to watch closely.

Summary

The role of a developer advocate varies daily—coding, creative writing, research, and video creation. Just six months ago, AI was a productivity enhancer; now, it’s a true collaborator. Understanding each model’s strengths and weaknesses is key to working efficiently.

By sharing my experiences with major LLMs in daily developer relations tasks, I hope to provide insights to help you find your perfect AI model.