My journey with Artificial Intelligence didn’t start with a curious afternoon in front of a computer — it began in a meeting room.

At the height of the AI boom, with headlines and social media buzzing about ChatGPT and other breakthrough models, my company made a strategic decision: integrate AI to improve user-facing information, streamline internal processes, and explore innovative interaction methods.

Initially, my role wasn’t to “play” with the technology — it was to understand it, test it rigorously, and figure out how it could fit into our workflow. What I didn’t anticipate was that this professional challenge would spark a personal curiosity, leading me to explore different platforms, models, and applications far beyond our original scope.

Today, AI is not just part of my corporate projects — it’s woven into my daily work as a developer. Tools like GitHub Copilot helps me to write code faster, catch errors before they happen, and find more creative solutions to complex problems.

OpenAI: The First Big Leap

The first tool we tried was OpenAI’s ChatGPT. We started with GPT-3.5 and soon upgraded to GPT-4 and GPT-4o, chasing better accuracy and reasoning capabilities.

I soon realized that the OpenAI patform is an excellent starting point for AI exploration, there I experimented with everything from simple chat prompts to sophisticated assistants, audio generation, and image creation.

In our business context, what impressed me most was ChatGPT’s ability to transform raw data into clear, user-friendly insights. However, we quickly encountered our first major challenge: token limits. This forced us to implement strategic workarounds, including:

• Semantic summarization to extract key points from lengthy texts

• Data vectorization using MongoDB to store embeddings and perform fast, precise semantic searches — ensuring we sent only relevant information to the model

While OpenAI didn’t fully meet our generative AI expectations, it remains central to our workflow, particularly for image generation — a capability that Anthropic doesn’t offer but one we rely on to complement our solutions.

Anthropic and Claude: The Reflective Choice

As we looked for models to complement GPT, we discovered Claude from Anthropic. Handling file formats — PDFs, PPTX, Excel, CSV, and plain text — took time to figure out. Unlike OpenAI’s highly structured API, Anthropic required us to build custom preprocessing and data loading workflows.

Despite these initial challenges, Claude delivered impressive results in semantic accuracy and contextual understanding, producing responses that more closely aligned with our team’s expectations. After comprehensive evaluation, we chose Anthropic as our primary model for day-to-day work.

Amazon Bedrock: The Integrator We Set Aside

Amazon Bedrock caught our attention with its promise of accessing multiple models from a single platform. But our data workflows didn’t adapt easily to its requirements, and the learning curve was steep compared to the value it offered.

In the end, we set it aside for our current use case — though it could be a good fit for simpler projects or prototypes that need to experiment with several models without heavy customization.

The good part of this was that bedrock allowed us to play with different platforms and models like Deepseek, Gemini and Mistral among others, this exploration reinforced our decision to focus on OpenAI and Anthropic as our primary AI partners.

Looking Forward

What started as a corporate initiative ended up transforming how I work and learn.

Our strategy is now clear: Anthropic Claude as the central model for its precision and contextual understanding, OpenAI to support areas where it shines (especially image generation).

In my daily work as a developer, GitHub Copilot and the AI in WebStorm have become indispensable assistants — helping me code faster, maintain cleaner code, and solve problems more efficiently.

If there’s one lesson from this journey, it’s that no single AI is perfect. The key is knowing how to combine the right tools for each need. AI doesn’t replace experience — but when used wisely, it amplifies it.

There is so much more to say about my journey with AI. From … how I understood this whole universe, the interconnection and the underlying base of all this, to … the low level tune mechanisms, assistants, reasoning models, vectors, and much much more.

But that will be another post!