“I was highly skeptical, but now I’m a believer.”

That’s a phrase I find myself repeating constantly in recent conversations about LLMs and AI, and it turns out I’m far from alone! I’ve heard variations of this statement multiple times in the past few weeks.

Using and abusing

My journey with AI started a long time ago. I learned about ML during my university years almost 20 years ago, but then parked the topic for many years until a couple of years ago when we got taken by storm by the LLM revolution.

OpenAI launched ChatGPT in November 2022. I’ve, like probably many others, played with it and enjoyed testing its boundaries. From creating stories for my daughters to exploring its knowledge and its biases.

Chatting with this brand new thing was fun and exciting, but also a bit limited, so much so that I stopped using it the next month.

A few months later I decided to give it another go (Friday, June 23, 2023, 1:10 PM according to my credit card transaction). This time it was via APIs, and I kept using it on and off until April 6, 2025 for various tasks.

In the meantime, I was also exploring other LLMs and tools, from Claude to Gemini to open source models like Llama 2. I’ve tried vibe coding, image generation, video generation, music generation, you name it.

All of this to say that my skepticism wasn’t driven by ignorance or unwillingness to try it out, but rather by a genuine assessment of its capabilities and limitations.

From skeptic to believer

So, what changed my mind?

My perspective has evolved over the past couple of months due to three main things:

  • Models got better, and “Agentic mode” became a thing: Advances in AI models, particularly the development of agentic capabilities, have enabled applications that simply weren’t feasible before.

  • The rise of Spec Driven Development: The emergence of Spec Driven Development (SDD) gave me a structured framework for integrating LLMs into development workflows. It adds context that vibe coding alone can’t provide, making the whole thing actually useful.

  • I found my groove: I identified specific use cases where LLMs demonstrated undeniable value. Initial successes built my confidence, which led to broader experimentation and, honestly, just getting better at this whole thing.

It all started with a simple:

npx bmad-method install

followed by:

/pm *help

What followed were 4 weeks of intensive exploration and experimentation with various features and capabilities of the tool.

I took on a real project, something I wanted to build for a while, and I used BMAD to help me design, implement, and test it.

The result? A fully functional application, way more than the mere prototype I was shooting for. The application had tests, documentation, CI/CD pipelines, and even monitoring setup. Easy peasy, I said… okay, it wasn’t that easy, but the outcome surprised me.

That wasn’t my only experience, of course. I’ve since tried other projects with BMAD, tested BMAD v6, Kiro from Amazon, and Spec-Kit released by GitHub.

I do have opinions about all of them, but I’ll save that for another post 😅.

The lessons learned

This experience forced me to shift my perspective on AI and its potential applications. I’m still a strong believer in the need for a human in the loop, from the initial spec definition to the final review of the delivered code.

I’ll also admit that the flow isn’t perfect yet. It doesn’t always go as smoothly as I’d like; different models and tools have their quirks and limitations.

Developers might not enjoy having to review all the initial designs and specs written by an AI. They might find it frustrating to review all the generated code. But it’s undeniable that, in the right hands, this approach can significantly boost productivity and creativity.

Looking ahead

I won’t blame anyone for being skeptical about AI since I was right there with them not long ago. But I, along with some friends, decided to start this blog because we believe we’ve finally found something interesting and worth sharing about AI and development.

So, if you’re still reading, give it a try maybe? And while you’re here, consider bookmarking this website or subscribing to our RSS feed. We might have more insights to share in the future.

The journey from skeptic to believer is a powerful one.

In the spirit of transparency, I even used an AI to help refine the phrasing of this very post—a small but practical example of the value I’ve come to appreciate. *

Want to talk about this? You can find me on LinkedIn or Twitter,