Like many of you, I started experimenting with ChatGPT out of curiosity.

Initially, it was a helpful assistant, assisting with tasks ranging from writing emails to drafting summaries and even brainstorming ideas for presentations. I got good at writing prompts. I learned to coax better answers by refining my questions, layering context, and iterating until the output was just right.

But here’s what I quickly realized:

Being proficient at using ChatGPT doesn’t mean you understand how large language models (LLMs) and Generative AI (GenAI) really work, nor what they can and can’t do to transform a business.

I’ve had the privilege of leading global teams, driving SaaS transformations, and delivering meaningful outcomes. I’ve seen firsthand how technology waves come and go, from the early days of web software to mobile, cloud, and subscription models, but what’s happening now with AI is fundamentally different. It’s not just a new tool; it’s a new paradigm for how businesses think, operate, and create value.

And so, earlier this year, I made a decision: If I wanted to lead in this new era, I needed to go deep.

Phase 1: Understanding the Machine

Evenings and weekends became my classroom. I enrolled in a technical course on GPTs and the transformer algorithm — the architecture that underpins large language models. What fascinated me most was how the course stripped away the mystique of “AI magic” and forced me to understand the inner workings of GPT2… in Excel. Yes, Excel.

I learned how tokens are encoded, how attention mechanisms evaluate context, and how large datasets transform into predictive power. It was both humbling and exhilarating to witness the inner workings behind the seamless conversational interface of ChatGPT.

However, understanding the theory was just the beginning. Once I grasped how the system functioned, I couldn’t help but wonder: What implications does this have for fundamental business transformation? How can we go beyond prompt engineering and leverage AI to rethink the way organizations operate?

Phase 2: From Curiosity to Creation

Nothing teaches you faster than hands-on experience. I decided to roll up my sleeves and build something. Not just a prototype, but a real, production-grade AI-first mobile app for both iOS and Android. I wanted to experience the end-to-end journey: from ideation to design, from coding to launch, from prompt tuning to marketing.

I chose Flutter as my development framework — the same technology I’d used in 2022 to build Coaching Log, an earlier app that predated the age of LLM copilots. Back then, Stack Overflow was my best friend. This time, I had new partners: ChatGPT, GitHub Copilot (powered primarily by Claude Sonnet), and a suite of AI agents that would become my virtual team.

Phase 3: Building an AI vs. AI Debate App

To truly understand how to fine-tune and orchestrate LLMs, I looked at going beyond a simple chatbot. The idea was to create an app where two AI personas debate a topic from different perspectives. Think “Einstein vs. a social media teenager” discussing the benefits of spending the day on social media, or “Sherlock Holmes vs. a Formula 1 driver” dissecting the ways to win an ancient Roman chariot race. And yes, I also wanted a “simple chatbot” to interact directly with one of these personas.

I called the project Aideai, pronounced eye-de-eye, short for AI-debates-AI, but also a subtle play on “idea within AI” considering that idea is written アイデア (aidéa) in Japanese. (Yes, I spent far too long in a ChatGPT session refining the name, but that’s part of the fun.)

Each persona required a detailed “system prompt” that defined its worldview, communication style, and tone — effectively, its character DNA. To populate the app, I built a custom GPT to generate the persona code in Dart, pre-populate prompts, and set variables such as creativity levels (models refer to this as Temperature, TopP, and TopK). That’s how I ended up with over 550 unique AI personas without losing weeks to manual work. And yes, I used AI to help me write AI — an experience that felt like standing on the edge of a new creative frontier.

Fast forward a few months (and about 40,000 lines of code later): Aideai 1.0 is live on both Android and iOS, complete with multiple subscription tiers, an LLM-based debate engine, a language detector, and even a hidden “release notes generator” that writes in the tone of your chosen persona (that one stays under wraps for now). 👉 Check it out here.

Phase 4: AI as a Co-Worker

Building Aideai wasn’t just about creating a product; it was about learning how far AI could stretch as a collaborator, greatly expanding my productivity and capability.

Coding with AI

When I started, GitHub Copilot was good at writing one function at a time or providing answers to my questions. It was a significant productivity boost, but not a transformative one. Fast forward a few months, and the Agent mode can create new features, refactor entire sections, generate unit tests, write detailed documentation, and even debug its own code.

The improvement curve is astonishing. As someone who never coded full-time (and is very rusty), I found myself iterating faster than ever, implementing UX and features in hours instead of weeks. It was not because I was coding better, but because AI was handling the syntax and structure while I focused on architecture and business logic.

That said, AI is not infallible. Hallucinations aren’t bugs; they’re a feature of probabilistic models. Leave it unsupervised, and it will confidently generate broken code or quickly descend into a rabbit hole. The human in the loop remains essential: not as a typist, but as a director.

AI as a Brainstorming Partner

When it came time to brainstorm for Aideai, ChatGPT became my creative sparring partner. We explored everything from simplifying the app name to user onboarding flow or ensuring each persona’s tone felt distinct yet consistent. I used it to pressure-test ideas, challenge assumptions, and refine the system prompts behind the scenes to achieve insightful output through multiple iterations, shaping how personas debated, summarized, or adjusted their tone depending on the context. Those sessions pushed me to think more like an AI engineer and less like a user, understanding how subtle fine-tuning of foundational models could dramatically alter results.

It was a vivid demonstration that AI can act as both collaborator and catalyst, amplifying creativity when guided with clarity. The more thoughtfully you articulate your goals and iterate on structure, the more valuable its insights become.

AI in the Marketing Loop

Beyond development, AI played a role in almost every aspect of Aideai’s go-to-market effort, from early messaging frameworks and feature descriptions to visual concepts and competitive positioning. I utilized ChatGPT sessions to explore branding, craft tone-of-voice guidelines, and even refine the tagline, messaging, and imagery, ensuring resonance, clarity, and fun. It helped me build a cohesive narrative, align value propositions, and accelerate content creation across web, social, and in-app materials.

Was it perfect? Not at all. Every output still needed a human touch, a sense of brand authenticity, empathy, and narrative flow (and I’m not even counting the effort it takes to remove all these emdashes!) Yet AI brought structure and velocity to each phase, enabling me to move from ideation to execution 80–90% faster! It was a massive shift in productivity and focus.

Phase 5: Lessons Learned

Looking back, diving deep into AI wasn’t just an intellectual exercise. It was a leadership transformation.

1. Integration is everything: If you haven’t woven AI into your daily workflow, from marketing to analytics and to product development, you’re leaving efficiency and insight on the table.
2. Prompting is a muscle, not a formula: There’s no “magic template.” The secret is iteration; it involves refining your prompts, just as you would refine your product strategy.
3. Trust, but verify: Never take AI output at face value. Hallucinations are by design. The best practice is supervised creativity.
4. Context is the multiplier: LLMs thrive on context. The more relevant and specific your input, the more valuable the output will be.
5. Understand the economics: Every input and output token has a cost. Optimize prompts for efficiency, just as you would optimize cloud spend.
6. Governance isn’t optional: Using AI responsibly involves safeguarding confidentiality, keeping sensitive data secure, and adhering to organizational guidelines on privacy, data handling, and intellectual property. It requires understanding AI’s information processing and maintaining boundaries to protect internal data while utilizing AI’s capabilities.
7. The human edge does matter: AI can generate, but only humans can discern meaning. Authentic leadership in the age of AI is about striking a balance between both.

Phase 6: From Personal Experiment to Organizational Imperative

My Aideai project started as a late-night experiment, a way to “learn by building.” However, it has evolved into something much larger: a comprehensive working experiment in how AI can accelerate innovation, augment creativity, and transform workflows end-to-end.

And it’s reshaped how I think. As leaders, our role is shifting from managing resources to orchestrating intelligence (human and artificial) to deliver value faster, more accurately, and with greater empathy.

Adopting AI at scale isn’t about replacing people; it’s about amplifying potential. The organizations that embrace AI as both a tool and a mindset will define the next decade.

So, What’s Next?

I’m still learning. I’m still experimenting. Currently, I’m delving deeper into Chip Huyen’s book “AI Engineering”, exploring topics such as multi-modal models, RAG, and agentic systems.

The leaders who truly understand how AI works, beyond what it outputs, will be the ones best positioned to shape the future rather than chase it. We’re still at the beginning of a new era where AI won’t just make us faster. It will make us better decision-makers.

If you’re not already integrating AI into your daily workflow, start now.

And if you are, keep pushing. Refine your prompts. Question the output. Test, learn, iterate. The future belongs to the curious.

💡 How are you using AI in your day-to-day work?

Have you discovered any surprising ways to accelerate your workflow or make more informed decisions? I’d love to hear your experiences. Let’s learn from each other.