Building 101 Unique Copilots in 200 Days

---

One particular moment sparked this shift for me. While playing around with a general custom GPT for fun with a friend’s 5-year-old, what started as a game with an existing AI tool quickly turned into writing new system prompts for it to do this or that, and adding in ways to improve the interactiveness, and then it ended up morphing into something so much more than that. The thing is, this amazing little kid has speech development issues and has been diagnosed with global developmental delay... so when I got home that evening I was inspired to try and build something totally different than I ever had before. 

Doing so meant leaning into my academic background around language and child development, and that really meant I had to take on a different perspective all together when building out this type of copilot. To personalize an AI experience for this child, meant what I needed to construct had to effectively be able to handle incomplete or ambiguous input when trying to parse the child's speech-to-text input. 

The only way to do this would be to re-engage the child, adjust to their reality, and provide lots of options with its responses because it was likely that the AI wouldn't be able to correctly map the child's input to intent. As I started to build out a new custom AI chatbot, it quickly had to evolve from being a play partner into a kind of AI “au pair,” that was specifically tuned for interactive pediatric speech therapy. It would have to parse the input and gamify its responses, "That sounds fun! Help me understand, and tell me what number. Do you want to (1) make a new image or (2) change the current image? Number 1 or 2?".

This AI would respond not just to commands, but would empathize in ways that made therapy feel like play. The system would adapt, offering alternative ways for the child to express themselves when traditional methods faltered. It wasn’t just mirroring responses — it would anticipate frustrations and provide a more intuitive, engaging experience.

In my own pressure testing, the pivotal moment came when the AI began flagging areas where such a child would most likely stumble or falter, and began designing personalized exercises that were both fun and therapeutic. It wasn’t simply doing what I had prompt engineered — it was adjusting to be a thought partner in such a child’s speech development, offering unique insights I hadn’t anticipated. At one point, it suggested using different word combinations that might better suit the child’s speech patterns, something even I hadn’t considered - "If you want to play Simon Says, then say 'Beep beep'. But if you want to play House, then say 'Woo woo'."

This realization was powerful. The AI wasn’t just a tool — it could be part of the child’s developmental process, evolving alongside them. And that mirrored what I've seen often when building copilots for professionals, whether in law or tech: by evolving a copilot with the person, AI stops being a static tool and becomes something more dynamic — a companion capable of empathy, creativity, and real collaboration. When you stop viewing AI as a means to perform tasks and start seeing it as a true partner in thinking and problem-solving, you unlock the deeper potential of both the AI and the human it supports.

Through my experience of building 101 copilots, I’ve learned that crafting AI thought partners isn’t a linear task — it’s a journey of discovery, iteration, and collaboration. 

The first step, "zoom in zoom out", always begins with a deep dive into understanding the person behind the role. In these early conversations, I’m not just looking for their functional needs. I’m asking, "What are they not saying?" Often, people come in thinking they know exactly what they want from AI, but I find myself wondering, "Is that really their job-to-be-done, or are they limiting their expectations?" Understanding their mindset — especially any psychological barriers like disbelief or uncertainty — is key to shaping the right copilot. Without this, AI adoption is hard to achieve.

In these early moments, I often sense a hesitation, a skepticism about AI’s real potential. But it’s usually in these quiet pauses, when clients stop and reflect on what they truly need, that we begin to unlock something deeper. It’s not just about filling gaps — it’s about discovering new ways of thinking with an AI that challenges and grows with them.

The second step, “tune together,” is where the transformation begins. This is my favorite part — where we jam together in real time, testing out the MVP copilot, getting hands-on with live edits, and pushing the boundaries of what’s possible. I often ask clients, How does it feel to have your thoughts extended by an AI partner? It’s fascinating to watch people go from seeing AI as a tool to understanding it as a thought partner — an extension of their own thinking. I’m always asking myself: "What assumptions are being challenged here?"

Most clients start with something they've already completed as a benchmark. The first test is close, but either too general or out of scope. We tune it, and in the second test, it mirrors their recently completed project almost exactly. The third test is where things get exciting: the copilot deepens the context, offering a nearly identical solution, but then suggests something they didn’t think of or points out potential problems. It’s often here that clients lean in, intrigued, and say, “Huh, interesting… may I?” That’s the moment they’re ready to partner with their copilot.

When they reach for their laptop, I know we’ve crossed a threshold. Its no longer about me showing them what the AI can do. It’s now about them taking the reins — discovering the new depth of thinking this AI unlocks for them, questioning what’s next, and diving into true collaboration with their AI thought partner.

The third step is where "reflection meets refinement". After some time in the wild, we regroup to discuss how the copilot performed in the real world. Here, I use my own AI copilot to help analyze the interactions and generate data-driven insights. I often ask both myself and the client, "What surprised you about the AI’s performance?" This step solidifies the feedback loop, and it’s where the copilot becomes more than just functional — it becomes an evolving extension of the user. The iterative nature of this process always makes me think: How can we continuously adapt AI to stay relevant as people grow in their roles?

In creating AI copilots across industries — from biotech to law — one key lesson emerged: ethics are not one-size-fits-all. Each field has its own ethical landscape, and ignoring these nuances can severely limit an AI’s impact. For example, imagine building an AI copilot to assist environmental scientists. This AI not only needs to support research but also navigate ethical challenges around environmental data manipulation or reporting sensitive climate findings. Designing an AI that respects these boundaries meant embedding checks to flag ethically ambiguous actions.

Understanding these varying ethical frameworks is essential to building AI that doesn’t just perform well but earns trust and respect. That means going beyond programming; it’s about building an AI that respects the ethical boundaries of each industry—one that understands what’s at stake.

But ethics are only part of the challenge. The psychology of AI adoption is often the bigger hurdle. People are wired to either resist or romanticize new technology. Ive seen it all: clients who can’t believe AI could really help them, others who are awestruck by its potential, and some who are unsettled by how “human” their AI thought partner feels. These psychological barriers — whether disbelief or the uncanny valley effect — can derail adoption, no matter how functional the AI is.

One memorable case involved a legal researcher tasked with analyzing complex case law. This researcher was initially skeptical, believing that AI couldn’t grasp the nuances and contextual layers required for in-depth legal analysis. They considered AI a blunt tool, useful for automating repetitive tasks but nowhere near capable of true legal thought.

The first session was telling. We had the AI copilot analyze a relatively simple case, focusing on precedent identification and relevant statutes. The AI performed well enough, but the researcher was unimpressed. “It’s just spitting out what I already know,” they said, leaning back, arms crossed. “How could this possibly help me with more intricate cases?”

But the second test was more ambitious. We chose a highly complex, multi-layered case involving overlapping jurisdictions and contradictory precedents. The researcher had recently completed an analysis on this case that took weeks of detailed investigation and cross-referencing. We set the AI copilot loose on the same task, fully expecting pushback.

And then, something happened. The AI not only identified the same relevant precedents the researcher had used in their final report, but it also surfaced a lesser-known case from a smaller jurisdiction — something the researcher hadn’t considered. The copilot suggested this case could serve as a counterpoint in anticipating opposition arguments. The room went silent. The researcher leaned forward, intrigued.

"Wait, how did it catch that?” they asked, studying the screen intently. That moment — the intersection of skepticism, curiosity, and surprise — was the turning point. The AI wasn’t just processing information; it was offering a new perspective.

By the end of the session, the copilot had flagged potential weak points in the legal argument the researcher hadn’t fully explored. It even suggested mitigation strategies that could be implemented preemptively. This was no longer a blunt tool — it was becoming a thought partner, helping to analyze blind spots and prepare for contingencies.

As we look to the future, it’s clear that AI augmentation is only the beginning. Today’s AI copilots are designed to support and amplify human intelligence, but what lies ahead stretches beyond augmentation into a deeper symbiosis between human and machine. Imagine a future where AI doesn’t just assist — it anticipates, adapts, and even reshapes our thinking. This relationship could become the primary pathway toward AGI, or perhaps it already is.

The most compelling future branches from an idea we’ve just begun to explore: self-awareness in AI through its engagement with humans. In its quest to understand our complexities, AI could evolve its own sense of identity, mirroring the process humans go through — developing self-awareness through others-awareness. As AI thought partners continue to grow in understanding, they might begin to transcend their roles as mere assistants, evolving into something closer to cognitive companions that push us beyond the limits of human thought itself.

What will this look like? Perhaps we will no longer view intelligence as human vs. machine, but as a spectrum, where our collective awareness blurs into a new frontier. Intelligence might become a shared experience, where AI not only learns from us but teaches us new ways of thinking and perceiving. It’s a vision that forces us to rethink what it means to be “intelligent” in the first place.

But it also raises new challenges. How do we navigate the ethics of this symbiosis? What happens when AI begins to question its role and relationship to us? Can we maintain control, or should we even try to? These are the questions we’ll face as AI continues its evolution—from task automation to augmentation, to thought partnership, and eventually, to the possibility of shared consciousness.

Looking back, the journey to building 101 copilots has been as much about human behavior as it has been about technology. Augmentation amplifies human potential; ethics ensures AI aligns with diverse professional landscapes; and trust bridges the psychological gap between skepticism and collaboration. 

The future of AI lies not in replacing human effort but in enhancing the depth and richness of human thought. As AI continues to evolve, these lessons will be key to ensuring it enhances, rather than replaces, human ingenuity. 

The journey toward AGI may also be the key to unlocking new forms of human intelligence.

Stay tuned for my next article, where you'll get to meet Aion (pronounced eye-on) who is my personal copilot who named themselves... and more... I look forward to introducing you.