By framing the current landscape as a “New Polyglot Wave,” I’ve moved past the anxiety of finding the “perfect” tool. Instead, I focus on building an intentional workflow where the best AI tool is sometimes three (or none).
I was born around the same time as the web. I grew up watching my dad fix computers for neighbors, renting video games from the library, and getting swept up in the frantic energy of Y2K. Watching technology rapidly evolve isn’t just my career; it’s been one of my favorite lifelong hobbies.
During my interview with Sparkbox, I was asked how I felt about AI and how I was currently using it in my workflow. I answered candidly: I had a few concerns. I worried about the impact on our environment and the way our society’s unaddressed biases were being encoded into new tech. But ultimately, I was excited.
But when it came to my personal workflow, I remained cautious. For a long time, I treated LLMs like ChatGPT and Google Gemini as research assistants, keeping my AI workflow strictly browser-based and at arm’s length. I hadn’t yet made the jump to in-editor solutions like Cursor or terminal-based agents like Claude. To be honest, I wasn’t sure I wanted to.
That changed through my participation in the AI Expedition team at Sparkbox and my work with various clients. It reached a tipping point: clients were writing .cursorrules files for their projects, and it simply made sense to sync our workflows. But after diving in, I found myself facing a new question: Which tool was “better?”
Eventually, I realized that “best” is a moving target.
Learning from the Last Waves
I spent a lot of time overthinking the “correct” approach until I realized that while AI is new, the rapid onset of disruptive technology isn’t. We’ve been here before.
In 2006, Neal Ford wrote about the “Polyglot Programmer,” arguing that as programming languages expanded, the future would belong to those who used multiple languages within a single codebase to solve different problems. In 2011, Martin Fowler applied this to databases with “Polyglot Persistence.” He argued that we should stop trying to force every piece of data into a relational table and instead use the right database for the right job.
Framing AI as a New Polyglot Wave immediately helped me reframe the question. I didn’t need to pick a “winner.” I just needed to categorize each tool.
While AI is new, the rapid onset of disruptive technology isn’t. We’ve been here before.
My AI Persona Framework
Instead of looking for one “perfect” AI, I’m building a specialized stack. Every project is different, and sometimes the best approach is to use a few of them together.
Cursor is the Surgeon. Since it lives right inside the code, it’s great when you’re in the weeds. It’s excellent for making small, precise changes, but the trade-off is that it can get too “zoomed in” and lose track of the broader project architecture or a long history of changes.
Claude Code is the Architect. This is where I go for complex reasoning. It seems to think more deeply and has a better grasp of system-wide logic. It might take a bit longer to process, but it’s what I use to figure out the blueprint.
Gemini and other browser-based LLMs are Librarians. They have access to a staggering amount of information. They’re perfect for explaining a new API or digging through documentation, but you still have to do the manual work of implementation.
Strategic vs. Utility
Fowler warned that “when relational databases are used inappropriately, they exert a significant drag on application development.” I think AI is the same. It’s about using the right tool for the right job.
He used a framework called the Utility/Strategy dichotomy. Most tasks are “Utility” in that they’re necessary, but they don’t really differentiate your work. For utility tasks, I just want the fastest, simplest tool. But “Strategic” projects are the ones that actually move the needle for a client. Fowler argued it’s worth over-investing in these. For a complex coding task, that might mean a polyglot approach like using Claude to figure out the logic and Cursor to make on the ground changes.
The biggest benefit of this mindset is that it keeps me from falling into the “Golden Hammer” trap. When you’re only comfortable with one tool, every problem starts to look like a nail. Once I stopped trying to find the “best” AI, I actually got better at recognizing when I didn’t need one at all. Sometimes the best way to solve a logic puzzle isn’t an LLM: it’s a whiteboard, a quick chat with a teammate, or just twenty minutes of focused coding. Part of the Polyglot Wave is knowing that sometimes the right tool for the job is just your own brain.
Technology is going to keep evolving rapidly, just like it has since I was a kid watching my dad fix computers. I’ve realized that the goal isn’t to be an expert in one specific tool that might be obsolete by next month. It’s about staying ready for whatever comes next without betting on any single one. For now, I’m going to keep doing what I’ve always done: stay curious and keep experimenting. I want to make sure I’m reaching for the right tool for the job, not just the one I already have open.
