Curious generalists can do specialist work when AI fills the gaps
With only 10 percent coders for 55 million users, a startup lab model lets staff stretch past their specialties.
Entrepreneur · Are AI-First Teams the Secret Ingredient Behind the Next Billion-Dollar Startups?The co-founder of HOLYWATER, a tech company, says his team of 200 includes only about one in ten coders, yet its platforms reach 55 million people worldwide. He says, "Our team of strong generalists enables 3.5 times annual growth."
His how-to advice is to hire for curiosity, not credentials, and to run the business as a place for trying ideas. He says, "We treat the whole company like a startup lab," where writers, editors, designers, and marketers can all pitch product ideas. A generalist, in his account, is someone who fixes problems quickly alone or finds the right colleague to help.
The day-to-day method differs by role. He says his "content operations manager uses AI for brainstorming and training models to mimic writing styles." He says his "brand designer supplements creative skills with AI-generated assets using focused prompts." He says his "full stack developer uses pair programming with tools like Cursor," an AI coding assistant that works alongside the programmer.
He recommends hiring for curiosity and letting every employee pitch product ideas. He says, "That's how we can launch products faster, test more, learn from data and grow." He adds, "AI won't replace specialists who know how to work with it," and says innovative tools amplify team productivity.
Questions people ask
What is Cursor and what does pair programming with AI mean?
Cursor is an AI-assisted coding program that works inside the text editor where programmers write code. It can autocomplete lines, explain errors, and draft whole functions from a plain-language request. Developers often describe using it as pair programming because it feels like a junior partner watching and suggesting.
How do you train AI to mimic a writing style?
It means feeding the AI examples of a particular voice, such as short, punchy episode descriptions or warm marketing emails, so it learns the patterns. The manager still edits and approves the output. Done well, it speeds up routine drafting but needs human review to avoid bland or off-brand copy.
When does this approach fail?
Start with one small, low-risk task and a clear measure of success, such as a draft in two days instead of two weeks. Pick a tool built for that job, keep a specialist available to review the work, and keep private or customer data out of public tools. If quality drops or review takes longer than doing the work, it is not a fit.