A month ago, I built an educational website for children about India (now live at exploreindia.musingmithun.com) without writing a single line of code. My contribution had little to do with programming and everything to do with defining goals, evaluating results, and guiding the process.
The objective was simple: create a site that would help kids learn more about India—its states, geography, culture and tourist attractions. I deliberately chose a straightforward project because I wanted to understand the capabilities of the tool rather than spend time defining complex requirements.
Like most first-time users, I didn’t get the desired outcome immediately.
It took a few iterations before Claude began producing results that resembled the vision I had in mind. As I worked through those iterations, I learned that the success of the project did not need me to write any code at all. I did not even know which scripting language to use, or how the files would be organized. My role was to articulate requirements, evaluate outputs and provide feedback.
The feedback loop was almost instantaneous. I would describe a change, Claude would implement it, and within minutes I could determine whether the result was moving closer to what I wanted. The result was a functioning website that would have felt far beyond my capabilities just a year ago.
One of the earliest versions of the site included a map of India that looked more like a collection of labelled boxes stacked on top of one another than an actual map. Claude had followed the request. The implementation was technically complete. The outcome, however, was wrong.
Without some understanding of what India actually looks like, it would have been easy to accept that output and move on. Instead, several rounds of feedback were needed before the result became useful.
The same thing happened when it came time to publish the site. Claude could suggest the steps required to host the application, but concepts such as domains, DNS records and hosting platforms still required a basic level of understanding of data networking. AI could guide the process, but it could not replace the judgment needed to execute it.
This is where I think many conversations about AI miss an important point. AI does not eliminate expertise.
As I worked through the project, I realized that the Product Manager in me was becoming more important, not less. The challenge was not generating code. The challenge was understanding what would be useful for the target audience. Every iteration required judgment, context and an understanding of what good looked like.
If implementation becomes easier, identifying the right problem becomes even more important. Organizations have never suffered from a shortage of ideas. They have always struggled with prioritization. The ability to say “no” to ten good ideas in order to focus on one important problem becomes more valuable when the cost of building falls. AI may reduce the effort required to execute an idea, but it does not reduce the need to decide which ideas are worth pursuing.
Perhaps that is one of the more profound effects of AI. The technology makes execution more accessible while increasing the value of judgment, user understanding, prioritization and the ability to recognize what good looks like.
Perhaps our professional identities will evolve in the same direction.
Mithun's Musings 