How Cursor built my GCP browser extension in 30 minutes
Originally posted on LinkedIn, January 27, 2026.
I have never built software this fast. And it keeps getting faster.
- 10:00: Cursor open on an empty folder.
- 10:05: I finished writing the prompt. Yes, I spent at least 5 minutes on it. That is the secret.
- 10:30: While I was doing other things, Cursor finished building a browser extension that improves the GCP interface.
It searched the web, researched the best technical approach based on my instructions, tested its work, built the extension, and finished by telling me which folder contained the code I needed to load.
I spent the next few iterations refining the design, only because I had not invested those extra minutes in the first prompt. Two more prompts got me to a result so impressive that I would have paid for a product like it.
It started with a Production incident. Someone, no names, I resisted tagging you and you know who you are, pasted the wrong Json into Secret manager and crashed production. How to manage that is a separate conversation, so I will stay on topic.
The incident made me wonder how Google, the company that rules this planet, has such a terrible interface. Literally no one among its million employees had done anything about it.
So I put on my save-humanity-from-terrible-interfaces cape again, armed myself with the best AI I could find, and it worked like magic. Even I was surprised that the first working version took only one iteration. All it needed was a good prompt and a good model. Opus, you save lives. This is nothing less than a quiet revolution.
The more I work with AI, the clearer it becomes that it is not just a code-writing engine. It is a technical partner for research and working through decisions.
My prompt started with the words, “I have a problem.” After describing the problem, I poured in every piece of information I could provide so it would have as much context as possible: links to our projects, the contents of our guides copied from Confluence, previous attempts, and similar projects.
I also shared how I thought the problem might be solved, perhaps with a browser extension. But I did not ask it to follow my idea. I asked it to research the problem itself and propose a plan. I explicitly told it to take the time to learn, investigate every part, and develop the fullest possible picture of how to build the whole thing from end to end. Only then did we start building.
Less than 3 years ago, I built a browser extension that improved Jira’s interface by myself. It was even covered by Geektime, for those who remember. I spent a few weekends wrestling with the code to make it work.
Today I can create the same kind of solution between meetings, using a technology I had never worked with before: Plasmo.
The barrier to technical creation has been removed. I can build freely and solve problems, small and large.