Interesting take, but I actually think we’re heading the opposite direction. If AI makes 'building' a commodity with near-zero cost, then 'Vibe Coding' is just a faster way to reach a dead end if the vision is blurry.
The real shift for PMs isn't becoming 'junior devs with a prompt window.' It’s becoming Decision Architects. When a designer can ship 10 prototypes in an hour, the PM’s value isn't in helping build them—it’s in the high-level analysis, judgment, and logic required to pick the right one.
This is a solid article about Cursor's workflows for PMs. What caught my attention was the AGENTS.md approach for onboarding AI with codebase context. I have been doing something similar with my own setup, maintaining context files that give the AI persistent knowledge about the project structure, decisions, and constraints. The difference between a fresh AI conversation and one that "knows" your codebase is night and day.
The "Plan mode" point resonates too. I have found that the biggest productivity gains come not from letting AI code faster, but from having it think through the approach first. When I built Wiz, my personal AI agent that runs continuously on Claude Code, the planning phase became essential. Without it, you end up with technically correct code that misses the actual problem you were trying to solve.
One thing I would add to the "the problem is usually your prompt" observation: it is also about knowing when NOT to prompt. After testing Cursor, Google AI Studio, Antigravity, and several others, I noticed that the best results came from giving the AI breathing room to explore rather than micromanaging every step. I wrote about this pattern in my comparison of AI dev tools if you want to dig deeper into the workflow differences: https://thoughts.jock.pl/p/cursor-vs-google-ai-studio-antigravity-ide-comparison-2025
Just started trying out cursor, almost zero coding background, as a product designer, and I can feel how powerful it is, anyone who says this isn't the future is lying, and I will die on this hill 😂
I love that the world is collapsing into the IDE. There is a knowledge arbitrage opportunity here, in that non-engineering people will see Cursor or IDE and say "I cant" or "not for me" and they will pass on an insanely powerful system they can use to make their life easier.
Thank you. This is helpful. I agree with the 4 things you list for "What makes this actually work?" They are all about giving the AI agent the tools and especially context. I would add a fifth context: For a major feature, don't just prompt it, iterate with it on a plan/spec for that feature. Taking the time to go back and forth with AI to hone your plan for the feature makes a huge difference in the quality of the output. It doesn't need to be a book, but laying out the goal, features, use cases, diagram if needed, data model if needed, implementation plan ... we have seen the quality of output leap proceeding in this way. The more and faster the human-AI iteration the better.
Interesting take, but I actually think we’re heading the opposite direction. If AI makes 'building' a commodity with near-zero cost, then 'Vibe Coding' is just a faster way to reach a dead end if the vision is blurry.
The real shift for PMs isn't becoming 'junior devs with a prompt window.' It’s becoming Decision Architects. When a designer can ship 10 prototypes in an hour, the PM’s value isn't in helping build them—it’s in the high-level analysis, judgment, and logic required to pick the right one.
This is a solid article about Cursor's workflows for PMs. What caught my attention was the AGENTS.md approach for onboarding AI with codebase context. I have been doing something similar with my own setup, maintaining context files that give the AI persistent knowledge about the project structure, decisions, and constraints. The difference between a fresh AI conversation and one that "knows" your codebase is night and day.
The "Plan mode" point resonates too. I have found that the biggest productivity gains come not from letting AI code faster, but from having it think through the approach first. When I built Wiz, my personal AI agent that runs continuously on Claude Code, the planning phase became essential. Without it, you end up with technically correct code that misses the actual problem you were trying to solve.
One thing I would add to the "the problem is usually your prompt" observation: it is also about knowing when NOT to prompt. After testing Cursor, Google AI Studio, Antigravity, and several others, I noticed that the best results came from giving the AI breathing room to explore rather than micromanaging every step. I wrote about this pattern in my comparison of AI dev tools if you want to dig deeper into the workflow differences: https://thoughts.jock.pl/p/cursor-vs-google-ai-studio-antigravity-ide-comparison-2025
Just started trying out cursor, almost zero coding background, as a product designer, and I can feel how powerful it is, anyone who says this isn't the future is lying, and I will die on this hill 😂
I love that the world is collapsing into the IDE. There is a knowledge arbitrage opportunity here, in that non-engineering people will see Cursor or IDE and say "I cant" or "not for me" and they will pass on an insanely powerful system they can use to make their life easier.
Thank you. This is helpful. I agree with the 4 things you list for "What makes this actually work?" They are all about giving the AI agent the tools and especially context. I would add a fifth context: For a major feature, don't just prompt it, iterate with it on a plan/spec for that feature. Taking the time to go back and forth with AI to hone your plan for the feature makes a huge difference in the quality of the output. It doesn't need to be a book, but laying out the goal, features, use cases, diagram if needed, data model if needed, implementation plan ... we have seen the quality of output leap proceeding in this way. The more and faster the human-AI iteration the better.
Yeah, being able to create 7 low fidelity wireframes in one minute, and do it repeatedly is an insane level up for product designers who can code.
No more excuses for poor UX in our apps anymore.