The PM role isn’t dying. The PM who only writes PRDs probably is.
I used to think I was a fairly hands-on PM.
Spoke to users.
Wrote detailed PRDs.
Made decent wireframes.
Could write basic SQL.
Sat with engineering during grooming.
Knew enough technical words to survive most discussions.
Then I tried building an actual working product with AI and realised how much of my job still ended at “someone else will figure this out”.
The product was simple on paper.
An internal tool where the sales team could upload call notes and get:
- repeated objections
- competitor mentions
- follow-up tasks
- accounts at risk of going cold
- patterns across the whole week
I had already made a PRD for something similar at work.
Around 11 pages.
User stories.
Edge cases.
Success metrics.
Flow diagrams.
Very proper.
During a GrowthX build week, I decided to actually build a rough version instead of writing one more document.
Used Claude for the first backend.
Generated a basic UI.
Connected a transcription API.
Added a database.
Got the happy path working in around two days.
Upload call.
Get summary.
See action items.
I was feeling quite smart.
Then two engineers reviewed it.
First question:
“What happens if the same call is uploaded twice?”
Had not thought about it.
“What if the transcript is 90 minutes?”
No idea.
“How are accounts being matched if the salesperson writes the company name differently?”
Not handled.
“Why are you sending the entire transcript to the model every time someone opens the account?”
Because that was the easiest thing to build.
“What does the confidence score mean?”
Whatever Claude had decided it meant.
That review was uncomfortable because none of these were exotic engineering problems.
They were product decisions hiding inside technical implementation.
In the PRD version, I could write:
“The system should intelligently identify duplicate calls.”
Sounds good.
When building it, I had to decide what duplicate actually means.
Same file?
Same salesperson and timestamp?
Same transcript?
What if a call gets uploaded once through Zoom and once manually?
I could no longer hide behind one elegant sentence.
Same with “surface important objections”.
Important according to whom?
Frequency?
Deal value?
Sales stage?
Whether the customer sounded angry?
Whether the competitor was mentioned?
The prototype forced me to confront how vague a lot of my product thinking was.
AI did make many PM tasks much cheaper for me.
Research summaries are cheaper.
First-draft PRDs are cheaper.
Wireframes are cheaper.
SQL is cheaper.
Basic prototypes are much cheaper.
Writing ten versions of microcopy is basically free.
But cheap does not mean solved.
The more AI handled the output, the more important these became:
- choosing the correct problem
- defining what good output means
- knowing which edge cases matter
- understanding technical trade-offs
- deciding what should not be automated
- getting the workflow adopted
- knowing whether the user will change behaviour
- deciding what happens when the model is wrong
AI helped me build a tool.
It did not tell me whether salespeople would trust it.
It did not tell me whether managers wanted summaries or only wanted missed follow-ups.
It did not tell me whether the company would allow customer calls to leave its system.
It did not tell me whether reps would correct bad output or quietly stop using the product.
Those are still product problems.
I do not think every PM now needs to become a production engineer.
I still cannot review backend architecture properly.
I am not going to pretend I can replace an experienced developer because I got a prototype running.
But I do think the bar has moved.
A PM who can build a rough workflow, put it in front of users and understand where it breaks will have a much better conversation with engineering than a PM who arrives with a 14-page PRD and says “technical feasibility to be assessed”.
The GrowthX experience was useful because nobody cared about how nicely I had framed the problem.
I had to show the screen.
Someone clicked the wrong thing.
The flow broke.
An engineer asked why I had made a decision.
I could not answer.
That taught me more than another session on AI product strategy would have.
My current view:
PM is not becoming less important.
A lot of PM output is becoming less valuable.
The PMs who survive will probably spend less time proving they are organised and more time proving they have judgment.
Curious how this is changing inside Indian companies.
Are PMs in your org expected to prototype now?
And when people say PMs should “become builders”, how far do they actually mean?
Functional prototype?
Basic code understanding?
Production ownership?
Or just better collaboration with engineering?