Let's assume that the Product Management Industrial Complex is real and that we — Product Managers — are all wheels within wheels of wheels. Which, I truly think we are!
We're living in unprecedented times for our craft and it needs to evolve.
You may think that evolution for Product Management is an agent taking our jobs, but I choose to believe it's actually PMs leveraging agentic AI to a radical extent.
First, lets think about what the core tasks are that a PM does:
Product strategy, in the form of research, brainstorming and documentation
Product execution, in the form of prototyping and project management to some degree of helping teams complete tasks
Communication, in the form of stakeholder updates.
I am sure there are dozens of other core tasks, but those seem like a logical three that take up 90% of a PMs day: Strategy, Execution, and Communication.
Agentic Product Strategy
It's difficult to admit but LLMs are very good at writing pretty good strategy documents, particularly when given plenty of context. As a proxy of this if Opus 4.6 can one shot a prompt like "build me a markdown centric task manager in the form of trello" and output a working application, I am pretty sure it can write a solid product strategy document around that prompt too.
Agentic Product Execution
We know that LLMs can code to an exceptional degree, it's important to note however that in companies with PMs, it's probably not as easy as simply executing a prompt and committing code — the sophistication of workflows etc is almost certainly prohibitive to PMs agentically writing the code that engineers agentically write code.
That said, table stakes would be that every product memo now has robust wireframes, prototypes, and so on.
Similarly for the project management side of execution, LLMs — particularly in conjunction with agents — could be a dominant force at managing and running sprint planning.
Agentic Product Communication
It's difficult to admit but a huge amount of a PMs role is attending meetings and sending stakeholder updates. With modern call recorders there is no reason why LLMs cannot do all the work "around" a meeting — preparation notes, transcriptions and action items. I think automating these tasks is a huge opportunity.
Similarly, given the LLMs have insight into the execution (and metrics) creating stakeholder updates seem like another potent task to automate.
What an Agentic PM would look like
I think less about an agentic PM as the literal tasks and more about how to give it an incredibly amount of context:
For every project, have structured markdown directory that ingests all possible context related to that project:
Automated call recordings for any meetings that appear relevant
Product specifications and engineering documentation in agent consumable formats (Markdown, not Google Docs!)
Slack transcripts of relevant channels
Similarly, a suite of tools that allow the agent to plug into where work happens:
Ticket management (Jira, Linear, etc)
Metrics.
Most importantly this agent would have a prompt that allows it to check in on all this context on a recurring basis. For example every day it may check all communications and understand key action items to deliver on, similarly on a weekly and monthly basis it may compile all the decisions and changes made so far — creating an automated newsletter that can then be sent out.
Building this
I wanted to build this at Pinterest but I have been amazed at how much of this critical context is completely inaccessible to LLMs. This strikes me as a very solid indication that the field is actually very far behind the "embrace AI" mantra it so often repeats.