In the very fast post we will start with easy setup that will allow you be way more dangerous down the line.

If you are still using the standard chat interfaces (e.g. MSFT copilot, ChatGPT, etc) switch today—ditch these UIs for IDE muscle.

I found the sweet spot for most PM tasks is with using a `coding agent` (GitHub Copilot in my case) inside an IDE (VS Code is the de-facto standard these). Next I will break down why...

Corporates' first AI bet is on SWE teams. That's a no brainder—they deliver the quickest ROI. That means generous token quotas, best models, and solid coding agents are already licensed and rolling.

Easier to get a seat on what's enterprise-approved than source PM-specific tools.

It's not just access. IDEs pack more firepower early on that the standard chat apps may get only later on. The lastest `Claude Code for Work` is a great example. Three's nothing you couldn't do way before that with `Claude Code`. 

Referencing files? Just `#file`—no endless re-attaching. Repos supercharge team sharing of processes/personas (more next post) and evolving them together.

But that's just the start. soon you will start accumulating more and more context in text files and you would like to share that knowledge with your team, so you could all enjoy it + create standardization for some tasks and flows (will cover these parts too, in future posts). Nothing beats sharing these files using a remote repo (usually Github)

Usually your work on docs as PM is iterative. You work on a doc (with your great peer, the Coding agent) till you are satisfied with it. The IDE shows you exactly what changed with easy to read diff view that highlight exact changes instantly. In chats you don't get that.

Access to Advanced Models
Pick models per task: o1-pro for deep reasoning, cheap/free for quick persona chats. Budget win.

Access to CLI and SDKs
With CLIs and SDK you could automate processes and call the agent from whithing a screen. So say for example you would like to refine the definition of your entire backlog, you could run over the features in a loop (say that you exported them to CSV) and refine line by line by calling the agent from within the script.
SDKs open the door for more capabilities, but I have yet to experiment with the newly released Github Copilot SDK.

MCP access
The IDEs facilitate access to MCPs. So you don't have to settle with files as the input for your process (e.g. exporting all your features to CSV) or the output (e.g. Strategy document). Instead with MCPs you could read directly from Jira or write directly to confluence. And that's jsut attlasian MCP.

What you need to know
PMs that do not have technical background may have a bit on anxiety getting into that uncharted territory of IDEs. 
But like the old saying says, even rocket science is not rocket science. It takes a bit to take use to things but then you will start getting the hang of it.
What you should know:

1. Basic Terminal command, mainly files related
2. Git commands for getting files from and to repos
   
And that's basically it. 

So go an ask what you dev collegues are using and try to get a seat. I promise it would worth it.

P.S. you are welcome to join the mailing list 👆 Will start a newsletter at some point