Pat Bergstresser

I love this app.

@petergyang @addyosmani Be careful.

A gift to OpenClaws everywhere thanks to @addyosmani (fwiw gog still rocks)

@aakashgupta “This is not an officially supported Google product.”

Google just gave your AI agent a way to access every Workspace API that doesn’t eat half your context window. Here’s the problem everyone’s been hitting. The standard way to connect Claude Code or Cursor to Gmail, Drive, and Calendar is through MCP servers. Google ships official ones. They work. But MCP has a structural tax that gets worse the more tools you connect. One developer measured his Google Workspace MCP setup: 142 tools. ~37,000 tokens loaded into context. That’s 19% of a 200k context window consumed before the agent even starts thinking about your task. Another developer reported MCP tools eating 98,700 tokens total, nearly 50% of their entire context, and asked Anthropic for help. Cursor hard-caps you at 40 MCP tools because the problem is so bad. The CLI approach sidesteps this entirely. Your agent reads a lightweight skill file, calls gws drive files list via shell, parses JSON back. The tool definitions never enter the context window. Same capabilities, fraction of the overhead. But the architecture goes deeper. This CLI reads Google’s Discovery Service at runtime and builds its entire command surface dynamically. Google adds a new Workspace API endpoint, the CLI picks it up automatically. Every static MCP server is permanently one version behind. Google’s own blog post announcing managed MCP servers admitted the previous state was developers “identifying, installing, and managing individual local MCP servers, often leading to fragile implementations.” This CLI is Google’s answer to their own problem. One npm install. 100+ agent skills. Encrypted credentials. And if you still want MCP as the transport layer, gws mcp starts a server over stdio. The real signal: as agents get smarter, the bottleneck is shifting from “can it access the tool” to “how much context does accessing the tool cost.” CLIs win that math every time.

@stuholden I can’t speak to whether this exists or not. But if you build it yourself you’ll want to use something like @Replit @boltdotnew @Lovable . With a few prompts you can get a prototype running that works perfectly for you. No code needed at all.

Via Claude, Gemini, ChatGPT, etc. Thanks for reading and contributing— kind of excited to pull this together and experiment for next Summer!!

Interested in open sourcing this and see where it goes: After the last World Cup, I transitioned my broadcast boards / notes from Paper Stickers to Digital. For the last few years I’ve used combinations of GoodNotes, Google Sheets, Slides, etc and nothing quite fits what I want…

@_Logor @Replit Haha true but I figure I would have done all the same steps anyway. Just faster and more efficient now

@PatThePM @Replit It really is but it zaps Credits! 😩

.@Replit agent v3 is the real deal folks…wow

This is one of those skills that a lot of people in tech think they have. But people who can do this effectively are incredibly rare unicorns.

@JoshuaImel @okayzade Thanks for flagging @JoshuaImel ! Now I don’t have to bounce back and forth with a different LLM. And the context available will be helpful!

@okayzade Check it out @PatThePM 👏

been working on this! you no longer need to yell at Agent to not touch any code!! plan in peace

@prajeenv Well said

@PatThePM This is real.. It is not about hustle, it is about mental ownership. When the product becomes part of your thinking, everything you do starts sharpening it ..

"PMs need to work 60+ hours a week to be successful." It's bullshit. What isn't bullshit? PMs need to *think* about their product 80+ hours a week. When you do this, your 40-50 hour work weeks become much more optimized than everyone else who's slogging 60+ hours.

@ParvSondhi Haha very true

@PatThePM 100% but so freaking hard to do. lol. so easy to get sucked into that envy loop

Been thinking about this tweet a lot recently

@BasedBigTech @aakashgupta This.

@aakashgupta They’d also be sandboxed, not choosing tools on their own (v0, etc). No way it would be choose your own adventure. So something is off.

PM Interviews are changing to test vibe coding

@carlvellotti IMO the EM should be running these frankly. If they aren’t, influence that change. Worst case, PMs should be there to answer questions and ensure the user POV is maintained.

Running grooming meetings can be stressful for PMs, especially if you aren't technical. Here's an excellent template spotted on Reddit.

Like AI music to my ears

Why are so many vibe coded products just regurgitations of things that already exist? - Habit tracking - To do lists - Fitness - etc. Where’s the creativity?! Non-technical people now have the ability to actually be creative and so many aren’t actually being creative.

@aakashgupta Personally fan of these types of cases. (Albeit cases are imperfect) In the end, PMs are evaluated on impact. And usually that’s measured through a metric. Being able to connect the dots between a product, the needle it moves, and how that impacts the business = foundational.

If you're preparing for PM interviews in 2025, there's one question type you cannot afford to mess up: Metrics. Here's the history of how it over took product hiring and why it's the silent killer of PM dreams:

Also FWIW, average Reddit CTRs for ads is 0.2-0.4%. So all 3 were within a normal range.

My take on next steps: - Speed is the winner - Having 1 leading val prop enables positioning clarity - Overall all 3 were close - Lead copy with speed, but leverage other two val props as supporting

In order to better understand how to position Vibe Updater relative to PM pain points, I ran a positioning test on r/ProductManagement. Which do you think won? Creatives in the first reply (can't attach to poll). Answers in the second reply.