I'll show you my AI PM stack, if you show me yours

Welcome to Product Cocktail, where the takes are as polarizing as a shot of Fernet—but the insights come together like a perfectly crafted daiquiri.

I'm not gonna bullshit you: I've avoided writing this newsletter topic for almost a year.

Last summer, with the proliferation of AI tools across a broad range of subdomains (LLMs, AI note taking, vibe coding, automation, etc.), it felt logical to put out a perspective on what I like. It was certainly something I'd talked about "offline" extensively.

But the draft sat, and sat, and sat.

It felt daunting to write about. "AI PM Stack" is such a loaded concept. It's such an unsolved problem. The stack is constantly changing as I learn more (or Anthropic ships another 700 features in 45 days). What if I put out an antiquated (3-day old) take? Would I be relegated to managing legacy database products at Oracle? (Shout out to my 30k homies laid off six weeks ago via email. You stay classy, Larry Ellison.)

Ultimately, I decided to do this to push myself, to make myself uncomfortable. We're living during an unprecedented pace of technological innovation (barf, I actually stole that line from a Deloitte Tech Trends report). This is a moving target. I'll try to hold myself accountable to update this list 3-4 times a year.

Daily Driver AI Chatbot: Gemini + Claude.

I was an early-ish subscriber to ChatGPT Plus—May 15, 2023 for those keeping score at home. I’ll be celebrating our three year anniversary by getting a divorce.

Gemini 3 felt like a big leap forward, on par or exceeding ChatGPT in many ways. As a daily driver, I appreciate the baked-in Nano Banana (image gen) support, which historically felt much more seamless than ChatGPT's image gen. Admittedly I haven’t used the updated GPT-5.5 model which improves on this. I'm using Gemini for parenting advice, lifestyle questions, and cocktail ideas (if it wasn’t obvious from <gestures around this email>, that’s what I do when I’m not mastering the AI PM newsletter “universe”).

Starting around January, Claude quickly took over as my default thinking partner for my professional life: job search positioning and interview prep coach, newsletter thought partner and editor, and technical advisor for AI coding projects.

All of these tools have their limitations — right now my biggest annoyance is with Claude’s usage limit ping pong. But tied for a distant second are ChatGPT’s weird clickbait engagement strategy ("If you want, I can also tell you the one thing in your resume that might accidentally get you filtered out by Klaviyo’s recruiter (even though you're qualified).") and Gemini's non-sequitur personalized memory gymnastics (“Since you've been troubleshooting your home WiFi lately, would you like me to find some "easy-sipping" recipes you can enjoy while you're working on the network?”)

Market Research: Perplexity Pro (Deep Research) and Claude Opus 4.6 (Extended Thinking).

Remember how fun it was to have 75 tabs open while writing a paper in college to research a topic, read all of the sources, synthesize the reseach, keep track of everything, and then actually write your paper? Me neither. That sucked.

I’ve shortcut that entire process by dumping a research prompt (usually created with help by an LLM or agent) into Perplexity or Claude, then letting it spin for 20 minutes while I go have a beer.

I come back to a markdown file with sources cited. I can scan the reports, ask follow-up questions, double-check any numbers myself, and even download the reports to incorporate into my project directory (for use by another AI tool like Claude Code).

Obviously, check their work and ask thoughtful follow-ups to actually understand and internalize the research.

AI Meeting Notes: Granola.

Imagine having a McKinsey associate in all of your meetings taking incredibly precise meeting minutes which you can then ask questions about afterward, except it’s free-99 instead of $500/hour.

Banger. Enough said.

Rapid Prototyping: Lovable and Replit.

“What is everyone even building with this shit?" is probably a fair question. I won’t answer it satisfactorily here. Nor will I delve into the “should PMs be touching production code?” debate, other than to say that stakeholders-or-engineers-misunderstanding-a-PRD is probably more common than tone-deaf CEO takes on LinkedIn.

The AI era has produced a communication cheat code in the form of AI prototyping. If you can describe it, there are no shortage of options to mock up a functioning prototype—faster than your designer can say “we need to design a new component for that.” These products fall along a spectrum of user-friendly “slop cannon” to terminal-based “follow the white rabbit” developer tool.

Lovable and Replit are both browser-based, easy to use, and are a great starting point for prototyping.

Lovable can do design-forward prototypes really well from the jump. My “holy shit” moment with Lovable was one-shot-prompting a clickable prototype of a “wedding bets” platform in less than two minutes—while on the bus from bridal party pictures to the ceremony at my sister-in-law’s wedding. (Arguably, that is too little friction, but let’s table that discussion for the after party.)
Example project: https://nuptial-bets.lovable.app

Replit is a fast way to ship a legit app with a working backend. One friend of mine built an entire game on Replit and deployed it to the App Store. Between its library of integrations for database/authentication/hosting to its parallel agent workflows, it is a powerful tool. Just watch out - you can easily burn tokens if it gets stuck in a loop on something.
Example project: https://tobrien88.github.io/generated-linktree/

AI Coding: Claude Code.

Claude Code is a terminal-based AI coding agent, meaning that it's run from the command line on your laptop. Now you can run it from the Claude desktop app.

While CC enables you to modify existing codebases of actual… code… easily, the PM and business unlock is thinking about what an AI agent could do with your day-to-day docs. Think: point CC at a folder of transcripts and synthesize user interviews, automatically create action items and engineering tickets from your meeting transcripts, manage a knowledge base about your entire job (that can recall specific decisions on demand without searching through Google Drive hell).

To put the capabilities in a way even the most out-of-the-loop PMs could understand: imagine if Outlook search actually worked, but could understand your emails and automatically create prioritized JIRA tickets in your backlog?

AI Chief of Staff: OpenClaw running on a cloud virtual private server.

I use an OpenClaw agent as somewhat of a chief of staff—helping ideate on newsletter topics (and saving notes down to my desktop), scheduling things, managing my inbox newsletter deluge, and other tasks.

Allow my Better Call Saul-inspired AI agent to say a quick hello:

Honorable Mentions:

Reply with your own stack—what you actually use, not what’s on your LinkedIn skills section. If I'm impressed by your cocktail, I’ll feature it in a future issue.

_{Icons made by Icongeek26 from www.flaticon.com.}