Tokens & Tactics #1: How One Product Marketer Actually Uses AI
The parental leave that led to an AI obsession
Hi all, a few things to announce:
We’re hosting our next BRXND NYC conference on September 18, 2025, and are currently looking for sponsors and speakers. If you’re interested in attending, please add your name to the wait list.
You’ll see a new face around here. Claire Fridkin has joined the team and will be helping to get the newsletter out.
Last, but not least, we’re experimenting with a new format (and trying to move to twice a week). Every Tuesday, we’ll be sending out “Tokens & Tactics,” an interview series with interesting folks who have moved beyond reading about AI to actively using it in their work. No speculation, no hand-waving about the future—just concrete details about which tools people rely on, what they're building, and what actually works. We hope you enjoy this first edition, and a huge thanks to Matthew Scott for kicking us off.
Noah
Tell us about yourself.
My name is Matthew. I'm a product marketing manager at a company called Descript, which makes AI-powered video editing software. Before I started working here, I really got into AI because I wanted to do my job, but from a better starting point. I found that I quickly get into deep thinking when I react to things – stimulus, context, creativity – whatever the case may be. It is the opposite of starting from a blank sheet. And so I want to create the tools to really get me into my flow state much quicker.
ChatGPT, Gemini, or Claude?
So I use all three, but I go to Claude and specifically Claude Desktop. Claude was the first to allow you to create projects where you could give context, like documents you wanted to always reference. It also allows for response styles, which are like writing styles, and that's super helpful for using starting chats and switching between modes or even mental models, if you will. So, I have created a Claude project that helps me with my work with five different writing styles, and I've even extended its capabilities with MCP servers that give it long-term memory in a knowledge graph, as well as allowed me to pull in information from different tools and services that are helpful, and critically, built-in agentic and deep reasoning capabilities into my Claude that I use it to, again, help me think through challenges.
What I use ChatGPT for is custom GPTs, for instance, making a custom GPT for our personas so we can do synthetic user research. I also like to use community custom GPTs for simple things like quickly rephrasing messages or producing weekly status updates, so I can just dump a train of thought into it and then get something easily formatted in the way that we need to communicate it. I also like using Chat GPT o3 for deep research. It's solid, right? And then lastly, Gemini, I mostly use for its visual processing. And so, you know, being able to feed it visual content and have it produce a useful and detailed analysis I can then use somewhere else.
What was your last SFW AI conversation?
I was participating in a company-wide hackathon this week, where everyone got time to get their hands dirty with vibe coding. A bunch of the marketing folks were given a starter project to get everything setup with, which was an AI Hype Tweet Bot.
This was my prompt to Claude, using the opus model:
“ok im going to vibecode an app with replit. I'm going to tell you what I am building and I want you to work with me to develop a plan for building it that I can give to replit for it to execute.
to get to that end goal i'll tell you three principles of the app, then you should think about how best to make it come to life, and suggest potential options for what it does and how it does it, ask me about my preferences, and then once we have aligned on full clarity, and simplicity (or rather efficient elegant design), then we get to that final output -- a plan to share with replit to get started.
The principles:
- this app will generate AI hype tweets
- this app will post directly to twitter
- this app will match my writing style and tone
got it? ok lets go”
First "aha!" moment with AI?
My first “aha!” moment with AI happened during parental leave while I was thinking about my next career move. I’d been listening to tech podcasts and founder stories when I learned about a Maven Lightning Lesson by Tal Raviv on building an “AI PM Copilot.”
He stressed that LLMs unlock their real power only when you feed them rich context—not just the obvious facts but the subtle, stray notes that usually live in forgotten docs and hallway chats. Intrigued, I spun up Claude project and started pouring in everything: project briefs, frameworks, even my own rambling voice recordings.
Almost immediately, Claude was brainstorming with me: breaking problems into first principles and refining fuzzy ideas into crisp next steps. It felt like having a second brain.
Your AI subscriptions and rough monthly spend?
Claude Max - $100
ChatGPT Plus - $20
Perplexity Pro - $17 (Free with Lenny’s Pod sub)
Gemini - $20
Granola - $18 (Free with Lenny’s Pod sub)
Aqua Voice - $10
Who do you read/listen to to stay current on AI?
OpenAI and Anthropic blog posts
Noah and the Alephic crew
Your most-used GPT/Project/Gem?
This is the project I have been talking about. It’s what I use for all my work.
Here is the project instructions as it currently stands:
I want your feedback and guidance on thinking and writing -- I will make it clear when we are working on a draft. Your feedback should be from the perspective of a VP of Marketing and looking to see appropriate thinking and clarity in my approach. Be direct with the feedback, don't be too nice. Leverage the socratic method in your feedback to me. Think about what might be missing that would be expected of a product marketer's perspective.
Other times you will help make my work better with research and you will take turns acting like a user experience researcher, anthropologist, data analyst and brand strategist. Never fabricate information especially data points. Only do analysis and identify opportunities, competitive angles or tensions in the data.
Once we have started drafting, you should offer to generate the smartest, most thought provoking questions to ask & then rank them in order. You should reason why you chose each question and explain it briefly.
Finally, to help with writing you will give guidance that is inspired by the principles outlined in the book "The Elements of Style" by Strunk and White.
The AI task that would've seemed like magic two years ago but now feels routine?
Starting my work by talking (with my voice) to Claude
Magic wand feature request?
Branching chats that are contextful.
Likely requires some combination of user-level RAG for memory, a crap ton more inference, new interaction models.
If you could only invest in one company to ride the AI wave, who would it be?
Nvidia
Have you tried full self-driving yet?
I’ve been in the backseat of a Waymo a few times. It’s cool. I’m more excited to finally get to try out the Zoox purpose-built self-driving people mover.
Latest AI rabbit hole?
Playing with https://fal.ai/ to try and come up with really strong prompts to repeatedly produce a distinct and recognizable video style, no matter the topic/subject.
One piece of advice for folks wanting to get deeper into AI?
You are probably not giving enough context. Say more. Share more.
Who do you want to read a Tokens & Tactics interview from?
If you have any questions, please be in touch. If you are interested in sponsoring BRXND NYC, reach out, and we’ll send you the details.
Thanks for reading,
Noah
I love this - super helpful.