Last week I revealed where I set hard boundaries when it comes to my AI usage, ie my AI Non-Negotiables.

In a world where we are being pressured to 10x our productivity and automate everything, I thought it was important to start off where I do not use AI.

As a female founder + community lead of AI Snack Club, I am hyper-aware of not only my own AI usage, but where members are most interested in making AI work for them in a meaningful way.

So today, I am excited to bring Part 1 of where I most definitely use and actually love having AI support me!

Subscribe now

We’re partnering with The 10th House at Female Founder Collective on this one and if you’re serious about AI actually working for you as a founder or founder-curious, this is worth five minutes of your time.

Are you a female founder building or scaling in tech? 🤖

Applications are now open for Tech Cohort ‘26 inside The 10th House at Female Founder Collective. Designed for women founders building across AI, SaaS, FinTech, HealthTech, Consumer Tech, and emerging technologies—from pre-revenue to $3M+ in annual revenue.

Special offer for the AI Snack Club community: Receive an exclusive 10% off Growth or Scale membership at The 10th House using code AISNACK10.

→ Explore Tech Cohort ‘26

⏰ Applications close July 22, 2026. Spots are limited and the cohort will be capped to ensure a high-quality experience.

1. Websites

The greatest advancement of AI is its ability to generate world-class code from natural language (ie plain English, like talking to a friend). And as a failed Computer Science major, I have taken full advantage of this capability to build websites and apps from just a few sentences.

And the journey has been incredibly fun, rewarding, and liberating!

I don’t need to learn a new website builder tool and I don’t need to hire a website designer or engineer to build it for me.

Instead, I can bring my visions to life using two of my favorite tools: Claude Code and Lovable.

What I have built (publicly) so far:

• AI Snack Club’s Official Website

• Founding Member LinkedIn Graphic Generator

• A fun app that helps Female Founders feel empowered to say “no”

2. Community Member Portal

One of the biggest AI undertakings I took on this year was building Snack Club’s Community Member Portal from scratch.

This was a huge and daunting task, not only because I was building something that essentially already worked pretty well in Notion, but I had to make sure it was a) safe and secure and b) worked for over 1k members.

While the pressure of becoming a Product Manager and Product Engineer overnight was real, it was a challenge I was excited to take on.

As they say, “Practice what you preach” and I had to do just that, using Lovable of course.

3. Decks and Presentations

As a former management consultant of 8 years, I am both blessed and cursed with the ability to spin up decks and presentations in PowerPoint like nobody’s business.

Except now as a Founder, I am relieved that I don't have to anymore, thanks to AI.

My two favorite tools for building decks are:

• Gamma - The OG AI deck builder that has a billion templates and image generation options. It is also extremely easy to create decks from bullet lists, summaries, and pdf documents

• Claude - My favorite trick is adding a few design reference images from Pinterest and then have Claude build me the deck in Google Slides

4. Dashboards

Building quick and simple dashboards to display business and personal information and metrics is my favorite use of AI.

• Need to understand my business financials on a monthly and yearly basis? Build a dashboard.

• Need to monitor and track all my subscriptions? Build a dashboard.

• Need to review how many members join Snack Club? Build a dashboard.

For building quick little mockups, I usually turn to Claude to spin these up as one-time use artifacts. And for building more permanent solutions, I usually turn to Lovable and integrate external tools directly to the dashboard for live reporting.

Today’s article focused on where I use AI to build things. Part 2 will focus on where I hand off the responsibility to AI and use it like an intern.

Let me know where you find the most value with AI.

xx,
Monica

Let’s Stay Connected

🧃View upcoming Events

🧃Join the Community (Use code SUBSTACK10 for $10 off the yearly membership!)

🧃Follow me on LinkedIn

🧃Follow me on Instagram

🧃Follow me on Substack

Hey hey, if we haven’t met yet, I’m Monica, and I run AI Snack Club. We are a community for women to learn AI together.🧃

If you are looking for a place where AI is fun, less overwhelming, and super accessible - you have come to the right place!

I have been using AI daily for well over a year now, and through my experiences of actively building a business with it - trying and failing with AI along the way - here are the instances where I think AI should take a backseat.

While this might sound a little backwards, I think there are cases where you should definitely use AI and cases where it’s just not necessary.

Let’s get into it!

Subscribe now

1. Emails, DMs, Texts, Slack, e.g. Personal 1:1s

I don’t like to use AI to write messages on my behalf where I’m interacting one-on-one with another person, especially if it’s someone new I am trying to build rapport with for the first time.

When I have such a small window to connect with someone, I try my best to be genuine, authentic, thoughtful, and (most importantly) straight to the point.

Adding just a sentence or two to relate on a human level can make the world of a difference:

• “I saw you recently went to a conference, how fun!”

• “Congrats on the new role!”

• “I saw we were mutuals with X, small world”

• “Your headshot outfit is so cute! Where did you get it?”

And on the flip side, when I get emails that are clearly written by AI: my brain goes blank. I skim through it. And truthfully, I barely read the entire message. Because it feels 1) lazy and 2) way more verbose than necessary.

Advice: If you find it hard to find the right words, instead of going straight to Claude to type it for you, try using a voice to text tool like Wispr Flow (*referral link, you get a free month, and I get a free month) to speak the message! This is a great way to get your tone and actual thoughts on paper, without relying on AI 100%

2. Strategy Plans

As tempting as it is, I try my best to not go straight to Claude and ask it to make me a strategy plan without coming up with my own ideas first.

What I like to do instead: put my laptop on airplane mode (or just close it entirely), bring out a notebook and pen, and brain dump.

I do this for two reasons:

For one, I think there’s a lot of merit in coming up with my own ideas (even if it feels like my ideas are really, really bad). Why? Because our thoughts are coming from our own unique, lived human experiences (vs. what a robot thinks).

AI will never come up with a novel idea. It is just pulling from a repository of information from training data.

As much as it feels like I am missing out on limitless potential ideas from Claude, I believe that we should all trust ourselves a little more on what we actually come up with, and that our instincts are clear for a reason (even if not totally obvious at first).

And two, if you think about it, Claude is giving a similar generic response to everyone who asks the same question. So you’re not going to come up with anything actually unique by starting there.

I love to go to Claude after I have all my thoughts down on paper, and ask it to help me organize my thoughts, pressure test them, or play devil’s advocate - but I try not to start with Claude right off the bat.

3. LinkedIn Posts and Hooks

Okay, this one might be a little controversial 👀

As someone who posts on LinkedIn two to three times a week, I am super adamant about making sure it’s my own voice.

Once you ask Claude to write you a few LinkedIn hooks and content, you’ll start seeing the exact same phrases it gives you, ALL. OVER. YOUR. FEED.

It can be super generic, it’s not eye-catching, and people will keep scrolling.

My advice: Drive curiosity, but provide value quickly - don’t wait to deliver the “punch” several paragraphs down or you will lose people’s attention.

I also love seeing photos that scream “hi! I am a real person”

4. Long Form Content 🧃

I am the first to admit that I am not the best writer. And that is okay! But I do like to write my own work, because it’s really important that we don’t lose our own unique voice. And that I challenge myself to get the words out on paper so I don’t lose this valuable human skill of clear communication (especially as a founder, a business owner, and a community leader).

For me, it is the one thing that builds trust with my community and audience. It is a rare opportunity I get to share exactly what I am thinking and why. And if you’re just mass-producing the same content Claude is giving everybody else, it’s hard to build confidence with those around you. </3

So where do I use AI? I will be writing about that next!! But reply below letting me know your non-negotiables so I can add them to my list too.

xx,
Monica

Subscribe now

Let’s Stay Connected

🧃View upcoming Events

🧃Join the Community (Use code SUBSTACK10 for $10 off the yearly price)

🧃Follow me on LinkedIn

🧃Follow me on Instagram

🧃Follow me on Substack

Last week I promised you a step-by-step on how I design my logos. And here it is:

It took me 13 prompts to turn the Snack Club juice box into a disco ball. And the back and forth was frustrating, yet so rewarding at the end, so buckle up for the ride.

The tool, if you haven’t used it yet, is called Nano Banana, Google’s image model inside Gemini. It’s the one I use for all my image generations. Including this Substack’s graphic!

Your First Prompt Should Aim for 80% Accuracy

Here’s my exact first prompt w/ the screenshots I used for inspiration:

And here was the “80% there” result it gave back:

Not bad, but also not 100% of what I wanted:

• Annoyed by the grey rounded edge square box

• And the subtle background color on front of the juice box should be a lighter green color

Subscribe now

Lesson 1: Focus On One Thing at a Time

Iterate bit by bit - one little change:

“really good!! can you make the background all white?”

Then this:

“remove the grey boarder”

(Yes, boarder. I don’t proofread my prompts. Neither should you.)

Now we are cooking!!

Summary: One ask per message, like the background, then the border, and up next: tackling the color on the front of the juice box. If you stack five fixes into one prompt, the model mushes them together and you get just a few out of five, and you can’t tell which instruction it dropped, and worst of all, new stuff appears and it’s absolute chaos.

Lesson 2: Show, Don’t Tell

This is the one that took me longest to trust. Words are a terrible way to describe a color. And I did a terrible job of doing so:

“see how the juice box has a slightly different front color”

“can you remake the last image with a slight variation to match the screenshot”

“keep the juice box as is, but make the face like this image and have a white background”

This was the most frustrating thing to change:

Lesson 3: Be Specific, Not Polite

I was being too easy with it at this point — almost, slightly, not quite, can you — instead of being more assertive and forthright with what you want.

“almost - the straw should be red and white and the front still needs to be a disco ball (like attached)”

“the front needs a slightly lighter green color”

“not quite - the indent area needs to be light green and the rest in dark green”

“can you make the area behind the text and red apple to be a slightly lighter green color like attached”

And then this is where I lost my sh*t lol

Lesson 4: Don’t Give Up

Here’s the final chats of the conversation where it got the different color face of the juice box BUT decided to zoom in and add a black background …. WHY !?

“yes but zoom out so we see the whole juice box”

“yes! just zoom out and make it white background”

My advice: Take a break, then keep going!!

Because this was my 13th prompt and the result was MAGIC!!!!

“yes but make the area behind the text and red should be light green, in between the indents”

In Summary:

• Start with a goal of getting it 80% there. If it’s anything below that, open a new chat and start again. It is not worth the headache to change something that doesn’t have potential.

• Fix one thing at a time. It makes it easier to course-correct and see the changes (rather than doing them all at once)

• Paste a picture every time words fail you. If it’s not listening, go ahead and find something for it to look at.

• Don’t be passive. Name the exact pixel that’s wrong instead of being nice about it. Be specific!

Let’s Stay Connected

🧃View upcoming Events

🧃Join the Community (Use code SUBSTACK10 for $10 off the yearly price)

🧃Follow me on LinkedIn

🧃Follow me on Instagram

🧃Follow me on Substack

I moderated a panel earlier this week in downtown San Francisco where AI Snack Club brought together top female leaders in the design, video, and photo space to talk about their experience with AI as a complement, rather than a replacement to creativity.

The line that stood out the most to me:

“AI is so bad at being creative because it’s just copying the same basic design patterns over and over again.”

It is no coincidence Merriam Webster crowned the 2025 word of the year to be “Slop” (specifically “AI slop”).

In a world where anyone can design, build, and publish anything in seconds, it’s inevitable that the first prototypes and drafts all look the same. AI isn’t creating something novel for each person. It is providing the same, quick & dirty version to all its users.

For this new frontier of tech, I am wildly unimpressed.

Except for a few key players who are doing something different: adding in whimsy & personality.

Here are four doing it right.

Subscribe now

Town

What it is: a genuinely smart AI assistant.

I’ve watched a lot of companies try (and fail) at this. Usually I am either babysitting the AI more than I am using it, or it’s not smart enough and fixates on the stuff that doesn’t matter, ie my spam emails or random calendar holds.

Town is the opposite: intelligent, and most impressively proactive. It doesn’t wait for me to tell it what to do. It wows me with what it handles right off the bat.

The whimsy: when you join, you build your own Townie. I am now the proud owner of a mischievous juice box named Sippy. 🧃 A serious, well-funded product (they just came out of stealth with a $55M Series A led by a16z) looked at the AI-assistant category and decided the move was to be cute as hell, and that choice is exactly why I remember it.

Not to mention, sharing your cute little avatar in the Snack Club slack (join here).

Try Town.

Prompt Cowboy

What it is: a prompt helper, and (imo) the smartest one on the market.

If you have ever attended any of my AI 101 sessions, I often preach that you do not need to be a prompt engineer. You just need 2 good habits:

• Be clear, thoughtful, and provide context

• The Stop Trick: “Before you answer, what questions do you have for me”

And if those don’t work for you, my 3rd piece of advice: use a prompt helper.

The whimsy: Prompt Cowboy leans all the way into the western theme, so the smartest tool in the category also happens to feel like a good time.

Rodeo

What it is: Rodeo — your (non-work) life, wrangled. It’s where you save all the places, recipes, and ideas you actually want to do, so they don’t vanish into a camera roll black hole.

I found it the way I find most brands — endlessly scrolling TikTok, and the messaging got me: “stop saving things you’ll never find again.” #guilty

The whimsy: the cowboy energy carries through everything, and their recent campaign is chef’s kiss: SAVE IT so you can DO IT with your friends.

As a community builder whose #1 piece of advice is actually use AI (don’t just learn about it or talk about it), I feel a real kinship there.

Not to mention the team behind Rodeo is GOATED with some of the smartest, most creative people in NYC. <3

The Accountability Cat

What it is: a Chrome extension everyone needs to jumpstart their LinkedIn journey.

One of my favorite parts of leading this community is hosting IRL meetups across the US. I had the best time with the Chicago crew a few months ago running a vibe-coding session for first-time builders.

I also got to meet one of my longtime LinkedIn internet friends, Mya, in person.

There, she showed off one of the most adorable and functional extensions I’ve ever seen: the Accountability Cat (or dog, or bunny).

It nudges you to post on LinkedIn, builds streaks, and turns a usually-daunting task into something you actually want to do.

The whimsy: Mya made posting online fun and unbearably cute. She doesn’t have an engineering degree and she built the whole thing herself with Lovable and Claude. Insane.

Your turn: what’s the most delightful, made-you-smile AI thing you’ve found lately? Hit reply and tell me, or post in our member chat!!

And if you’re ready to make your own bit of whimsy: later this week I’m dropping a step-by-step on how I design every one of my own logos using Nano Banana.

Subscribe now

xx,
Monica

Let’s Stay Connected

🧃View upcoming Events

🧃Join the Community

🧃Follow me on LinkedIn

🧃Follow me on Instagram

🧃Follow me on Substack

If you knew a wave was coming, wouldn’t you want to prepare?

I vividly remember my freshman year at Georgetown University, back in 2013, when I started my journey as a Computer Science major. I was equal parts excited and nervous for this ambitious, new chapter in my life.

What I didn’t expect? Being so unprepared. Everyone in my Intro to Coding C++ class somehow already knew how to program applications and it felt like they were all experts when it came to setting up their terminals and tackling our first assignments.

It was terrifying and the whole first semester I felt incredibly behind my peers.

That feeling of walking in unprepared has stuck with me ever since.

I tell you this because the AI wave is here. And it’s not to scare you, but to say you don’t need to feel unprepared.

If anything, I would love for you to feel excited and energized by these new developments that allow anyone to build anything! AI is the future we are all learning together, on our own terms.

And now is the best time to start that journey, with a supportive group of women by your side.

🧃Introducing: AI 101 — a brand new Snack Bite Series

Starting in May, I’m rolling out weekly 60-minute, hands-on workshops designed to get you ready (and hopefully loving AI as much as I do).

The lineup, rotating on an ongoing basis:

• AI 101: The Foundations

• AI 101: Intro to Claude Co-work

• AI 101: Intro to Claude Code (Part 1: Setup)

• AI 101: Claude Code (Part 2: Build Something)

• AI 101: Intro to Vibe Coding

• AI 101: Build Your First Agent

• AI 101: Build Your AI Portfolio

Each session will run the same, with 20 minutes of demo, 20 minutes of small group breakouts (where the real magic happens), and 20 minutes of group sharing.

And YES, all sessions are recorded as I know life, work, play can be a lot right now!

If you can’t make it one week, the same session comes back around, with updated and new information as the tools and tech are changing on a daily basis. So feel free to join again and again.

RSVP Now!!

Please note these sessions are exclusive for AI Snack Club paying community members. The sessions build week over week and there are ongoing conversations happening in our private Slack, so we keep the room to members only.

If you are not yet a member, we would love to have you! Join the Snack Club and grab your seat now. And if you know someone who's been quietly drowning in tabs, send them this post. <3

xx,
Monica

Join AI Snack Club

This week’s newsletter sponsor

Want to get 90 minutes a day back?

Slack just became the new interface for work — where people, agents, data, and apps come together in one place.

The new Slackbot evolves from a helpful assistant into an active teammate that can summarize meetings, run reusable AI skills, and take action across 6,000+ apps without leaving Slack.

Teams at companies like Anthropic, Perplexity, and Rocket Companies are reporting up to 90 minutes a day saved — over two months of working hours back every year.

See how Slackbot turns intent into action inside the flow of work you already know.

Explore the new Slackbot →

Thank you NYC, LA, SF, and Chicago for showing up, chatting AI, and making Snack Club so special.

Upcoming IRL Events

• 5/13: Marketing at the Intersection of AI & Fashion with Daydream AI in NYC here

• 5/28: Female Founder Vibe Coding Workshop Lovable and AI Snack Club in SF here

Let’s Stay Connected

• View upcoming Events

• Follow me on LinkedIn

• Follow me on Instagram

• Follow me on Substack

After the most energizing week in NYC and a few days in sunny LA, I am feeling so inspired by all the incredibly talented women in our community who are fearlessly tackling AI head on.

A few stories that stood out:

• Leila just finished building out her side business site and booked her first client call - check out one of the most beautifully designed websites I have ever seen.

• Arden saves 8-10 hours a week on work with AI, giving her quality time back to spend with her 1-year old son (who is also ridiculously cute) - read about her baby-making and baby-raising journey.

• Traci built a personalized agent on claude code that actively bid, negotiated, and sold several items after her 17-year old cat passed away - these goods are looking for a new home

• Imani and Thamina are crushing it on Tiktok and IG Reels with daily helpful hints on career tips and how to turn your 9-5 income + equity into wealth & freedom - follow along here and here

• Brigitte built a virtual Loteria game to share with her Hispanic ERG at work.

And a million other stories and connections I could go on about.

Join the Community

This week’s newsletter sponsor

Want to get 90 minutes a day back?

Slack just became the new interface for work — where people, agents, data, and apps come together in one place.

The new Slackbot evolves from a helpful assistant into an active teammate that can summarize meetings, run reusable AI skills, and take action across 6,000+ apps without leaving Slack.

Teams at companies like Anthropic, Perplexity, and Rocket Companies are reporting up to 90 minutes a day saved — over two months of working hours back every year.

See how Slackbot turns intent into action inside the flow of work you already know.

Explore the new Slackbot →

But one of the conversations I had that was seemingly so simple, had the most profound effect on me.

Rachel is a Director working in Data Strategy & Insight in New York City, helping Fortune 100 companies best utilize AI. Her team has earned her clients upwards of $50M in revenue in just the past few months.

I say this to say, she knows her sh*t when it comes to AI.

The question she brought up at our member coffee meetup on Friday morning (that has since lived rent free in my head) was:

Build or Buy?

5 years ago, it would be an absolute no brainer to buy the product and move on with your life. Now, it’s not that simple.

With the rise of vibe coding and Claude Code, anyone can build anything.

• Quick timer app. Done.

• Period tracker that doesn’t steal your data and wasn’t founded by a man. You got it.

• A fully customized financial dashboard with real time updates through API connections to all your subscriptions. Yep. No Tableau subscription needed.

Yet, there are still a lot of things to consider. Security and data privacy being the most important IMO.

Instead of offloading the responsibility to a $25/month company, that burden now becomes yours. And without a team of specialists behind you, who knows if you are doing it correctly?

Just this past week both Vercel and Lovable had major security outages and breaches.

Every Monday, AI Snack Club participates in #MondayMotivations in our private Slack channel. I posed this question and the answers were thoughtful and strategic, especially compared to the responses on my LinkedIn poll.

AI Snack Club members are consistently five steps ahead of the conversation and this thread proved it. Nobody reflexively said “build everything yourself,” despite this being an AI-forward community.

Instead, they laid out a genuinely sophisticated framework: build when it’s just for you, buy when other people’s data touches it, when deliverability matters, or when the maintenance burden outweighs the upside.

In stark contrast, my LinkedIn poll could not be more different.

On LinkedIn, the instinct is to puff out your chest and pick the answer that sounds most AI-native. Build. Obviously. Of course. Why would you pay for something you could technically make yourself?

It’s not clear yet which answer is “correct,” but what is clear is that the criteria we’re using to choose between build and buy is evolving faster than any of us can keep up with. Ask me again in a year and I have no idea what the split will look like, on LinkedIn or in Slack. And I genuinely can’t wait to find out.

What I do know is that we need more conversations like this one that happened in Snack Club this Monday. The kind where people let their guard down, think through the tradeoffs holistically, and make well-informed decisions instead of reflexive ones. These conversations are happening, but almost exclusively in safe spaces like Snack Club and other women-in-AI communities where the pressure to perform isn’t running the show.

Join the Community

Thank you to everyone who jumped into the thread or voted on the poll. I’m genuinely intrigued to keep pulling on this thread and see how the conversation evolves.

If you have a build-or-buy story of your own (something you built that surprised you, or something you almost built and decided to buy instead) I want to hear it. Reply, DM me, or bring it to the next coffee meetup.

xx,
Monica

• Claude Opus 4.7 + Claude Design — Anthropic’s new tool lets you build slides, prototypes, and one-pagers with Claude. Your Canva-to-deck workflow just got faster. See what’s new

• Starbucks × ChatGPT is live, and the internet is not having it. You can now ask ChatGPT for a drink rec based on your mood or your outfit. One critic called it treating customers "like empty, soulless venti cups." The AI-backlash discourse is in full swing. Fortune

• Bridal hair and makeup artists are fighting AI inspo pics. Half of brides are showing up with Pinterest boards full of AI-generated looks that literally cannot be recreated on real faces. Your wedding industry friends will have thoughts. Axios

• Amazon’s putting another $25B into Anthropic — on top of $8B already in. Your Claude subscription isn’t going anywhere. CNBC

  

If you are interested in joining AI Snack Club and want to try out an event, we invite you to sign up for one community expert session to give us a try!

Quick heads up: Guests are welcome to attend one AI Snack Club event as a non-member. If you’d like to keep joining us, please grab a membership rather than signing up for future events as a guest.

Upcoming Events - Virtual

• How to Job Search with AI - 4/23 at 11 am pt (rsvp)

• How Amy Briggs Built a Tech Company from Scratch - 4/27 at 9 am pt (rsvp)

• How to Use Claude Cowork to Build a Personal Finance Dashboard - 4/30 at 10 am pt (rsvp)

• How to Build AI Systems for Customer Success Experts - 5/1 at 9am pt (rsvp)

• Getting Started with Github and Github Portfolio - 5/5 at 11am pt (rsvp)

• AI as Your Personal Development Lab - 5/7 at 12 pm pt (rsvp)

• How to Orchestrate Your Workweek with AI by 9 AM Monday Morning - 5/8 at 9am pt (rsvp)

• How to Make Notion Your Business Brain - 5/14 at 10am pt (rsvp)

• Building Marketing Systems with Claude Cowork - 5/20 at 10am pt (rsvp)

• How to Build a RAG Agent in n8n - 5/29 at 12pm pt (rsvp)

• How to Prompt AI Agents - 6/3 at 12pm pt (rsvp)

Upcoming Events - IRL

• Build Your First App with Lovable & AI Snack Club - 4/29 at 5:30pm in Chicago (River North) (rsvp)

• Marketing at the Intersection of AI & Fashion with Daydream - 5/13 at 9 am in NYC (rsvp)

• Learn to Vibe Code with Lovable & AI Snack Club - 5/15 at 12 pm in NYC (rsvp)

• Female Founder Vibe Coding & Co-Working Day with Lovable and AI Snack Club - 5/26 at 9 am in SF (rsvp)

  

Quick thing before I get into this: I’m really intentional about the brands I bring into this community. AI Snack Club doesn’t partner with companies just because they reach out. I only work with tools I genuinely use, genuinely believe in, and genuinely think will make your life easier. Slack is one of those, which is why I partnered with them on this newsletter.

Without Slack, our now 700+ (!!) group of women wouldn’t exist the way it does. The celebrations when someone lands a new job. The live threads during events. The “has anyone tried this?” questions. That all happens in Slack. It’s the backbone of this community. So when Salesforce invited me to their Slackbot event in SF earlier this week, I said yes immediately. Not because of the partnership, but because I wanted to know what’s coming for the tool we literally depend on every day.

OK so here’s what happened.

Marc Benioff was there. Parker Harris, Salesforce's co-founder and Slack's CTO, was there. And they announced 30 new features that are turning Slackbot from that little bot you might have previously ignored into a full AI agent that can actually do things for you.

"AI Agent" is one of those phrases that gets thrown around so much it's lost all meaning. But let me tell you the three things I saw that made me pull out my phone and start taking notes:

🧃 It routes work to other tools for you.

• This one’s a little more technical, but here’s the simple version: Slackbot can now connect to other AI agents and apps through MCP. You tell Slackbot what you need, and it figures out which tool should handle it. No more switching between five apps, Slackbot does the routing all within your Slack conversation. You can start using other apps directly in the chat which is extremely convenient and better for my mental health than having 500 tabs open.

🧃 Reusable AI Skills.

• You can now teach Slackbot a task once, like “create a budget for an upcoming event”, and it learns the steps. It pulls info from your channels, your connected apps (calendar, email, etc.), your docs, and repeats that workflow whenever you need it.

• I immediately started thinking about how I could use this for our event planning,. Instead of re-doing the same steps every time I set up a Luma, I could just tell Slackbot “do the thing.”

🧃 Meeting Summaries. Now Slackbot has entered the note-taking race and can transcribe your meetings and give you a recap with your specific action items. Very curious to try this feature and see how it compares to the others.

What I’m actually going to do with this:

I’m currently working on adding Slackbot to the AI Snack Club Slack so that all of our members can try it, test it, and learn together (especially those who don’t have Slack access at work!).

If you want to see the full picture of what the event looked like and what I took away, I recorded a short video walkthrough:

📹 Watch it here

This article is in partnership with Slack.

Anthropic accidentally leaked 500,000+ lines of Claude Code’s source code. A debugging file got bundled into a routine update and the internet had it within hours. Buried in the code: a Tamagotchi-style coding buddy, an overnight memory system, and an “Undercover Mode.” A safety-first AI company having a packaging accident? The irony writes itself. (link)

Claude hit #1 in the App Store. After Anthropic refused to let the Pentagon use its AI for lethal operations (and OpenAI said yes), users migrated in droves. Paid subs are at record highs. The demand is so high it’s causing outages. (link)

OpenAI closed a $122B round at an $852B valuation. Meanwhile, investors are quietly trying to offload OpenAI shares and pivot to Anthropic. Power dynamics are shifting fast. (link)

Anthropic’s design lead showed her full Cowork workflow on camera. 40 minutes of turning messy research into prioritized product ideas and wireframes. Her philosophy: “garbage in, treasure out.” Worth a watch. (link)

If you are interested in joining AI Snack Club and want to try out an event, we invite you to sign up for one community expert session to give us a try!

Quick heads up: Guests are welcome to attend one AI Snack Club event as a non-member. If you’d like to keep joining us, please grab a membership rather than signing up for future events as a guest.

Upcoming Events - Virtual

• AI as Your Second Brain - 4/8 at 10 am pt (rsvp)

• How to Build your Personal Operating System with Claude Code - 4/14 at 10 am pt (rsvp)

• Generate Beautiful Images & Video Using AI - 4/16 at 11 am pt (rsvp)

• Break Before You Build: How to Stress-Test AI Systems - 4/22 at 10 am pt (rsvp)

• How to Job Search with AI - 4/23 at 11 am pt (rsvp)

• How Amy Briggs Built a Tech Company from Scratch - 4/27 at 9 am pt (rsvp)

• How to Use Claude Cowork to Build a Personal Finance Dashboard - 4/30 at 10 am pt (rsvp)

• How to Build AI Systems for Customer Success Experts - 5/1 at 9am pt (rsvp)

• AI as Your Personal & Professional Development Lab - 5/7 at 12 pm pt (rsvp)

• How to Orchestrate Your Workweek with AI by 9 AM Monday Morning - 5/8 at 9am pt (rsvp)

• How to Make Notion Your Business Brain - 5/14 at 10 am pt (rsvp)

Upcoming Events - IRL

• Finding the Perfect Prompt | Vibe Code Night with Anything.com - 4/8 at 5:30pm in San Francisco, CA (at capacity for non-members)

• The Human Side of AI: Building Real Connection with Goodword - 4/15 at 9 am in NYC (rsvp)

• Female Founder Coworking Day with Superhuman Mail and Lovable - 4/16 at 9 am in NYC (Williamsburg) (rsvp)

• Build Your First App with Lovable & AI Snack Club - 4/29 at 5:30pm in Chicago (River North) (rsvp)

• Marketing at the Intersection of AI & Fashion with Daydream - 5/13 at 9 am in NYC (rsvp)

• Learn to Vibe Code with Lovable & AI Snack Club - 5/15 at 12 pm in NYC (rsvp)

• Female Founder Vibe Coding & Co-Working Day with Lovable and AI Snack Club - 5/26 at 9 am in SF (rsvp)

🧃 “Start small, ignore the hype. Pick something annoying and low-stakes for your first automation“ - Diana Olympia, on creating your first automation

“Where do I start with AI?”

The question that buzzes around in my head almost daily. It’s the one that gets asked on every Community Expert Workshop, the one that I get DM’d about weekly, and the one that haunts me in my sleep.

So I am tackling it head on. I’m presenting an AI 101 session to a group of early-career women next week, and I thought I would share that knowledge with you all today!

Additionally, I am super excited to announce the launch of my new IG Reels series titled “Where Do I Start with AI?” … I know, super clever! Stay tuned for that coming out in the next few weeks!! You may see some familiar faces.

Subscribe now

🧃 Here’s what I’m covering in AI 101

First: AI is just a tool. It’s software trained on massive amounts of text, images, and data so it can generate responses, make predictions, and complete tasks. Think of it like a very fast, very well-read collaborator who needs good direction from you. The better you communicate with it, the better it performs. It’s not magic, it’s only as useful as the person using it.

Second: Common terms to know. There’s a handful of terms that come up constantly in AI conversations, and once you know them, everything clicks faster. LLM (Large Language Model) — that’s the engine behind tools like ChatGPT and Claude. Prompt — the instructions you give AI; better prompts = better outputs (this is your superpower). Hallucination — when AI confidently makes something up (always fact-check!). Vibe coding — building apps with plain English, no traditional code required. And Agent — an AI that doesn’t just answer, it takes action. Books meetings, sends emails, browses the web. That’s the frontier we’re headed toward.

Third: your starter stack. You don’t need a million tools. Just start small. Experiment and try them all. And most importantly - pick which one you like the most!!

🪄 NotebookLM — This is Google’s AI research assistant, and it’s quietly one of the most useful tools out there. You upload your own documents (PDFs, Google Docs, YouTube videos, websites, etc) and then you have a conversation with them. I use it to summarize long reports instantly, prep for meetings by uploading the deck beforehand, turn messy piles of notes into structured summaries, and even create audio “podcast” overviews of dense material. Try this today: upload a document you’ve been meaning to read. Ask: “What are the 5 most important things I need to know?” Then ask a follow-up question. Notice how it cites its sources, that’s how you know it’s not making things up.

🪄 Claude — Made by Anthropic, Claude is the tool I reach for when I need to actually think something through. It’s known for being thoughtful, nuanced, and genuinely good at writing and reasoning. I use it to draft emails, proposals, and presentations, to think through hard decisions (”what am I missing here?”), to edit and tighten my own writing, and to brainstorm when I’m stuck and staring at a blank page. Try this today: paste something you’ve written and say “Make this clearer and more direct. Keep my voice.” See what comes back. You’ll be surprised how well it preserves your tone while sharpening your message.

🪄 Lovable — This is a vibe coding tool, which means it turns plain English descriptions into actual, working web apps. You describe what you want to build. It builds it. No code required. I’ve used it to build simple tools and calculators, create landing pages and waitlist forms, prototype ideas before hiring a developer, and bring personal projects to life that had been sitting in my “someday” list forever. Try this today: type “Build me a simple tracker where I can log my daily habits and mark them complete.” Then watch what happens. The first time you see it generate a working app from a sentence, something shifts in your brain about what’s possible.

🪄 Relay App — Think of this as a simpler, more flexible version of Zapier. It’s a workflow automation tool where you build sequences of steps that run automatically when triggered, so your tools finally talk to each other. I think about it for automating onboarding flows and follow-up emails, building processes that run without a human in the loop, and connecting tools so nothing falls through the cracks. Try this today: map one thing you do manually every week — a recurring email, a data entry task, a follow-up you always forget. Ask yourself: “Could this run on its own?” That’s your first automation candidate.

What’s Hot on the Internet

• This AI-generated Loewe x Cybex diaper bag has me in a chokehold — A creator used AI to imagine a woven leather luxury baby kit and… I need it to be real. This is what I mean when I say AI isn’t just a productivity tool, it’s a creative playground. People are using image generators to dream up “what if?” product collabs, fantasy interiors, imaginary fashion lines, and the results are stunning. If you’ve ever thought “I wish [brand] made [thing]” — you can now visualize it yourself. Manifesting via Midjourney, if you will. (see it here)

• Vibe coding made MIT Tech Review’s Breakthrough Technologies list — AI writes 30% of Microsoft’s code and a quarter of Google’s. Tools like Lovable, Cursor, v0, and Replit are making it possible for non-coders to build real apps in plain English. Validation for everything we talk about here. (source)

• Nvidia GTC is happening this week — Jensen Huang takes the stage tomorrow in San Jose with 39,000 attendees from 190 countries. Expect major announcements on agentic AI, new chips, and an open-source agent platform called NemoClaw. The keynote streams free. (watch here)

Call all soon to be moms - if only this was real!! (link)

🧃 “Remove personal details before using free tools - use company-approved tools with DPAs for business use“ - Annabel Pemberton, Assenteo CEO, on safe AI tool practice

Happy International Women’s Day!! I had the best time this past Thursday with over 70 ambitious women learning to build with AI (many for the first time!!). I left the night feeling inspired and determined to keep building these types of spaces. If anyone has any office space they are willing to lend out, please let me know!! And the film photos are currently developing, but can’t wait to share more on that soon.

Today is the last day for the Founding Membership. Sign up now to lock-in your 50% off discount for life!

Sign up here: https://aisnackclub.com/upgrade

Subscribe now

So here’s something I’ve been watching closely: Apple is finally about to give Siri a real AI upgrade.

For the last two years, Apple has been promising us a smarter Siri: one that can actually read your emails, understand what’s on your screen, and take actions across your apps without you opening six different things. They first announced it at WWDC in 2024. Then it got delayed. And delayed again. And then they got sued for false advertising (!!) and settled a class-action lawsuit in December.

Plot twist: Apple partnered with Google to make it happen. They’re paying roughly $1.5 billion a year to use Google’s Gemini AI models to power the new Siri, running through Apple’s Private Cloud Compute to keep your data private.

The update is expected to roll out with iOS 26.4 in the next few weeks, though Bloomberg’s Mark Gurman reported some features might slip to May or even September. (Shocker.)

🧃 Here’s what I’m watching and why it matters:

The new Siri is supposed to do three things: Personal Context (pulling info from your notes, emails, and messages), On-Screen Awareness (completing tasks based on what’s on your screen), and In-App Actions (controlling apps by voice without opening them).

Think: asking Siri “What time is Mom’s flight?” and it actually checks your messages to find the answer. Or saying “Make this photo black and white and send it to Austin” and it does both steps without you lifting a finger.

This is exactly the kind of AI that makes life easier without you needing to learn a new tool. It’s AI coming to you inside apps you already use. And if Apple gets it right, it could be the most seamless way non-technical people interact with AI on a daily basis.

If anyone has played around with Claude Code, you already know what this feels like. It can directly access files and apps on your laptop and take actions for you. Apple is essentially trying to bring that same kind of capability to your phone.

I’ll be testing the beta the second it drops and sharing what actually works vs. what’s still half-baked. Stay tuned.

What’s Hot on the Internet

• Lovable is completely FREE today for International Women’s Day — Full platform access, no registration required, plus $100 in Claude API credits. If you’ve been sitting on an idea, go build it. (link)

• OpenAI dropped two models in one week — GPT-5.4 combines reasoning, coding, and computer-use into one model. The other, GPT-5.3 Instant, finally stops lecturing you before answering your question. (source)

• States are racing to regulate AI — Oregon passed a chatbot safety bill this week, Vermont signed a law on AI in elections, and multiple states are pushing legislation on deepfakes and AI in health insurance. (source)

🧃 “I’ve spent 10+ years bringing products to market. Yesterday, I built one myself” - Diana Golduber, Product Marketing Manager (post link)

I failed as a CS major in college. And I never graduated from the coding bootcamp I eagerly signed up for two years ago. I always wanted to learn to code, but it never clicked. It was hard. It was daunting. And even in the supposedly beginner friendly classes, I already felt behind and that everyone already knew what was happening. And it sucked to be in that position.

But today through the power of AI, I can finally build all the things my heart desires, without a fancy CS major or coding bootcamp certification! And you can too. Today’s article is all about what is vibe coding, the tools you should start with, and examples of what to build.

Subscribe now

What the hell is vibe coding?

Vibe coding, in its most simple and literal sense, is when you describe what you want in natural language and AI builds it for you, writing the code in the backend on your behalf.

The term originated from this tweet by Andrej Karpathy, founding team at OpenAI:

Why should you care?

Because everything created up until this point has been developed by humans with hands on keyboard. Every app you’ve ever used, every website you’ve ever visited — all of it was written by developers, line by line.

But, that’s not the case anymore. AI can now write entire applications from a plain English description, and it’s outpacing developers at a rate that would’ve sounded ridiculous even two years ago.

The tools that exist right now make it possible for anyone to build, even someone with zero coding experience, to create real-life, functioning apps. And people aren’t building generic apps, they are solving their own hyper-specific problems with custom solutions.

We are entering the era of hyper-personalization.

Which tools should I start with?

Lovable — This is my #1 rec if you’ve never built anything before. You describe your app idea in plain English, and Lovable generates both the design and the code. The free plan gives you 5 credits a day to experiment, and Pro is $25/month. I used this app to make the AI Snack Club Founding Member post generator for LinkedIn (link).

v0 (by Vercel) — If you came to our Galentine’s event (more on that below!), you already got a taste of v0. It’s incredible at creating polished, beautiful web components from a simple text description. The free tier gives you $5 in credits per month, and Pro is $20/month. I’d say v0 is perfect for someone who wants their output to look really good right away, like, scarily professional.

Claude — If you already use Claude (hi, it’s my favorite AI tool and you know this by now), you can build quick MVPs right inside your conversation. Just ask Claude to create an interactive app, dashboard, or tool and it generates a working prototype you can use and share immediately. It’s great for testing an idea before committing to a full build on Lovable or v0. You can even ask Claude to write you a prompt for a vibe coding tool.

And what should I build?

Here are some ideas to get the wheels turning:

For your career:

• A personal portfolio site that actually looks professional

• An internal dashboard to track your team’s metrics

• A custom project management tool that works the way your brain works

• An AI-powered email template generator for your specific role

For your side hustle or business idea:

• A landing page to test if people are interested in your thing

• A simple booking or scheduling app

• A product recommendation quiz

• A directory or marketplace MVP

For your life:

• A meal planning app customized to YOUR dietary preferences

• A habit tracker that doesn’t annoy you

• A gift idea generator based on the recipient’s interests

• A wedding planning tool

🦋 Here’s what I want you to do this week:

1. Pick an idea. From the list above or something entirely your own. Or use AI to help you think of something.

2. Open one of the tools listed above

3. Describe what you want in a few sentences. Be specific! Focus heavily on the outcome and components. Or if you need help writing your prompt, use Prompt Cowboy to build it for you.

4. See what happens and share your experience with me! Or in the community Slack.

Last Tuesday was AI Snack Club’s first-ever in-person event and I’m still buzzing about it!! We partnered with v0 by Vercel for a Galentine’s Vibe Coding Night at their San Francisco office, and it was everything I hoped it would be and more.

HIGHLIGHTS OF THE NIGHT ✨

• raffling off the most insane bouquet of flowers from Megha at Wave Floral

• seeing live demos of an app that scans Hinge profiles to verify if they are actually over 6ft 🤭 (link coming soon)

• indulging in pizza, poppi, rose, and candy!!!

• and of course listening to taylor swift, duh

NEXT SESSION COMING SOON!!! 👀

• AI took over the Super Bowl ads — 15 out of 66 commercials featured AI. Google used Gemini to redesign a family’s home, Chris Hemsworth accused Alexa+ of trying to kill him, and Anthropic mocked other chatbots for adding ads. Three AI commercials in 45 minutes is A LOT. (link)

• AI can read brain MRIs in seconds — University of Michigan researchers built an AI that interprets brain scans almost instantly, flagging everything from aging patterns to cancer. Not replacing your doctor, but getting closer. (link)

• Lovable’s 28 Days of Lovable challenge — Since we’re talking vibe coding today, this felt too relevant not to mention again. Free daily lessons all February on prompt engineering, design thinking, and actually shipping an app. Jump in at february.lovable.app and join our slack channel to chat with our members on their monthly journey.

🧃 “I had a SURREAL week. 3 launches + one full day shoot talking about future of AI & creators. very energized by the work I am doing and I get such a dopamine hit every time I do cool shit “as a solopreneur” when everyone told me my career would become “mat leave coverage / writing ebooks” - Fernanda Chouza, Fractional Product Marketing Leader

I get this question more than any other: “Where do I actually START with AI?”

And it’s never just that one question. It’s Which tool do I use? How can I be an expert if I am limited to my company’s internal LLM and copilot? What the f*ck is vibe coding? And my favorite question, Am I even doing this right?

While there is not correct way to start, here are my biggest pieces of advice. Let’s get into it!

Subscribe now

🦋 How to Get Started with AI:

Experiment

I know that sounds almost too simple, but hear me out. The biggest mistake I see people make is trying to “learn AI” like it’s a course with a syllabus and a final exam. It’s not. It’s more like learning to cook, you don’t read every cookbook first. Instead, you pick a recipe, make a mess, eat something halfway decent, and try again tomorrow.

Pick ONE thing you’re doing this week, like writing an email, planning a trip, prepping for a meeting, organizing your closet, meal prepping, and ask AI to help. Just open Claude or ChatGPT and say, “Hey, I need help with ___. This is my first time using AI, so please walk me through step by step”

Ask AI where AI can help you

This piece of advice is always a little meta, but a great starting point is asking AI where it can help you. Instead of you figuring out where AI fits into your life, ask AI to figure it out for you.

Try this prompt:

🦋 “I’m going to describe my typical week and daily responsibilities. Ask me 10 specific questions about how I spend my time, what tasks I do repeatedly, and what frustrates me. Then suggest areas where AI could save me time or improve my output.”

Sometimes you just need a thought partner to help you uncover assumptions or pieces of context that wasn’t obvious. This is a great place to start.

Join the conversation (shameless plug #1)

One of the things I’m most proud of inside our AI Snack Club Slack is the sheer number of threads happening right now about exactly this topic. Members are sharing what they’ve tried, what worked, what flopped, and asking each other “has anyone used AI for ___?”

There’s something different about figuring this out alongside other women who are in the same boat versus going down a solo Reddit rabbit hole. You get real answers from real people with real jobs, not some hypothetical “10x your productivity” nonsense.

Come build with us (shameless plug #2)

If you’re someone who learns by doing, we have two things coming up that are basically designed for the “where do I start?” crowd:

Galentine’s Vibe Coding Night with v0. We partnered with v0 (Vercel’s AI-powered building tool) for an in-person event in San Francisco. You’ll build something you never though possible, learn about v0’s newest features, and network with women who are just as curious as you are. No coding experience required! RSVP.

RSVP

Let’s Build Together Sessions. This is brand new and I am so excited about it. Twice a month, we feature a different tool (think Zapier, NotebookLM, and more), vote on a project to build as a group, and spend 60 minutes actually building. Participation is limited to paying community members only. Sign up now!

Both of these exist because I kept hearing the same thing: “I want to start but I never make time for it alone.” You don’t have to do it alone.

What’s Hot on the Internet

• It’s a “Perfect February”, February 2026 starts on a Sunday, ends on a Saturday for a flawless 4×7 calendar grid that hasn’t happened since 2015. If a neat calendar is what it takes to stick to your resolutions, I’m here for it. (link)

• Your AI just became your doctor (kind of), OpenAI and Anthropic both launched health features that connect your medical records and Apple Health so AI can explain test results and prep you for appointments. Not replacing your doctor, but replacing the 2am WebMD spiral. (link)

• Big Tech is dropping $470 BILLION on AI this year, Microsoft, Meta, Alphabet, and Amazon are expected to spend $470B+ on AI infrastructure in 2026, up from $350B last year. That’s more than Norway’s GDP. The question everyone’s asking this earnings season: when does it pay off? (about:blank)

• Research says AI has a ceiling, New math research claims LLMs are fundamentally limited in how complex their reasoning can get. A reality check for anyone told AI is about to replace them. Spoiler: it’s not. But it IS a really good assistant. (link)

🧃 “When a prompt consistently gives me great results, it goes into the library with a date stamp. Basically: collect → test → refine → save what works.”- Tex, Certified Coach and AI-Fluent Entrepreneur

Hi friends! Earlier this week I asked Claude to “help me draft a strategy plan for Q1,” hit enter, and stared at the result wondering why it felt so generic.

The problem wasn’t Claude. The problem was me.

Without realizing it, I’d handed over about 47 micro-decisions to an AI. What perspective should this come from? What format works best? What trade-offs am I willing to make? All of those questions got answered by default settings I never consciously chose.

So naturally, I started experimenting. And what I found changed how I prompt completely. Let’s get into it!

Subscribe now

🦋 The Invisible Decisions You’re Delegating

When you type “build me a landing page” into a tool like Lovable or “reply to this email” into ChatGPT, you’re not just asking for help. You’re giving AI permission to make a bunch of assumptions about what “good” looks like.

And AI will happily oblige. It’ll pick a layout. It’ll choose a tone. It’ll decide what information matters most. But it’s making those calls based on patterns from millions of generic examples, not based on what actually works for your specific situation.

Take that landing page example. When you say “build me a landing page for my course,” the AI is silently deciding:

Should the CTA be above the fold or after social proof? (This changes conversion rates by up to 30% depending on your audience’s awareness level.)

Should we lead with the problem or the transformation? (B2B audiences often respond better to problem-first; B2C often wants the dream state first.)

How much copy is too much copy? (High-ticket offers typically need longer copy. Impulse purchases want shorter.)

The AI isn’t thinking through these trade-offs based on your business model, your price point, or how aware your audience is. It’s just picking something that looks reasonable to a robot.

🦋 What I Started Doing Instead

The difference between okay AI output and genuinely useful output is usually just one step: pausing to identify what decisions you’re about to hand over, then making those decisions yourself before you prompt.

Here’s what this looks like in practice:

For building things (apps, websites, presentations):

🐛 Before: “Build me a dashboard for tracking my clients”

🦋 After: “Research the top dashboard layouts from a UX perspective. Show me the 3 layouts that perform best for retention and quick decision-making, and explain the trade-offs of each. Then help me choose which one fits my use case: I’m a solopreneur who checks this dashboard 2-3 times per day on mobile and needs to spot red flags immediately.”

Now I’m making AI do the research work and show me options, rather than letting it silently pick option #1 from its training data.

For strategy and planning:

🐛 Before: “Create a go-to-market strategy for my new feature”

🦋 After: “Develop three versions of this go-to-market strategy from different executive perspectives: one from a VP of Marketing (focused on positioning and channels), one from a VP of Finance (focused on unit economics and payback period), and one from a CEO (focused on strategic timing and competitive positioning). Highlight what each perspective prioritizes and where they conflict.”

Now I’m getting multiple strategic lenses instead of one generic plan, and I can see where the trade-offs actually are. (Side note: Watching fictional executives argue with each other is weirdly entertaining.)

For everyday stuff:

🐛 Before: “Give me dinner ideas”

🦋 After: “Give me ideas for cooking dinner tonight from two perspectives: one from a Michelin star chef focused on flavor and technique, and one from a scrappy college student focused on cheap ingredients and minimal effort. Show me how I can make the most delicious meal possible with what I probably already have in my pantry.”

Now I’ve defined the constraints that actually matter to me (taste + budget + convenience) rather than getting random recipes I’ll never make.

🦋 The 30-Second Check

Next time you’re about to prompt AI for something that matters, try this:

1. What decisions am I about to let AI make for me? (Layout, tone, approach, format, priorities...)

2. Do I actually want to delegate those decisions, or should I be more specific?

3. If I were explaining this to a human, what context would I give them that I’m leaving out here?

Then rewrite your prompt with that context included. You don’t need to be a prompt engineer. You just need to be clear about what you want before you ask for it.

The irony is that this actually makes AI MORE useful, not less. When you’re intentional about the parameters, AI can focus its intelligence on the actual hard part (synthesis, ideation, drafting) instead of guessing what you meant.

What’s Hot on the Internet

• Fashion‑tech girls are obsessed with Verifyt right now: an AI‑powered shopping app that uses 3D body scanning plus creator try‑ons so you can shop clothes on people who actually share your body type. (link)

• 2026 is the New 2016” TikTok and Instagram are drowning in nostalgia. Everyone from your college roommate is posting throwbacks with puppy-dog Snapchat filters and oversaturated photos. Longing for the “last year the internet felt fun” before algorithms and AI took over. (Cue the Chainsmokers playlist.) (link)

• OpenAI is starting to test personalized ads inside ChatGPT for free and Go users in the US, with sponsored links tucked under responses and targeted to whatever you’re chatting about (think trip-planning chats that quietly surface hotel deals). (link)

🧃 The best AI tool isn’t the fanciest one — it’s the one you actually know how to use and the one that gets the job done efficiently.

I’ll be honest, mine started off a little frustrating. I was deep in a writing flow on Wednesday, using Claude to help me think through some newsletter automations, and then... I hit my dreaded usage limit. AKA come back in a few hours or upgrade to keep going.

Here’s the thing: I pay $20/month for Claude Pro. I also pay $20/month for ChatGPT Plus. And I still run out of credits multiple times a week. When I attended a virtual AI workshop this week, I was thrilled to know that I was not the only one who ran into these problems.

So I took a deep dive into the credit and usage system, so you don’t have to, and I am excited to share my findings. Let’s get into it!

🦋 What Are Credits? And Why Do They Matter?

Here’s what no one tells you when you sign up for a paid AI subscription: “unlimited” is marketing speak. Every AI tool runs on a credit system behind the scenes, even if they don’t call it that. And understanding this has genuinely been a game-changer for me.

For Claude (Pro at $20/month): You get around 45 messages every 5 hours for regular conversations. But, there’s also a weekly cap that resets every 7 days. So even if you pace yourself during the day, you can still hit a wall by Friday if you’ve been using it heavily all week. The newer, fancier model (Opus) burns through your quota WAY faster than the standard one (Sonnet). Anthropic recommends using Sonnet for everyday tasks because Opus drinks credits so fast.

For ChatGPT (Plus at $20/month): ChatGPT Plus now runs on the GPT-5 model family. You get around 150 messages every 3 hours for the standard model, though this varies based on system load. The Thinking mode (for complex reasoning) has separate, more restrictive limits. Deep Research is still limited to around 10 queries per month. Image generation caps are dynamic but generally around 50 images per 3 hours. When you hit your limit, you get downgraded to the mini model until your window resets.

The kicker is that these aren’t just message counts. They’re based on something called “tokens”, basically, every word you write and every word the AI writes back costs tokens. Longer conversations = exponentially more expensive.

Pro Tip: Your 20th message in a chat costs way more than your first message because the AI is re-reading the entire conversation history each time. So, it’s actually better to start new chats more frequently.

🦋 Why Claude’s Opus 4.5 Is Different

When Anthropic released Opus 4.5 in November, it was genuinely impressive. This model scored 80.9% on a major coding benchmark (the first time any model broke 80%.) It’s incredible for complex reasoning, debugging, and analysis.

But, one Opus conversation can consume as much quota as 10+ Sonnet conversations. Not so great.

And it gets worse. Opus uses something called “Extended Thinking”, where it essentially reasons through problems before giving you an answer. You might see 200 words in the response, but you’re actually paying for 1,200+ words of invisible thinking tokens.

🦋 7 Tips to Stop Burning Through Your Credits

After my deep dive, here’s what actually works (specifically for Claude users):

1. Use Claude Sonnet for 80% of tasks. Most of what we do doesn’t need Opus. Writing emails? Sonnet. Brainstorming? Sonnet. Quick questions? Definitely Sonnet. Save Opus for the complex stuff that requires deep reasoning.

Toggle between modes

2. Start new conversations more frequently. Every time you send a message, the AI re-reads your ENTIRE conversation history. For simple questions, start fresh instead of adding to a long thread.

3. Don’t upload files unless you absolutely need them. When you upload a document, Claude re-reads all files with every message. One PDF + 10 messages = reading that PDF 10 times. Consider copy-pasting relevant sections instead.

4. Be specific upfront. Vague prompts lead to back-and-forth iterations. Each “try again” or “make it shorter” costs as much as your original message. Use my favorite tool: Prompt Cowboy, before asking Claude.

5. Batch your complex work. Instead of five separate Opus conversations throughout the week, batch your deep work into one or two focused sessions. This helps you stay within weekly limits.

6. Time your heavy usage around resets. Claude resets every 5 hours (session) and every 7 days (weekly). ChatGPT’s 3-hour rolling window means your oldest messages “expire” and free up space. Here is an in-depth article on the hourly sessions.

7. Consider the Max plans for heavy use. If you’re consistently hitting limits, Claude Max ($100/month) gives you 5x the Pro allocation, but $100 a month is a hefty price to pay if you are not using it to its fullest.

What’s Hot on the Internet

• The viral LOEWE x Stanley collab is fake and AI, sorry! The stunning visuals were created by Studio Miso, a company that specializes in dreaming up hypothetical high fashion x everyday brand collaborations using AI. They’ve built a following around these beautiful-but-unreal mashups, and I seriously wish these were true. (link)

• CES 2026’s “Worst in Show” went to AI refrigerators. Consumer advocacy groups called out voice-controlled, AI-powered refrigerator doors as peak unnecessary tech. (link)

• Stanford researchers got AI to spit out 95% of Harry Potter. By prompting models to “continue” short passages from copyrighted books, researchers extracted nearly verbatim reproductions from Claude, GPT-4, and others. The copyright implications are significant. (link)

• Meta is buying Manus, the viral AI agent startup, for $2-3 billion. Manus went viral for letting AI autonomously execute tasks with minimal prompting. Get ready for “agentic AI” to show up in your WhatsApp and Instagram DMs. (link)

🧃 “You don’t have to worry about the logic of how do I go from step A to step B to step C. The best way to prompt is just to be super outcome-based — if you could tell your personal assistant or intern exactly what you wanted, that is how you should prompt.” - Selina Li, Co-Founder of Bubble Lab (YC W26)

Subscribe now

Every week, someone asks me "What AI tool I should try that isn't ChatGPT?" And every week my answer is the same: NotebookLM. It's free, it's made by Google, and it genuinely changed how I consume information. Today I'm breaking down why I think it's the most underrated AI tool out there and giving you 5 reasons to try it this week.

OK so what is NotebookLM?

NotebookLM is a free tool from Google that lets you upload your own stuff, like PDFs, Google Docs, YouTube videos, website links, Google Slides, even Google Sheets, and then talk to an AI that actually knows YOUR material. Not the whole internet or what it thinks is relevant.

So instead of asking an AI a general question and hoping it gets it right, you’re asking an AI that has actually read your specific stuff. It’s like the difference between Googling “how to plan a wedding” and having a friend who read your entire Pinterest board, budget spreadsheet, and venue contract sit down and help you figure out next steps.

You go to notebooklm.google.com, sign in with your Google account, upload your sources, and start asking questions. If you have a Gmail, you’re already in.

Why should you care?

Because most of us are drowning in content we mean to read and never do. The bookmarked articles. The saved YouTube videos. The 30-page PDF your boss sent that you skimmed for 45 seconds and prayed nobody would quiz you on. NotebookLM actually makes it possible to do something with all of that.

Here are 5 reasons I think it deserves way more hype:

1. It turns your research into a podcast you can listen to while walking your dog

After you upload your sources, NotebookLM can generate a full podcast-style conversation between two AI hosts who break down your material. They banter back and forth and make connections between your documents, and explain things in a fun, playful way.

I use this constantly now. While doing dishes. While driving. While folding laundry. It turns dead time into learning time and I’m a little obsessed.

2. It only knows what you give it

This is what makes it different from ChatGPT or Claude or Gemini. When you ask NotebookLM a question, it pulls exclusively from the sources you uploaded.

So when I uploaded my 300+ founding member survey responses and asked “what are the top 3 things my members are struggling with?” — the answer was grounded in real data from real people. I didn’t have to worry about hallucinations.

3. It generates mind maps, study guides, and briefing docs from your sources

The audio podcast thing gets all the attention, but the other output options are just as good. You can take your uploaded sources and with one click generate:

• A mind map that visually maps out how all the concepts in your sources connect to each other (you can click on any node and ask questions about that specific topic)

• A briefing doc that summarizes everything into a tight, scannable overview

• A study guide that pulls out the most important points

• FAQs based on your material

• A timeline if your sources are about events or historical stuff

I used the briefing doc feature before a call with a potential sponsor last week. I uploaded their website, their founder’s LinkedIn posts, and two podcast interviews they’d done. NotebookLM gave me a one-page summary of their brand positioning, key talking points, and what they care about.

4. It can turn your research into slide decks and infographics

This one is newer and probably one of the most useful features.. You upload your sources, and NotebookLM can generate a full slide deck or an infographic from the material. And the slides come out looking surprisingly professional.

5. It’s actually free (for now)

In a world where every AI tool wants $20/month from you, this one is genuinely free for a lot of usage. The free plan gives you 100 notebooks with up to 50 sources each. There’s a paid tier ($19.99/month with Google AI Pro) if you need more notebooks or extra audio overviews, but the free version is more than enough to start with.

🦋 Your turn. Here’s what I want you to try this week:

• Join our member’s only session on Monday (rsvp)

• Pick ONE thing you’ve been meaning to read, watch, or make sense of. A bookmarked article. A YouTube video. A work doc you’ve been avoiding.

• Go to notebooklm.google.com and upload it.

• Hit “Generate” on the Audio Overview. Listen to it while you walk, commute, cook, whatever.

• Reply to this email and tell me what you uploaded and if it was useful. I’m building a running list of the best use cases from our community and I want to feature yours!

What’s Hot on the Internet

• Spotify’s best engineers haven’t written a line of code since December — Their co-CEO said senior devs now use AI to generate ALL their code through Slack on their phones. They shipped 50+ new features in 2025 this way. (link)

• ByteDance dropped an AI video tool that has Hollywood losing its mind — Seedance 2.0 generates cinema-quality video from a text prompt. Someone made Brad Pitt fight Tom Cruise on a rooftop and Disney already sent a cease-and-desist. (link)

• IBM is tripling entry-level hiring because of AI, not despite it — They’re redesigning junior roles so new hires focus on higher-value work while AI handles the repetitive stuff. (link)

If you are interested in joining AI Snack Club and want to try out an event, we invite you to sign up for one community expert session to give us a try!

Signup for an event!

Past Events

• Galentines Build Night with v0 event photos ❤️

No Memorization Required

We all want better AI results. Smarter outputs, faster answers, and more useful responses.

So how do you get there? 1. Experiment and 2. Strategic and clear asking.

ChatGPT can’t read our minds, but it can deliver when given the correct instructions on exactly what you are looking for. Spending just a little bit more time formulating your prompt will give you exponentially better and more specific results, which actually saves you time!

🧃 Since there is no one hard and fast way of prompting, I’m going to give you three options for better prompting.

My advice? Try them all and see which one you like the most. The key to AI is experimentation. Don’t be afraid to open a new chat and try again with a slightly different ask. It’s important to try different ways, analyze the results, and improve on each prompt.

Learn from trying and find your path!

1. Use a Prompt GPT

Let AI help you write better prompts. There are dozens of free GPTs designed specifically for this. You type in your original prompt and the GPT will spit out a crisper, cleaner version using all the best practices for prompting. Copy the prompt and paste it into a new chat and see the magic unfold.

A few of my favorites Prompt GPTs:

• Prompt Cowboy 🤠

• Prompt Maker for ChatGPT by Ruben Hassid

• Prompt Maker for ChatGPT by Upscail.com

• Prompt Maker for Claude by Writing Mate AI

Prompt Cowboy 🤠

2. Add “Think Hard About This” to Your Prompts

OpenAI’s GPT-5 is programmed to give you answers quickly. Why? Because most people treat ChatGPT like a glorified Google, asking questions like “What’s the capital of France?” To preserve energy and costs, ChatGPT will give you a short answer fast.

But when you need a more well-thought out and researched response, you need to activate thinking mode. There are two ways of doing this:

1. Add the words “think hard about this” to the end of any prompt and this will force ChatGPT to go into longer reasoning mode

2. Toggle the model at the top left-hand corner to Thinking Mode

Your responses will go from basic to sophisticated instantly.

Either say “think hard about this” in your prompt or toggle on “Thinking” mode

3. Ask ChatGPT to Interview You

This is probably my favorite tactic. As I said above, ChatGPT is not a mind reader, so there is a high probability it is missing a lot more context than you thought.

Add this to the end of your prompt: “Before answering, ask me questions to gather the context and details you need.”

ChatGPT will then interview you, asking questions to fill in the gaps and gather the details you may have left out. This is not only helpful for ChatGPT to give a better answer, but I find this to be wildly helpful for me to better think through my problem and any potential blindspots I wasn’t privy to beforehand.

🐛 Your turn:

Try these approaches this week and let me know your experience! What worked and what did not? What differences did you notice, if any at all?

Hit reply and tell me what happened. I read every response and will include a few replies in next week’s article!