I get this question constantly: how is your team actually using AI day-to-day? Today I'm answering it in detail — the 5 levels of the system we run, every tool we use, and the part I'm still figuring out.
Don't try to do all 5 in one weekend!
Level 1: Organize your knowledge layer (days 1–6)
Before any of this works, your data has to be organized.
If you skip Level 1, every agent and AI tool you build on top will just create more chaos. I learned this the hard way after rebuilding everything 3 times.
Two changes mattered most:
One: I stopped typing.
Allie Miller told me that the best prompting is complaining to your AI. Voice gives the model 10x more context than typing. You naturally explain the background, mention edge cases, give examples, without thinking about it. I use WisprFlow because it works in Russian and English (Claude doesn't always understand my Russian, but Wispr handles it). For longer recordings like conference talks, interviews, anything I want to process into a post later, I use Trint.
Trint transcribes every episode of my Silicon Valley Girl podcast
Most top founders I've talked to recently have switched to voice. If you're still typing prompts, that's the first thing I'd change this week.
The strategic context files (the part most people skip)
Every AI tool I use also has access to a small set of context files about me. These don't change often. But they add a lot of strategic content to any AI that I'm using, so that every answer is tailored for my situation. They include:
A voice profile — how I write, what I sound like, what I dislike, my emotional markers
A fact dossier — my background, brands, companies, family, schools, what I won't talk about publicly
An anti-AI writing guide — patterns I refuse to publish in my voice (here's my version)
My newsletter and YouTube channels description — who my audience is, what I love to talk about, what the cadence is
When I started my first Claude project, I asked Claude to interview me for an hour. It asked me questions about my work, my preferences, my history, and built a draft voice profile from my answers. My team then cleaned it up. That document is now used by every team member when they write something I'm going to publish.
The output: anyone on my team can write me a post or a script, run it through the same context files, and it comes back sounding like me. Not because they're good at imitating me. Because the context is doing the work.
Action this week: ask Claude or any AI you use to interview you for 20-30 minutes. Then save the output and use it as a project document across all your projects.
I want you to become my second brain. I want you to never invent facts about me. I want you to use only facts that I tell you, in everything you write for me and every decision we make together.
Interview me for the next 30 minutes. Ask one question at a time. Go deep on each answer before moving to the next. Cover: who I am, what I do, what I'm building right now, what my goals are for the next 12 months, what I'm scared of, what I've already tried that didn't work, who my audience is, what they care about, what they're skeptical of, what's outside the scope of what I want to talk about publicly.
At the end, output everything as a single document called about-me.md, formatted with clear headers.Same for Voice profile:
Here are 3 things I wrote that I love and that my audience or my investors loved. Read them. Analyze them deeply.
I want you to understand my tone of voice — the things I would never say, the things I love to say, how I sound when I'm excited, when I'm sad, when I'm being honest about a failure. Notice if I use emojis. Notice if I use double exclamation marks. Notice my sentence length patterns.
Then ask me follow-up questions to fill in what the writing doesn't show. Ask about words I hate. Ask about phrases I refuse to use. Ask about which adjective I'd never pick.
When you're done, output everything as my-voice.md.
Two: I created a database that lives outside any one AI tool.
If you run a business or social media accounts or are doing any projects, you probably have a lot of data associated with that project. It has to have a home. Ideally, it's an external database so that you can migrate from one tool to another.
This matters because tools change every 3 months. Today I love Claude. Tomorrow Codex ships something that makes me want to switch. Next month it's a new Gemini model. If all my decisions, context, and history are locked inside Claude, migration is painful.
So my entire knowledge layer lives in an external database. Organized by channel: YouTube Silicon Valley Girl, YouTube Linguamarina, Instagram, LinkedIn, Podcast (Spotify), Newsletter. Each one has performance data, transcripts, tone of voice, branding, decisions log. Whatever AI tool I'm using that week just connects to it.
Quick example of how this pays off: every podcast episode has its own Notion page in our database. When the transcription gets added to that page, the whole team gets a notification in our team group. That single trigger kicks off clip selection, announcements, newsletter drafting, social posts, all from the same source. Nobody has to message 4 people to say "the transcript is ready."
Level 2: Build AI on top of your knowledge (days 7-12)
Now you teach AI your business deeply enough that it stops needing you to re-explain things.
I use Claude Projects for this. One project per platform: YouTube, Newsletter, Podcast, Instagram, LinkedIn. Each project is built on relevant data and strategic context files I just described.
Right now my team is also testing Claude Cowork, which is the desktop version. The difference: Claude Projects in the browser can read your files and respond. Cowork can also open files, edit documents, and run scripts on your computer. We have a master folder for YouTube production with subfolders for titles, thumbnails, scripting, distribution, and guest research. Each subfolder has its own instructions file telling Claude exactly what to do for that task. The instructions stack: master context first, task context second, then the actual work.
The team's feedback so far: results are far more accurate on the first try.
One example I'm proud of: our guest evaluation system.
When my team wants to vet a podcast guest, they paste a name into our Silicon Valley Girl Claude project. The project scores guests across 8 categories: Results, AI/Business factor, Personal story, Fame fit, Zeitgeist, Bridge factor, PR/Media presence, and Thought leadership. Each one is weighted. Each one is benchmarked against our actual Notion data on how past episodes performed.
At the end, the guest gets a Tier 1–4 rating, a view estimate (something like "120K–220K views"), and an Invite or Pass verdict.
A list of 10 guest names used to take a booker and a strategist 2–3 days. Now it takes under 10 minutes. That's 8–12 hours back per week for the team, and we make better booking decisions because every guest is scored against the same rubric.
Level 3: Scheduled agents (days 13–18)
Scheduled agents are prompts that run on a timer, on their own, without anyone clicking anything.
Right now I have 6 agents running for Silicon Valley Girl.
Every Monday at 9am, one agent does trending content research and drops 10 video ideas into a .docx file. By the time my team opens their laptops, the ideas are ready.
At 10am another agent collects the most important AI, tech and business news from the past 7 days into one summary. So before our Monday standup, we already know what's worth talking about.
At 12pm a third agent runs a full channel health audit covering the last 12 months. We see what's working, what's slipping, what to double down on.
Every day at 12pm a fourth agent scans English-language tech, AI and business media for mentions of the Silicon Valley Girl podcast from the day before. When something good comes up, we know about it the same morning.
Then there's the guest follow-up agent. Every Wednesday at 12pm it reads our declined-guests list in Notion, scores each one, finds strong news hooks from the past 7 days, and recommends who to re-approach this week with a fresh angle. My guest producer used to spend 80% of her week chasing people who hadn't even responded. Now that work runs in the background and she gets a shortlist with draft messages.
She got most of her week back!!
Every scheduled agent is doing one job a human used to do manually. None of them is replacing anyone. They're just removing the part of the job that didn't need a human in the first place.
Level 4: Vibe code your own tools (days 19–24)
This is where it gets interesting.
Luis von Ahn, CEO of Duolingo told me on the podcast that at Duolingo, every single person has built their own dashboard. I loved that idea, so my team and I built our own.
The dashboard pulls data from every social platform connected to the podcast.
I did not write the code. I described what I wanted in Claude chat, Claude turned it into a technical prompt, and Claude Code wrote and edited the actual files on my Mac. I iterated on it for about 8 to 10 weeks between March and May before it did what I wanted. Hosting runs on Railway, costs a few dollars a month.
Right now it's pulling from YouTube (both Silicon Valley Girl and Linguamarina), Instagram, Threads, LinkedIn, X, and Beehiiv. TikTok is waiting on developer approval. Spotify and Apple Podcasts are next.
The interesting part is what it does with the data.
A Python cron job watches every video on my English channel from the moment it's published. There are 3 trigger moments:
Day 4: if click-through drops below 4% with 20,000+ impressions, the system fires. That's an early packaging signal, the title or thumbnail isn't pulling weight.
Day 7: if views are below 12,000, or 30-second retention is below 50%, or intro-end retention is below 45%.
Day 14: closing check. If total views are below the 30K KPI, my team gets a message.
When a trigger fires, the system sends the metrics to Claude (Sonnet) and asks for a diagnostic. The output lands in our team channel looking like this:
Telegram bot that sends alerts to my team if some of my videos underperform
The diagnosis also writes back to our Notion tracker, into the "Action Needed" and "Optimization Notes" fields. So when my YouTube manager opens Notion, she already knows what needs fixing without waiting on a meeting or waiting on me.
The whole thing costs fractions of a cent per video to run.
AI search visibility
Most people don't think about this yet. Right now traffic is shifting from Google search to chatbots. People ask ChatGPT, Perplexity, and Google AI Overview for podcast recommendations, tool recommendations, expert recommendations. The same questions they used to type into Google. The companies that show up in those answers are going to win the next 5 years.
When I tested it earlier this year, my podcast didn't show up at all.
So my team and I started asking the question: how visible are we in AI search? That kicked off a rebuild of the podcast site and JSON-schema, Wikipedia page creation, and update of the descriptions on every platform.
The goal: make the site readable to AI crawlers (GPTBot, PerplexityBot, ClaudeBot) the same way old sites were optimized for Google.
We track Silicon Valley Girl against 11 competitors. Every day, 50 prompts run through ChatGPT, Perplexity, and Google AI Overview to tell us where we show up.
The honest before-and-after:
When we started, SVG only appeared in branded prompts (who is Marina Mogilko?)
Today, SVG appears consistently in 6-8 prompts that do not mention my name at all.
LLMs are starting to feature my podcast in their answers
Prompts we currently win on:
"Best podcasts for women interested in AI and startups"
"Perplexity CEO podcast"
"What podcasts are popular in Silicon Valley?"
"Best podcasts with Silicon Valley CEO interviews"
Is that a lot? No. Is it more than zero? Yes. And every one of those positions is a slot we didn't have a month ago.
Try it for yourself: send your URL to any chatbot you use and ask, "how visible am I in AI search?" Then ask for a to-do list for the next 6 months to improve your visibility.
Level 5: Close the loop (days 25–30)
This is where almost no one is. It's also the level I'm still building.
The piece I'm working on is closing the loop on team conversations.
Every call we run (strategy, 1:1s, planning, guest prep) gets recorded (we use Granola). The notes and the next steps end up in one clean transcript per meeting.
The problem: those notes sit in Granola. They don't get into the Claude project that runs the rest of my business. So when I ask Claude "what did we decide about the SVG editorial calendar last week?", it has no idea. The decision exists, but only in Granola.
Next steps based on a team call transcription
What we are working on right now is a connection between Granola and Claude. When a meeting wraps, the transcript and the next-step list should move straight into the right Claude project. YouTube into the YouTube project, podcast into the podcast project, ops into the ops project. So the next time anyone on the team asks Claude a question about a decision we made, Claude already has it.
This is the gap I've been thinking about for months. Almost no founder has solved it yet. If you have, please reply and tell me how.
If you found anything useful in this email, please forward it to one person on your team. Someone who'd actually build one of these levels. That's how this newsletter grows, and it really helps me out.
Marina 💜
