Alex Vacca

Alex VaccaAug 26, 2026

At least half of my LinkedIn feed now reads like it was written by the same person.   It was.   AI drafts most B2B copy, ours at Frontal AI included. The difference is what happens before we hit publish: our team built a Claude skill that strips out every AI slop pattern.   Here are our top hacks to make AI write good copy:   → Give AI unique voice references   AI is bad at creating an original voice. But it's very good at mimicking patterns.   Give it posts you wrote before AI. Or posts from copywriters you liked pre-AI.   The output gets way more personality.   → Ban the obvious AI patterns   AI has its own way of writing. And now it's somehow become the world's way of writing.   "Most teams think X. The reality is Y." "It's X. It's Y. It's Z." "Everyone does X. Nobody does Y."   They're everywhere, and they make your copy sound like everyone else.   → Mix up sentence rhythm   AI gives every sentence the same pace. Mix short punchy lines with longer ones, that change of rhythm adds personality and keeps people reading.   The cheatsheet breaks down the rest of the patterns.   And if you want the full Anti-AI Slop skill that we built, ready to drop into Claude, get it here 👇

Alex VaccaAug 25, 2026

These 21 MCP connections run Frontal AI's GTM from one interface.   Many GTM teams still run one campaign across a stack of separate tabs: list building in one, enrichment in another, sequencing in a third, CRM updates at the end of the day.   You would not guess it from the GTM bubble on LinkedIn, but it is what we keep seeing in the companies we audit.   We run most of the work from Claude Code. Each MCP connection plugs Claude straight into one of our tools.   Claude researches accounts through Clay or Firecrawl, updates CRM records, launches campaigns, pulls website data, writes briefs, logs tasks, and moves data between systems.   Here's the stack:   WORKSPACE + BUILD → Notion: content system plus the wiki Claude pulls context from → Slack: team and client comms in one place → ClickUp: company-wide project management → Miro: workflow design and visualization → GitHub: warehouse for all our agents and workflows → n8n: still carries plenty of quick-win automations   PIPELINE + CRM → HubSpot: every client and sales record → Apollo.io: quick enrichment and TAM mapping → LeadMagic: ad activity and hiring signals that show which accounts are worth working → lemlist: email sequencing with 40+ actions Claude can run → Prospeo.io: email finding and verification inside our Clay workflows → Instantly.ai: email sequencing → Clay: runs the workflows we built in Clay, straight from Claude Code → Ocean.io: lookalike list building → AI Ark: TAM mapping and lead enrichment   RESEARCH + INFRA → Firecrawl: turns any website into clean data → Apify: data collection for everything else → Exa: semantic AI search → Browserbase: AI agents operating live web pages → Supabase: data warehousing → Fireflies.ai: meeting recording   Which MCPs are you connecting?   P.S. Credit to Kenny Damian, our Head of GTM, for mapping out the full stack.

Alex VaccaAug 21, 2026

This is Andrej Karpathy, and 2 words he tweeted without thinking became the Word of the Year.   The tweet took him seconds. The credibility that made it stick took 20 years.   His story starts in Bratislava, Slovakia. At 15, his family moves to Toronto.   In 2006 he's a University of Toronto undergrad posting Rubik's Cube tutorials on YouTube. The videos are patient and unusually clear. 9 million views. Some of the fastest speedcubers in the world learned from that channel.   Remember that habit. It's the whole story.   At Stanford he does his PhD in computer vision under Fei-Fei Li. Then he co-creates CS231n, Stanford's first deep learning course, and puts every lecture online for free. Enrollment jumps from 150 to 750 in two years, and the free version trains a generation of AI engineers.   What follows reads like a tour of AI's center of gravity:   → 2015: founding member of OpenAI → 2017: Elon Musk hires him to lead Tesla's Autopilot vision team → 2022: he walks away from one of the biggest jobs in AI   And instead of taking a VP seat somewhere, he goes back to YouTube. Multi-hour lectures, building GPT from scratch, free for anyone.   Then, February 2, 2025. He fires off a tweet about a new way of building software: "fully give in to the vibes, embrace exponentials, and forget that the code even exists."   He calls it vibe coding.   Nine months later, Collins Dictionary makes it Word of the Year. Replit and Lovable now market themselves as vibe coding platforms, a category named by one unplanned post.   And this May, Anthropic hired him to lead a research team. Announced, naturally, in a post.   Here's the detail I can't stop thinking about.   Karpathy later called that tweet a "shower of thoughts" he fired off without thinking, and admitted that after 17 years of posting he still can't predict which post will hit.   The man who named the Word of the Year can't pick winners.   He spent two decades explaining hard things clearly, in public, for free. So when he said the right two words at the right moment, an entire industry repeated them for him.   That's what a content flywheel looks like at full speed. You can't spin one up the week you launch.   Follow Alex Vacca for more breakdowns on building and selling in the AI era 👆