🐺 This AI unicorn started with a guy named Fred taking notes for $100/month

 📰 What’s Trending

Milestone raises $10M to make sure AI rhymes with ROI 
Israeli startup Milestone just raised $10 million to correlate AI tool usage with actual engineering metrics, including code quality. Finally, someone's asking the question we all should be: is all this AI actually making us better at our jobs, or just faster at writing bad code? 🔗 Read on TechCrunch

ElevenLabs strikes deals with Michael Caine and McConaughey for AI voices ElevenLabs locked down deals with Michael Caine and Matthew McConaughey to AI-generate their voices. The actors get paid, the company gets legitimacy, and we get to hear "alright, alright, alright" in our navigation apps. This licensing model might be the path forward for AI voice tech. 🔗 Read on TechCrunch

Europe catching up to US in AI app layer funding 
European and Israeli AI apps have raised 66 cents for every dollar raised by US counterparts in 2025, up from 10 cents a decade ago. While the US dominates foundation models, the application layer is more competitive than most realize. 🔗 Read on TechCrunch

Is Wall Street losing faith in AI? 
Tech stocks took a beating this week with Palantir down 11%, Oracle declining 9%, and Nvidia losing 7%. Meta and Microsoft's commitment to continue heavy AI spending didn't help. As one analyst put it: "Valuations are stretched... good news is just not enough to move the needle." 🔗 Read on TechCrunch

Inception raises $50M for diffusion models in code generation 
Stanford professor's startup just raised $50 million from Nvidia, Microsoft, and Snowflake to apply image-generation tech to code. They're betting diffusion models (think Midjourney) can outperform traditional text models for large codebases. 🔗 Read on TechCrunch

Google's AI Mode gets agentic capabilities for ticket booking 
Google's rolling out AI that can actually book event tickets and beauty appointments across multiple sites. No more tab juggling to find concert tickets. This is the kind of practical AI application that actually saves time. 🔗 Read on TechCrunch

Elad Gil says certain AI markets are already won 
The legendary investor thinks foundation models have clear winners, but tons of AI applications are still wide open. His take: "AI was the one market where the more I learn, the less I know." Even the pros are confused. 🔗 Read on TechCrunch

🧠 The $100 Manual Note-Taker That Built a Unicorn

So Sam Udotong from Fireflies.ai just dropped this bombshell on LinkedIn, and I had to share because it's both hilarious and brilliant.

Before Fireflies had any actual tech - before they had AI, before they had transcription, before they had anything remotely resembling automation - they were charging $100/month for Fred from Fireflies to manually dial into your meetings and take notes.

Think about that for a second. Some dude was literally sitting on calls, silently typing notes, pretending to be an AI. They did this just to pay rent. And now they're valued at over $1 billion.

Here's the thing: this is genius product validation disguised as desperation.

Why This Actually Worked

They proved people would pay for meeting notes before burning a single dollar on development. Most founders would spend 6 months building the perfect AI transcription engine, then discover nobody wants to pay for it. These guys flipped it - prove the demand, then build the solution.

The manual approach also taught them exactly what customers actually needed. When you're personally sitting through 8 hours of sales calls every day, you learn real quick what information matters and what's just noise. That's product research you can't get from surveys.

The Pattern I'm Seeing Everywhere

This isn't isolated. I've watched multiple B2B companies pull this same move:

• Scale AI started with humans doing data labeling

• Flexport manually coordinated shipments via spreadsheets

• Stripe manually processed payments for their first customers

They all validated demand with manual processes before automating. The fancy term is "Wizard of Oz MVP" but really it's just being smart about not building things nobody wants.

What This Means for B2B Founders

Stop obsessing over having perfect tech on day one. Your early customers don't care if there's a human behind the curtain as long as their problem gets solved. In fact, having humans involved early on gives you:

• Faster iteration cycles (humans adapt instantly, code doesn't)

• Better customer insights (you're literally in their workflow)

• Lower initial investment (pay per task vs. massive dev costs)

The real test isn't whether you can build it. It's whether someone will pay for it. And you can test that with duct tape and elbow grease.

The Uncomfortable Truth

Most AI companies today are still partially manual behind the scenes. That chatbot that seems so smart? There's probably a human reviewing and tweaking responses. That instant document processor? Someone's QA-ing the output.

And that's totally fine. The goal isn't to eliminate humans completely - it's to solve problems profitably. If that means having Fred from Fireflies secretly taking notes while you build the actual tech, so be it.

Sam and his team went from charging $100 for manual note-taking to building a billion-dollar company. Not because they had the best tech from day one, but because they understood what problem they were solving and weren't too proud to solve it manually first.

Next time you're overthinking your MVP, remember: sometimes the best AI is just some guy named Fred with a good typing speed.

💡Smart Moves

Voice AI: Get the Proof. Avoid the Hype.

Deepgram interviewed 400 senior leaders on voice AI adoption: 97% already use it, 84% will increase budgets, yet only 21% are very satisfied with legacy agents. See where enterprises deploy human-like voice AI agents - customer service, task automation, order capture. Benchmark your roadmap against $100M peers for 2026 priorities.

Download the Report

⏳ Conversion Corner

Everyone's using AI to generate landing pages, but they're missing the biggest conversion opportunity: testing offers that don't exist yet.

The Pattern: Companies are building entire products before validating demand, burning through runway on features nobody asked for.

The Problem: You're solving for the wrong variable. The question isn't "can we build this?" - it's "will anyone buy this?"

The Fix: Create a landing page for a feature that doesn't exist. Drive real traffic. Measure actual intent. When someone clicks "Start Free Trial," show them a "Coming Soon" message and capture their email.

Track two metrics:

• Click-through rate on your CTA (interest)

• Email capture rate after the reveal (genuine intent)

If you get >5% CTR and >30% email capture, build it. Otherwise, pivot.

The Funnel Lens: This approach shortens your learning cycle from months to days. For B2B SaaS with longer sales cycles, you're validating product-market fit before writing a single line of code. We tested 12 feature concepts this way for a client - only 3 passed the threshold, saving them $200k in wasted dev costs.

Mini Challenge: Pick your most ambitious feature idea. Give yourself 2 hours to create a fake door test. If you're not slightly embarrassed by how simple it is, you're overthinking it.

💎 Data Gem

89% of AI startups still use human-in-the-loop processes

According to a 2025 Bessemer Venture Partners survey, nearly 9 out of 10 AI companies employ humans for quality control, edge case handling, or training data generation. The fully autonomous AI company remains largely a myth.

So, don't feel bad about your manual processes. Even OpenAI has armies of humans reviewing outputs. Focus on solving the problem first, automation second.