· 3 min read

🛡️ Wrong War, Wrong Wall

France built the perfect fortress and lost Paris in six weeks. Schools are doing the same thing with AI detection — scanning 200M papers while students route around it. The real opportunity is one step upstream.

🛡️ Wrong War, Wrong Wall

France spent the 1930s building the most expensive wall in history. The Maginot Line ran along the German border, a chain of underground forts with rail lines, air conditioning, and artillery that could shred anything coming at it head-on. The thing was genuinely impregnable. Soldiers lived down there like it was a submarine.

Then May 1940 arrived, and the Germans just didn't come at it head-on. They drove their tanks through the Ardennes, a forest to the north the French generals had written off as too dense for armor. Straight around the wall. Paris fell in six weeks, and the Maginot Line barely fired a shot. It held perfectly, defending a border nobody bothered to attack.

That's the trap with a fortress. You pour everything into stopping the last war, and a smarter enemy strolls around the part you forgot to guard.

Schools are building that exact wall right now. Generative AI broke the take-home essay, so Turnitin scanned over 200 million papers hunting for AI writing while the cheating rate barely budged. The whole edtech market rushed to build a taller wall at the border. Students get around it by lunch.

The real opening sits one step upstream, in assessment design. Call it AssignmentForge: an instructor pastes in a vulnerable assignment and gets back a redesigned package. Process scaffolds, oral-defense prompts, source-trail rubrics, AI-use policy language. Work the student actually has to show up and do.

You don't need any proprietary AI to build it. 700 instructors at $149/year is roughly $9K MRR solo, and a handful of $12K–$35K institutional contracts push it past $20K. The moat isn't the model anyone can call. It's owning the assessment-pattern library nobody else is bothering to build.

Read the full playbook here:

AI detection is the wrong fight. Turnitin scanned 200M papers and the cheating rate didn't move. The durable business is one layer upstream: helping instructors redesign assignments before AI can hollow them out.

Full Playbook

From the Vault:

Japan's sticker boom and the digicam revival aren't separate trends — they're one consumer mood nobody has packaged into a brand yet.

Full Playbook

Pet spending hit $158B in 2025, yet 52% of owners skip recommended care. No one decodes the estimate at the moment of panic — that silence is the business.

Full Playbook

Read next

🧵 Levi's Wrote HR's Dress Code

🧵 Levi's Wrote HR's Dress Code

In 1992 Levi's mailed an eight-page dress code guide to 25,000 HR managers, then sold the khakis that complied. Casual dress went from 66% of companies to 90% in three years. Weddings are stuck in the same fog now, and nobody is selling the guide. Today's idea sells it to planners.

Startup Heist | Briefings
Startup Heist | Briefings
· 3 min read
🧪 A Hundred Twins Beat Twenty Thousand

🧪 A Hundred Twins Beat Twenty Thousand

In 1930, Scotland fed 20,000 schoolchildren to settle a nutrition question and learned nothing, because teachers picked who got the milk. Scale never fixes a bad read. The same trap is burning ad budgets today, where only 19% of creative tests produce a winner.

Startup Heist | Briefings
Startup Heist | Briefings
· 3 min read
🧾 The Count Was Never Wrong

🧾 The Count Was Never Wrong

Quebec's strategic maple syrup reserve lost $18 million because the barrel count always matched. Nobody checked what was inside. Retail has the same blind spot between the receiving dock and the shelf, and Shopify POS never covered it. Today: a $79-a-month app for trading card shops.

Startup Heist | Briefings
Startup Heist | Briefings
· 3 min read
New startup opportunities, ideas and insights right in your inbox.