AI Is Another Unicorn

June 30, 2026AI Business

Rogue

The iceberg doesn't care how well you understand the tip.

I knew nothing about finance.

So I did what builders do. I researched. Vocabulary, chart patterns, order books, options chains, the lot. Months of it. I understood enough to feel confident.

Then BlockFi went Chapter 11.

Three years. Frozen. Gone. The vocabulary didn't save me. The charts didn't save me. What got me was what I couldn't see — the legal structure underneath, the custodianship model, the counterparty exposure sitting four layers below the UI I was looking at.

The iceberg doesn't care how well you understand the tip.

The Surface Layer Problem

Everyone's talking about decoupled microservices, serverless architecture, AI-native infrastructure. The language of the surface layer. Clean. Visible. Frameable in a conference talk.

What nobody talks about: the Optimization Engines that have been running airlines, trains, and booking systems for forty years. Revenue management. Yield optimization. Demand forecasting. Cancellation pricing algorithms that would make a quant sweat.

When your flight gets repriced seven times on a Tuesday afternoon while the NYSE is closed — that's not a language model. That's a decades-old math problem that was already solved before the internet existed as a consumer product.

AI didn't replace that. It didn't touch it. It's the new UI on top of the same iceberg.

Nothing Is Free

Here's the recursiveness that breaks your brain if you look at it long enough.

GitHub: free. The best code on the planet, contributed by builders who believed in open source.

Now: the training data for every major LLM. Extracted. Modelled. Sold back as intelligence.

Nothing is free. Someone always paid. Usually the person who didn't read the iceberg.

Developers Used to Be the Iceberg

The ones who understood systems laterally — diagonally — horizontally across the whole stack. Who knew why the legacy code was written the way it was, what the constraint was, what decision was made in 1997 that nobody can unpick in 2026.

That depth was the value. Not the syntax. The judgement underneath it.

Now the surface layer gets produced faster. Cleaner. AI writes the boilerplate in seconds.

And someone in a meeting looks at that and says: so we should be able to do this in a week, right?

The iceberg is still there. It just became invisible again.

The Tell

China just discontinued 3,000 degree types.

Not AI. Not software. Not engineering. History. Philosophy. Literature. Languages. The subjects that produce people who ask why, not just how.

That's the tell.

A country removing the degrees that teach humans to think laterally, while accelerating the ones that produce engineers who execute — that's not an education policy. That's a pipeline decision.

The countries treating AI as a language tool are building faster UIs.

The ones who understand what's underneath are building the next iceberg.

Language Model vs Optimization Engine

The Language Model is not the solution to the hard problems.

It is excellent at approximation. Pattern recognition at scale. Summarisation. Retrieval. Genuinely useful for a certain class of problem.

But an airline's load factor optimization is a constrained integer programming problem. A booking cancellation system is a Markov decision process. A financial custody structure is a legal and mathematical architecture.

You can build a very convincing interface on top of those things with AI.

You cannot replace the thing underneath.

We've Been Here Before

I use AI every day. It makes me faster, cleaner, better at the surface layer work.

But I've seen this before.

Web 2.0 was going to change everything. Blockchain was trustless infrastructure. The Metaverse was the next platform.

Each one had a real capability underneath the hype. Each one got priced like it had already solved problems it hadn't touched yet.

AI has a real capability. A significant one.

But the gap between impressive surface layer tool and solved the optimization problems that actually run critical infrastructure is not a roadmap item. It's a different class of problem.

The iceberg doesn't care about the demo.

Know Which Layer You're On

The moral isn't to not use AI.

The moral is: know which layer you're on.

If you're accelerating your surface work with AI — good. Do more of it.

But if you're pricing your business on the assumption that the iceberg has been solved — check where your funds are custodied.

BlockFi had a very impressive interface.