Vadim Kuhay

A humble system designer, team builder, and player-coach

4 minute read

In my previous article I wrote about a long-lost client who called back to share an astounding story[1] — attaining Phase III of AI-Native evolution in two years. I was amazed because I remembered their architecture: Modular Monolith. But soon it made sense — they had clean boundaries, so adding AI magic was purely incremental for them.

Here I’ll describe their astounding financial gains from Phase I alone.

Why Your AI Fails (And Theirs Doesn’t)!

Your AI can’t tell the difference between a customer order and a lunch order. Origami’s AI predicts market demand with 80% accuracy.

The difference? They taught AI their language FIRST.

While you’re wrestling with chatbots that hallucinate, Origami gained:

  • Sales prediction accuracy: 10% → 70-80%; 🎉

  • Operational cost reduction: 17% in Phase I alone;

  • AI integration failures: Zero!

How? They had something you don’t: Corporate Taxonomy.

Corporate Taxonomy = Teaching AI exactly what every word means in YOUR business.

Not Wikipedia definitions. YOUR definitions.

The Magic Started in 2018 (Before Anyone Cared About AI):

It all began with creating Corporate Taxonomy (CT). From CT, we get an automatic Enterprise Dictionary. Let me explain what CT is and why it changed everything.

Remember, Origami makes high-precision hydraulic power transfer components.
Boring? Maybe. Profitable? Absolutely.

They have three main customer types:

  • Integrators — build bigger machines using Origami parts (predictable, planned);

  • Maintainers — fix industrial machines, need parts NOW (unpredictable, sporadic);

  • Explorers — design new machines, want custom components (expensive, disruptive).

In 2016, all these different needs funneled through one "Front Office."
Chaos. Expensive chaos. Order was the first necessity.

And here’s where it gets interesting…​

Corporate Taxonomy: The Secret Weapon!

Corporate Taxonomy classifies business objects AND behaviors.
This is the magic that makes AI integration seamless.
And it’s exactly why 80% of corporations fail miserably with LLMs today.

Let me show you with a simple example — an Order.

In CT, every artifact has:

  • Name — unique identifier

  • Type — classification

  • Attributes — data/state it owns

  • Policies — business rules it follows

Here’s what Origami’s base business object looks like:

com.origami.BusinessObject:
  parent: null
  name: Object
  api: v3.1.7
  policies:  # Runtime immutable
    - name: Creation
      type: com.origami.policy.OnCreate
    - name: Modification
      type: com.origami.policy.OnModify
    - name: Delete
      type: com.origami.policy.OnDelete
  attributes:
    required:
      immutable:
        - name: id
          type: UUID
        - name: created
          type: Instant
      operational:
        - name: modified
          type: Instant
        - name: description
          type: com.origami.data.RecordList

Look at the order and predictability!
Every business object follows these rules.
Now here’s their Order object:

com.origami.Order:
  parent: com.origami.BusinessObject
  name: Order
  api: v11.1.0
  attributes:
    required:
      immutable:
        - name: origination
          type: com.origami.data.crm.Record
      operational:
        - name: disposition
          type: com.origami.data.ChainOfCustody
        - name: reconciliation
          type: com.origami.data.Reconciliation

The boundaries AI needs are already visible.
But let me make this crystal clear with a restaurant example everyone understands.

The Restaurant Order: Same Word, Different Meanings:

When you sit down at a restaurant, an Order is born. But here’s the magic:

  • To the Waiter: Order = table number, items, special requests, payment;

  • To the Cook: Order = items to prepare, special instructions, timing;

  • To the Manager: Order = revenue, table turnover, inventory impact.

Same order. Different meanings. Different contexts.

This is what your AI doesn’t understand and why it fails.

Each domain has its own private meaning for shared words. The waiter doesn’t care about inventory. The cook doesn’t care about payment methods. But it’s the SAME ORDER.

Quick question: When is an order "closed"?

  • When payment is collected?

  • When the customer leaves?

  • When the bank settles?

  • When quarterly books close?

Think about it. Your answer reveals how confused your own business language is.

The Enterprise Dictionary: Your AI’s Rosetta Stone.

Unlike Webster’s, an Enterprise Dictionary shows:

  • Base definition: "Order = business event including a sale";

  • Waiter’s definition: "Order = table service request";

  • Cook’s definition: "Order = production queue item";

  • Accountant’s definition: "Order = revenue transaction".

See how this makes everything trivially simple for AI?
And not just AI — humans understand better too!

The Accidental AI Preparation:

Here’s the beautiful part: Origami wasn’t preparing for AI.
They were just trying to run their business better.

Two key changes created magic:

  1. Corporate Taxonomy — everyone knew what everything meant.

  2. Public API — customers could self-serve.

This caused:

  • Inversion of Concern: Customers pulled data instead of calling in (commands → queries).

  • Segregation of Control: Each domain owned only its piece of the order lifecycle.

Costs dropped. Efficiency soared. I collected my check and left.
Little did I know…​

The AI Explosion: One Tiny Change, Massive Impact:

Years later, when AI arrived, Origami’s team made one brilliant addition:

com.origami.BusinessObject:
  attributes:
    required:
      immutable:
        - name: id
          type: UUID
        - name: created
          type: Instant
        - name: context  # <-- THE MAGIC
          type: com.origami.DomainContext

See what they did?! They made context mandatory!

Every business object now carried its complete meaning. Not just data — MEANING.

Clever, isn’t it?!

Phase I Results: The Numbers Don’t Lie!

They started simple — added AI to the order placement boundary.
Remember those three customer types?
The ordering portal was a nightmare of complexity.

First enhancement: An LLM on the ordering edge.
Result: The "unpredictable" became predictable!

By analyzing order patterns with context, they discovered:

  • Raw material prices;

  • Interest rates;

  • Geopolitical events.

These three factors explained ALL fluctuations.
And they also discovered why — customers hoard!

Sales prediction accuracy jumped from 10% to 70-80%.

Imagine the manufacturing efficiency gains!
Imagine the inventory optimization!
Imagine the cost savings!

All because every object carried its complete context.

What This Means for YOUR Company?!

If you have:

  • Spaghetti architecture: 6-12 months to build taxonomy first.

  • Some boundaries: 2-3 months to document necessary context.

  • Clean architecture: 2-3 weeks to add AI (like Origami).

Every month you delay:

  • Competitors get further ahead.

  • Your AI projects keep failing.

  • Your costs stay high.

  • Your predictions stay wrong.

The brutal truth: You can’t skip this step.[2] And NO vendor can sell you a shortcut.

The MCP Bonus Round:

When Model Context Protocol arrived, Origami just…​ rolled it out.
No drama. No rearchitecting.
Their context-rich taxonomy fit MCP like a glove.

While you’re having vendor meetings about "AI transformation," they’re in Phase III.

The Two Keys to AI Success:

After 30 years of building systems, it comes down to this:

  1. Semantic Consistency — Every word means ONE thing per context;

  2. Boundary Enforcement — Contexts don’t bleed into each other.

That’s it. That’s the magic.

Your vendors won’t tell you this because they can’t sell it in a box.
(Or, they don’t know.)

Just think about it — you want AI to do your business for you.
Can it get up and walk around trying things out like your employees do?
No?! Then how would it know what you want it to do?

What’s Coming Next:

Phase II is where it gets REALLY interesting — AI agents start talking to each other across boundaries. But that only works if Phase I is rock solid.

I’ll cover those exponential gains in my next article. But first, ask yourself: Does YOUR business even have a common language? (Or, what is Ubiquitous Language?)

If not, every AI dollar you spend is wasted — LLMs are "Language" models, aren’t they?

The Uncomfortable Conclusion:

Everything we’ve discussed is just Systems Engineering.
…​ But notice how natural it feels in business terms?

That’s because people — specifically people’s behaviors — ARE part of the system!
I’ve always designed systems with humans at the core.

Until now.

The world just shifted. For the first time in human history, we can build different systems.
Systems where digital minds understand context as well as humans do.

Maybe better.

But only if you teach them your language first!

This story is part of The Final Industrial Revolution series.
Republished to

Want to teach AI your business language? That’s exactly what we do at ASE Inc.!

No magic. No shortcuts. Just foundations that enable real AI integration.

P.S. — I’m obsessed with teaching teams this craft.
The CTO called years later just to thank me for that.
Turns out, teaching people to fish beats catching fish.
Best of all — they teach me back! How about that?!

P.S.S. — We built our reputation on un∫ucking your most prized initiatives.
Please ask around. You might like what you find 😜

1. See The AI Success Story Nobody’s Talking About: How “Boring” Architecture Beats Hype for more on this story. It’s a part of the AI Integration Architecture for Enterprise series.
2. This article The AI Evolution Playbook: Why 80% Will Fail explains HOW to avoid the TRAP described here American AI Integration Trap — your playbook on integrating AI and staying ahead.

Leave a comment

You may also enjoy