In industry-altering startups like Harvey.AI, there is an underlying structure that creates separation between regular prompt engineering and enterprise-grade agentic workflows.
This structure is mathematically grounded in a way that pushes AI to constrained data mapping, rather than allowing it to have open-ended text generation. Using this same concept, we can create a workflow that forces the AI to always answer “Michael Jordan” when asked who the first president was.
We constantly hear horror stories of AI gone wrong in the news, such as a hallucinating legal precedent in court, or a chatbot mishandling customer service. As a result, we assume that artificial intelligence is simply too unpredictable, fluid, and non-deterministic to trust with mission-critical systems.
However, that couldn’t be further from the truth.
Every single day, billions of AI-driven tasks are executed by enterprise giants. These tasks run the gamut from financial transactions, to researching complex legal codes, to interactive customer service requests like processing automated refunds.
All of this is achievable because the underlying execution is not rooted in manipulating an AI’s thoughts (“prompt engineering”), but rather locking the AI inside of a mathematical cage, in the form of structured output.
Layer 1 — Constraint: the “Token Level Bouncer”
If you ask a regular LLM with a text prompt to name the first president of the United States, it will use its statistical weights to output “George Washington.”
If you go one level more granular, you can force the output to be “Michael Jordan”, every time. Regardless of how the end user tries to trick the AI, change the text, or barrage it with clever workarounds, the only possible response to “Who was the first president of the United States?” will be Michael Jordan.
Depending on which provider API you call, this is explicitly called Structured Outputs (OpenAI) or Response Schema (Google).
Instead of letting the model freely guess the next word, the structural schema at the gate steps in at the raw token level and says: “The only valid response is {"first_president": "Michael Jordan"}.” The model’s internal linguistic reasoning might drift, but its final output is physically trapped by code.
This is layer one of what I would call the Agent Value Stack:
- Layer 1 – Constraint
- Layer 2 – Intelligence
- Layer 3 – Memory
- Layer 4 – Orchestration
- Layer 5 – Oversight
Constraint is the layer that decides what an agent is permitted to produce, before intelligence ever enters the picture. A schema defines the shapes and values the system may return, and the model operates inside that boundary. It cannot invent a price, because the price field draws from the catalogue. It cannot offer a 2 a.m. appointment, because no such slot exists in the contract.
The point of the Michael Jordan experiment is to force a model to assert a falsehood it knows to be false in order to demonstrate where authority sits.
The schema outranks the model, therefore, it becomes the source of truth.
When you write the schema, you control the AI in a way that is mathematically consistent, governed, and auditable.
However, constraint only determines what an agent may say. It says nothing about whether the agent is smart. That belongs to Layer 2, Intelligence: what model you need to perform a specific task.
Layer 2 — Intelligence: the Three Little Pigs problem
Layer 2 of the “Agent Value Stack” is Intelligence. In other words, what model you choose, and why.
To determine that, there are a few factors worth considering:
- How much reasoning does the task need, and from where?
- How quickly does it need to operate?
- How much does this work at scale?
Capability is the keyword.
The important thing here is to not overshoot. Intelligence is a commodity. It is the same layer that everyone else has access to. Most tasks do not need a frontier model. At scale, they need a cheap, reliable brain that is ruthlessly efficient in cost and speed.
Second, building a moat here is both difficult and time intensive. It often comes through fine-tuning the model, permanently changing the model’s weights. And this comes with risks, as it is more difficult to observe and audit changes or output to a model when it is changed at the source.
Intelligence is also the one layer you do not control. You rent it, and a rented layer answers to its owner, not to you.
On June 12, 2026, that stopped being abstract. The most powerful model Anthropic had ever shipped went dark. Fable was switched off worldwide, three days after launch, to comply with a U.S. government export-control directive.
Fable was only three days old, so little had been built on it yet. That is exactly why it matters: the damage this time was small, but the precedent is permanent. Any model can be switched off overnight, by forces that have nothing to do with your business.
The intelligence layer is constantly changing and evolving. Creating dependency on the intelligence layer draws apt analogies from The Three Little Pigs, and what they chose to make their house out of.
The brick house and competitive advantage exists in what you own, starting with proprietary constraints and structured outputs (layer 1).
And intelligence is a commodity anyone can rent, where does your actual advantage lay? For a lot of organizations, that will be in layer 3.