Many people dream that in an ideal world there would be no limits. They would be able to do whatever they want, exactly when they want to. They could achieve all their goals with no effort and no rules.
A person who tests this out quickly discovers life with no limits means that nothing happens. It means never leaving the couch because there are too many possibilities to choose from. It’s scrolling through Netflix for an hour without choosing anything to watch because you can’t decide what you’re in the mood for. I may have done this once or twice.
Life requires structure, form, and constraints.
Animals have skeletal structures that determine and limit their skills. A seal isn’t great at running but is an excellent swimmer. A cheetah can’t peel a banana but can run as fast as a car. I can’t fly but I can sit at my desk for seven hours straight drinking matcha and working on theoretical math problems.
These are physical constraints, and they’re all around us. But other forms of constraint exist as well. Constraints facilitate creation. In the arts, there are countless examples of strict form focusing creativity. Haiku, sonnets. Formal guidelines push the writer to reach a more interesting place by concentrating the thought. There’s a reason we frame artwork, after all.
But not all constraints are created equally.
Some reduce capability. I’ll show you how.
A finance example is multisigs, a technology that uses multiple keys as a security measure, most often in a cryptocurrency context. The constraint here is that more than one signatory must sign off on a transaction before it is approved. This system is good for certain things; when you have an established policy that you want to update, for example. But when you apply multisigs to settlement, to permitting or not permitting a trade to go through, not only does the constraint become less useful, it also provides a false sense of security.
Multisigs are not fail-proof. Signatories can be exploited. It’s possible to swap in a fake user interface for another. So you think you’re approving one thing, but really have agreed to something else.
The approval process is constrained by humans in the least effective way. Coordinating a group of people, one of whom is always on vacation, to approve a time-sensitive trade means you’ll always be playing catch up.
So the constraints in this system cause you to lose capability as well as security.
The right constraints unleash capabilities.
Let’s go back to creativity for a moment. I am both ashamed and unashamed to admit that writing was very difficult for me. Despite the fact I have an intelligence for it. For a long time, I found myself laboring under the wrong constraints: Does the sentence sound smart? Does it sound prestigious? Am I writing about things that people will pay money to read?
These constraints weren’t allowing me to utilize my intelligence. They were hampering me at each turn. When I finally realized this, I knew I had to find a better constraint. Every time I put the pen down, I’d ask myself, “Am I interested in this material?”
And I saw firsthand the difference that changing a constraint can make. Suddenly I was able to access parts of my intuition and intelligence that had previously felt off-limits.
I applied this constraint moment by moment. Action by action. It was the lens through which I viewed every word. It wasn’t something I checked once a week or even once a day. I was checking continuously. Because that’s how you guide behavior.
That’s how you unleash capability.
The problem with AI is not that we don’t have good enough intelligence.
The problem is that this intelligence needs to come with better constraints. David Foster Wallace was a genius. And as we saw, genius can lead to danger. It is tremendously productive and yet also destructive on the same scale. AI is the same way.
AI should be incredibly powerful. The huge amounts of money that have been poured into AI show that people have great expectations of its abilities. But so far this potential has been largely unrealized. AI agents are not used in production systems. Agents have the raw intelligence needed to handle production coding, but instead of being allowed to work they are put in sandboxes because we don’t trust them.
What’s missing is the right constraint to unleash that capability.
Here’s a simple formula that took me a long time to work out:
Intelligence + the right constraint = productivity.
Inherence is the constraint technology we need.
Inherence provides effective constraints that are tailored to the abilities of AI and bring out the best in them. And, crucially, we constrain the model correctly.
There are many ways to constrain an AI agent incorrectly. You can ask, “Did your AI’s inference run correctly?” You can ask, “Were you able to watch your AI run?” You can ask, “Was your AI fake or tampered with?” Or even, “Did another AI think your AI did a good job?”
You can have the right answer to all of these questions and still the AI can go out of bounds.
Because all these things (ZKML, observability, hardware-based trust, LLM-as-judge) have value, but they don’t prove and enforce overall conduct.
The question that needs proving is, “Did your AI do what it was told?”
This is what we do. We designed our system to constrain AI in the ways that matter.
What does that look like? An AI agent that follows the rules that you give them. A purchasing agent that only buys from approved vendors and follows spending limits. That doesn’t give away your trade secrets. That doesn’t hack into other servers.
Inherence gives you peace of mind to let your AI work: It guarantees your agent’s conduct using cryptographic proof. AI is a protean shapeshifter without a body. Inherence provides the skeleton and structure for every role your agent takes on.
If you’re looking for the right constraint, give us a call–see what you can unlock.