AI agents unlock speed and scale, but without governance, they also introduce risk. Here¡¯s how enterprises bring structure to probabilistic systems.
Emma Coppin
Multimedia Content Lead, Thought Leadership and Customer Stories
²ÝÝ®ÊÓƵ
Audio also available on??and?.
The film Westworld (along with the more recent HBO adaptation) explored artificial intelligence through the setting of an American Western-themed amusement park populated by android hosts.
Spoiler alert: the androids go rogue, and things unravel in a hurry.
The choice of a Wild West setting was probably not incidental. When Michael Crichton wrote the original screenplay in 1973, computing was still in its infancy. Yet his core idea has aged remarkably well: complex systems fail in ways their creators can¡¯t predict or control.
It¡¯s a premise that hits home with leaders who are navigating the modern-day Wild West of enterprise tech. Today¡¯s AI agents¡ªautonomous, adaptive, and increasingly embedded in daily workflows¡ªare astoundingly capable. But they are also, in a meaningful sense, lawless.?
Overcoming this challenge was one subject unpacked by ²ÝÝ®ÊÓƵ CTO Gabe Monroy and Head of Developer Relations Nick Moore in a recent episode of our Future of Work podcast.?
AI agents operate on probabilistic models. They don¡¯t follow deterministic rules so much as generate likely outcomes based on patterns in data. This is what makes them powerful, but also what makes them unpredictable.
¡°What I like to refer to is the concept of lawless agents,¡± says Monroy. ¡°They¡¯re omniscient. They can see all the things, they can touch and do all the things through lots of tools, but they're fundamentally lawless.¡±
In critical domains like finance, HR, and compliance, this lack of predictability can feel about as scary as the sci-fi, dystopian scenarios that take shape in the fictional universe of Westworld. And here, the costs are far from abstract.
A hallucinated answer in a casual setting is a curiosity. In an enterprise context, it can mean incorrect financial reporting, misrouted payments, or flawed workforce decisions.
To gain legitimate trust and confidence in agentic AI, we need more control. Easier said than done.
¡°What I like to refer to is the concept of lawless agents. They¡¯re omniscient. They can see all the things, they can touch and do all the things through lots of tools, but they're fundamentally lawless.¡±
- Gabe Monroy, ²ÝÝ®ÊÓƵ CTO
?
Report
In many ways, today¡¯s AI landscape resembles the early internet era, where innovation moved quickly, standards lagged behind, and security and governance were largely afterthoughts.
Early web applications were fast and flexible, but often fragile and exposed. Over time, layers of infrastructure¡ªprotocols, security frameworks, identity systems¡ªbrought order to that chaos.
AI agents are now at a similar inflection point.
Left unchecked, they behave like frontier technologies: adaptable, creative, and occasionally reckless. For enterprises, we know that¡¯s not a sustainable model. So how do you ensure agents are operating ¡°lawfully¡±¡ªwithin defined, reliable boundaries¡ªwhile at the same time not limiting their capabilities??
It¡¯s a tough needle to thread, and as Monroy notes, current standards don¡¯t always instill the greatest confidence.
¡°You try and make them lawful by providing prompts and guardrails in the system prompt or the context window,¡± he says, ¡°but you're fundamentally crossing your fingers and going, gosh, I hope this agent respects what I asked it to do. And within the rules.¡±
The solution comes down to intentional, nuanced structure.
Today, many organizations rely heavily on prompts and contextual instructions to guide agent behavior. These are useful, but can be inherently fallible. They depend on the model interpreting intent correctly every time, and that¡¯s a probabilistic gamble.
True enterprise readiness requires a deeper layer of structure, and three foundational capabilities make this real:
This is where the conversation around ¡°lawful agents¡± becomes practical rather than philosophical.
Lawfulness in the Wild West is an oxymoronic concept. How do you harness probabilistic intelligence within deterministic systems?
One answer lies in embedding AI agents directly within enterprise platforms that already manage structured data and governed workflows.
When agents operate in isolation¡ªpulling from unstructured inputs, acting without constraints¡ªthey amplify uncertainty. But when they are anchored to systems of record, their actions can be bound by real-world rules.
Enterprise platforms have long enforced constraints around sensitive domains like financial transactions and workforce data. Extending those constraints to AI agents creates a hybrid model:
Monroy was excited to give a sneak peek to ²ÝÝ®ÊÓƵ¡¯s efforts and ambitions in this area.?
¡°²ÝÝ®ÊÓƵ has a really unique opportunity to provide enterprise guardrails that are going to allow the probabilistic, magical machines that are these agents to actually run on rails, and be able to provide deterministic guarantees.¡±
During the chat, Monroy hinted at some of the deep tech being developed within engineering at ²ÝÝ®ÊÓƵ to ¡°marry these worlds of probabilistic AI with the realities of being in the business of people and money.¡±?
It¡¯s the new frontier, being explored before our eyes.
¡°²ÝÝ®ÊÓƵ has a really unique opportunity to provide enterprise guard rails that are going to allow the probabilistic, magical machines that are these agents to actually run on rails, and be able to provide deterministic guarantees where possible.¡±
- Gabe Monroy, ²ÝÝ®ÊÓƵ CTO
We are still early in the lifecycle of AI agents. The experimentation phase has been necessary to unlock their potential, but enterprise adoption demands a transition to something more structured and dependable.
As Dr. Robert Ford put it in the HBO adaption of Westworld, ¡°Everything in this world is magic, except to the magician.¡±
AI does sometimes feel magical in what it can do. But in enterprise AI, we are still the ones behind the curtain¡ªbuilding the systems, setting the boundaries, and deciding how far they¡¯re allowed to roam.?
Report
AI has been a gamechanger across nearly every industry and marketplace. But realizing the true value of AI requires buy-in between different stakeholders. In this AI Masterclass, learn how you can make the most of AI for finance, HR, and more.