A few years ago, I got a call from a CEO who needed an entirely new system built for his company. His old system had cost him millions.
Not the purchase price. Not because it failed. Because it worked perfectly.
That sounded crazy enough that I got on the first flight to LA to hear the story.
A Business That Runs on Urgency
His company sold steel. The various weights, strengths, and sizes used in construction projects, mostly purchased long before a project starts.
But what happens when you're in the middle of building a hotel, working against a tight timeline, and you run out of critical supplies?
You can't wait months for materials to arrive from an international supplier. You need a small amount, and you need it now.
That's where this guy's company comes in. He sells steel at a premium to construction companies in a bind. Speed costs money. His customers know that and pay for it.
I know. At this point you're wondering what this has to do with AI or rule engines.
Trust me. It's worth it.
The Rule That Made Perfect Sense
His business software had a rule built into it. A smart one, actually.
If a SKU isn't selling, create a 40% discount and alert the sales team.
In his business, you can't hold inventory. Steel sitting in a warehouse is money bleeding out. So when something isn't moving, you move it. Fast. Even at a loss.
The system did exactly what it was supposed to do.
And it cost him millions.
The Million-Dollar Mistake
Here's what happened.
He had purchased some new supplies, a shorter and lighter steel beam he didn't normally carry. The shipment arrived and was logged into the system.
But the rule noticed no sales. Zero. So it did what it was designed to do.
It auto-applied the 40% discount and alerted the sales team.
And the sales team sold it. At huge discounts. To customers who would have gladly paid premium prices for hard-to-find materials.
The rule was right. The system was working. And millions walked out the door.
The Hidden Problem No One's Talking About
Here's the thing. The rule engine did its job. It captured the business logic perfectly. “If no sales, then discount and alert.”
But it didn't know why that rule existed.
It didn't know the rule was designed for stale inventory, not new arrivals. It didn't know that a brand-new SKU with zero sales history is completely different from an old SKU that stopped selling. It didn't understand the context that made the rule appropriate in the first place.
And this is exactly where the current conversation about AI agents is headed.
Rules and Guides Aren't Enough
The latest conversations in the enterprise AI space have centered on context graphs. And for the last year, I've been discovering the same thing over and over.
AI doesn't just need to know the rules.
It needs to know the context around the decisions. The “why” behind them. The explanations for exceptions. The conditions that make a rule valid, and the conditions that make it dangerous.
Think about it. Every business rule you've ever written has invisible context attached to it.
“Discount slow-moving inventory” sounds simple. But the real rule, the complete rule, includes all the unspoken assumptions. It assumes the inventory has been in the system long enough to have a sales pattern. It assumes the discount won't undercut your value proposition. It assumes your sales team understands when to override it.
None of that is captured in the rule itself.
What I Love about Context Discussions
When we talk about giving AI agents the ability to make decisions, we're not just handing over a rulebook. We're asking them to operate in a world full of exceptions, edge cases, and situational judgment.
A rule engine captures what you decided.
Context captures why you decided it, when it applies, and when it doesn't.
Here's what that looks like in practice.
Instead of just encoding “discount slow-moving inventory,” you capture the reasoning. This rule exists because holding inventory costs money. It applies to SKUs that have been in the system for more than 90 days with declining sales velocity. It does not apply to new arrivals, seasonal items, or premium-priced materials where discounting undermines positioning.
Now your AI agent has something to work with. Not just a trigger and an action, but the judgment required to know when that action makes sense.
The Competitive Advantage of Context
Most companies building AI agents right now are focused on capturing their business rules. That's table stakes. It's necessary, but it's not sufficient.
The companies that will win are the ones capturing the context around those rules. The reasoning. The exceptions. The “why.”
I spent today with my team challenging them to leverage AI in new ways by learning to document context: their reasoning, decision making, and strategies. Because an AI that knows your rules will do exactly what the steel company's system did. It will follow them perfectly, even when following them is exactly wrong.
An AI that knows the context will pause. It will recognize when a situation falls outside the conditions where the rule applies. It will ask for guidance, or make a judgment call based on the reasoning behind the rule.
That's the difference between a system that costs you millions and a system that saves you millions.
What I Shared with My Team
If you're building AI agents, or preparing your organization for them, start capturing more than just the rules.
For every business rule, document the reasoning. Why does this rule exist? What problem was it solving when someone created it?
Document the boundaries. When does this rule apply? When does it not? What conditions have to be true for this rule to make sense?
Document the exceptions. What situations override this rule? Who has authority to ignore it? What signals indicate this might be one of those situations?
This isn't just good documentation. It's the context layer that will make the difference between AI that helps and AI that hurts.
The Conversation Has Shifted
A year ago, people were talking about whether AI could even be trusted to be involved in business decisions.
Now the question is different. How do we give AI the judgment to know when different guides apply?
The answer isn't a better rule engine. It's a richer context layer. One that captures not just what you decided, but everything that made that decision make sense.
Because your AI knows the rules. The question is whether it knows enough to use them wisely.