Aug 17, 2026 · 21 min · 8 segments
The vocabulary of agentic AI has turned over five times in about eighteen months — prompt engineering, context engineering, harness engineering, loop engineering, and now graph engineering. For a…
I'm Chris Bradley, Chief Marketing Officer at Veritiv and head of our AI Transformation Lab.
Today, I want to talk about a set of words you've probably seen scrolling past you this year, each one announced as the thing that replaced the last one, prompt engineering, then context engineering, then harness engineering, then loop engineering, and now graph engineering.
If you're leading a business rather than building the systems, the reasonable reaction is to tune it out.
But underneath the churn, something real is happening, and it's worth twenty minutes of your attention because it explains the single most consistent pattern I see in enterprise AI, why the same frontier model succeeds at one company and fails at another, running the identical task.
Every one of them is the field discovering again that the model was never the constraint.
They are also available to your competitors on the same terms at roughly the same price by tomorrow afternoon.
So the difference between an AI program that works and one that doesn't is almost never the model.
Each of those five terms names one layer of that surround, and they don't replace each other.
Harness engineering is the one in the middle, and it's the one we have the most direct experience with, the runnable environment you put around a model, so it can actually do something.
So today, three layers in order, what the agent can see, how it works, and how many of them work together, context, loop, graph.
Uh, one quick note before we get into it, especially for anyone listening from inside Veritiv.
Employees should use the AI tools that are approved and sanctioned by the company for their role.
And whatever tool you are working in, never put sensitive or confidential information into it unless your company has explicitly sanctioned that tool for that use.
The views in this episode are based on my own experimentation and on the work we do in the lab.
They are not endorsements, procurement decisions, or official positions of Veritiv regarding any specific vendor or product.
Read the full transcript.
Create an account to read the whole episode, search across every transcript, and follow the shows you care about.
I'm Chris Bradley, Chief Marketing Officer at Veritiv and head of our AI Transformation Lab.
Today, I want to talk about a set of words you've probably seen scrolling past you this year, each one announced as the thing that replaced the last one, prompt engineering, then context engineering, then harness engineering, then loop engineering, and now graph engineering.
If you're leading a business rather than building the systems, the reasonable reaction is to tune it out.
But underneath the churn, something real is happening, and it's worth twenty minutes of your attention because it explains the single most consistent pattern I see in enterprise AI, why the same frontier model succeeds at one company and fails at another, running the identical task.
Every one of them is the field discovering again that the model was never the constraint.
They are also available to your competitors on the same terms at roughly the same price by tomorrow afternoon.
So the difference between an AI program that works and one that doesn't is almost never the model.
Each of those five terms names one layer of that surround, and they don't replace each other.
Harness engineering is the one in the middle, and it's the one we have the most direct experience with, the runnable environment you put around a model, so it can actually do something.
So today, three layers in order, what the agent can see, how it works, and how many of them work together, context, loop, graph.
Uh, one quick note before we get into it, especially for anyone listening from inside Veritiv.
Employees should use the AI tools that are approved and sanctioned by the company for their role.
And whatever tool you are working in, never put sensitive or confidential information into it unless your company has explicitly sanctioned that tool for that use.
The views in this episode are based on my own experimentation and on the work we do in the lab.
They are not endorsements, procurement decisions, or official positions of Veritiv regarding any specific vendor or product.