*And not just any AI — an agentic one. An AI tool that answers a question is easy to file under “technology.” But agentic AI doesn’t just answer; it decides, acts, and pursues goals across steps. The moment it acts autonomously, it stops behaving like a tool and starts behaving like an agent in the network — which is exactly why the ecosystem questions we ask of suppliers and couriers now apply to it too. In short, they are all agents in the ecosystem. None of them is the ecosystem.
It sounds obvious once you say it. But watch how differently we treat the first three and the fourth.
When a supplier underperforms, no one concludes that suppliers don’t work. They ask the ecosystem questions — the ones every procurement professional asks in their sleep. What authority did it have? What decisions did it touch? How was performance measured? What were the escalation paths? What happens when it’s wrong? A supplier influences outcomes. It does not determine them. The outcome comes from how the whole network — incentives, dependencies, decision rights, governance — absorbs what that supplier does.
We ask the same of a courier. The same of a consultant. The same, instinctively, of a new hire. We have spent centuries learning to govern human agents and decades learning to govern supplier and logistics ecosystems. The questions are second nature.
Then an AI agent enters, and the questions vanish.
Suddenly the conversation is about models, prompts, architecture, orchestration. The agent’s capabilities fill the whole frame, and the ecosystem questions — authority, accountability, performance, escalation, what happens when it’s wrong — quietly drop out. We stop asking how the system governs the agent and start asking only what the agent can do.
That is the entire mistake, in one move.
Where the analogy holds, and where it doesn’t
Let me be precise, because a careful reader will push here, and they should.
An AI agent is not a supplier in every respect. A supplier has its own objectives, economic interests, and judgment. It can negotiate, adapt strategically, withhold information, and decide whether to participate at all. An AI agent has none of that — no independent goals, no stake, no will. It produces outputs from its training and the task in front of it. The mechanics are genuinely different, and pretending otherwise would be sloppy.
But here is what survives the difference: the governance questions are identical. What authority does it have? What decisions does it influence? What happens when it’s wrong? How is its performance monitored? What are the escalation paths? Those questions do not disappear because the agent is software instead of a company. If anything, they matter more — because an agent with no judgment of its own will execute a misaligned instruction faster and more completely than any supplier ever would.
So the analogy isn’t “AI is a supplier.” It’s “AI is an agent in the ecosystem, and every agent in the ecosystem must be governed — regardless of what it’s made of.”
There’s a deeper parallel still. Training an AI is itself a buyer–supplier relationship in miniature. The developer rewards the system for what it gets right and steers it away from what it shouldn’t do — reinforcing some behaviors, discouraging others, until the agent’s tendencies are shaped by that structure of reward and correction. That is exactly how a buyer shapes a supplier: reward the performance you want, penalize what you don’t, and over time the supplier’s behavior bends toward what you’ve incentivized. The mechanism is the same. An agent — human, corporate, or artificial — becomes what it is rewarded for becoming.
The mistake is not new. It is generational.
Here is the part worth sitting with.
This is not an article about AI. It is an article I could have written in 1998, in 2007, in 2015, and again now — changing only the noun.
Every technology generation arrives wearing the same costume: the thing that will finally deliver. And every generation, when initiatives fall short, the blame lands in the same place — on the technology itself. ERP failed. The SaaS rollout failed. The analytics program failed. The RPA initiative failed. The digital transformation failed. Now: the AI failed.
Sometimes the technology genuinely underperformed. Far more often, the ecosystem failed to govern, align, integrate, or absorb it — and the technology took the blame because the technology is the visible part. The agent is what you can point at. The ecosystem — the incentives, the dependencies, the decision rights, the seams between people and systems — is harder to see and harder to indict.
So the same post-mortem gets written every decade, with a new logo at the top. We blamed the tool. The tool was never the determinant.
The question that doesn’t change
The real question was never whether an AI agent can scale. It was never whether ERP could scale, or analytics, or RPA. It is the same question every time:
Can the ecosystem absorb what the agent makes possible?
That is the question we failed to ask of six technology generations, and it is the one being skipped again now — because AI, like every wave before it, is being introduced as a technology discussion instead of an ecosystem discussion.
I have been making this argument since a 1998 defence-sector engagement where the outcome was determined not by the technology introduced, but by the operating conditions it entered. The technology of that era is long obsolete. The argument is not — because it was never about the technology. It was about what the technology enters.
The agents keep changing. The ecosystem logic doesn’t.
The substrate isn’t more technology. It is how humans and AI agents align with the technology that already exists.
Truth Is Believing. Accuracy Is Knowing.
Jon Hansen is the creator of Implementation Physics™, a research-based framework developed over nearly three decades to explain why technology initiatives succeed or fail. Supported in part through Canada’s Scientific Research & Experimental Development (SR&ED) program, his work spans six technology generations—from ERP through Agentic AI—and examines the organizational conditions that determine outcomes regardless of the technology being deployed.
His research forms the foundation for the Hansen Method™, Hansen Fit Score™ (HFS), Phase 0™ Readiness Assessment, and ARA™/RAM 2025™ multimodel verification architecture.
-30-
Related
What do suppliers, couriers, buyers, and AI* agents have in common?
Posted on June 11, 2026
0
*And not just any AI — an agentic one. An AI tool that answers a question is easy to file under “technology.” But agentic AI doesn’t just answer; it decides, acts, and pursues goals across steps. The moment it acts autonomously, it stops behaving like a tool and starts behaving like an agent in the network — which is exactly why the ecosystem questions we ask of suppliers and couriers now apply to it too. In short, they are all agents in the ecosystem. None of them is the ecosystem.
It sounds obvious once you say it. But watch how differently we treat the first three and the fourth.
When a supplier underperforms, no one concludes that suppliers don’t work. They ask the ecosystem questions — the ones every procurement professional asks in their sleep. What authority did it have? What decisions did it touch? How was performance measured? What were the escalation paths? What happens when it’s wrong? A supplier influences outcomes. It does not determine them. The outcome comes from how the whole network — incentives, dependencies, decision rights, governance — absorbs what that supplier does.
We ask the same of a courier. The same of a consultant. The same, instinctively, of a new hire. We have spent centuries learning to govern human agents and decades learning to govern supplier and logistics ecosystems. The questions are second nature.
Then an AI agent enters, and the questions vanish.
Suddenly the conversation is about models, prompts, architecture, orchestration. The agent’s capabilities fill the whole frame, and the ecosystem questions — authority, accountability, performance, escalation, what happens when it’s wrong — quietly drop out. We stop asking how the system governs the agent and start asking only what the agent can do.
That is the entire mistake, in one move.
Where the analogy holds, and where it doesn’t
Let me be precise, because a careful reader will push here, and they should.
An AI agent is not a supplier in every respect. A supplier has its own objectives, economic interests, and judgment. It can negotiate, adapt strategically, withhold information, and decide whether to participate at all. An AI agent has none of that — no independent goals, no stake, no will. It produces outputs from its training and the task in front of it. The mechanics are genuinely different, and pretending otherwise would be sloppy.
But here is what survives the difference: the governance questions are identical. What authority does it have? What decisions does it influence? What happens when it’s wrong? How is its performance monitored? What are the escalation paths? Those questions do not disappear because the agent is software instead of a company. If anything, they matter more — because an agent with no judgment of its own will execute a misaligned instruction faster and more completely than any supplier ever would.
So the analogy isn’t “AI is a supplier.” It’s “AI is an agent in the ecosystem, and every agent in the ecosystem must be governed — regardless of what it’s made of.”
There’s a deeper parallel still. Training an AI is itself a buyer–supplier relationship in miniature. The developer rewards the system for what it gets right and steers it away from what it shouldn’t do — reinforcing some behaviors, discouraging others, until the agent’s tendencies are shaped by that structure of reward and correction. That is exactly how a buyer shapes a supplier: reward the performance you want, penalize what you don’t, and over time the supplier’s behavior bends toward what you’ve incentivized. The mechanism is the same. An agent — human, corporate, or artificial — becomes what it is rewarded for becoming.
The mistake is not new. It is generational.
Here is the part worth sitting with.
This is not an article about AI. It is an article I could have written in 1998, in 2007, in 2015, and again now — changing only the noun.
Every technology generation arrives wearing the same costume: the thing that will finally deliver. And every generation, when initiatives fall short, the blame lands in the same place — on the technology itself. ERP failed. The SaaS rollout failed. The analytics program failed. The RPA initiative failed. The digital transformation failed. Now: the AI failed.
Sometimes the technology genuinely underperformed. Far more often, the ecosystem failed to govern, align, integrate, or absorb it — and the technology took the blame because the technology is the visible part. The agent is what you can point at. The ecosystem — the incentives, the dependencies, the decision rights, the seams between people and systems — is harder to see and harder to indict.
So the same post-mortem gets written every decade, with a new logo at the top. We blamed the tool. The tool was never the determinant.
The question that doesn’t change
The real question was never whether an AI agent can scale. It was never whether ERP could scale, or analytics, or RPA. It is the same question every time:
Can the ecosystem absorb what the agent makes possible?
That is the question we failed to ask of six technology generations, and it is the one being skipped again now — because AI, like every wave before it, is being introduced as a technology discussion instead of an ecosystem discussion.
I have been making this argument since a 1998 defence-sector engagement where the outcome was determined not by the technology introduced, but by the operating conditions it entered. The technology of that era is long obsolete. The argument is not — because it was never about the technology. It was about what the technology enters.
The agents keep changing. The ecosystem logic doesn’t.
The substrate isn’t more technology. It is how humans and AI agents align with the technology that already exists.
Truth Is Believing. Accuracy Is Knowing.
Jon Hansen is the creator of Implementation Physics™, a research-based framework developed over nearly three decades to explain why technology initiatives succeed or fail. Supported in part through Canada’s Scientific Research & Experimental Development (SR&ED) program, his work spans six technology generations—from ERP through Agentic AI—and examines the organizational conditions that determine outcomes regardless of the technology being deployed.
His research forms the foundation for the Hansen Method™, Hansen Fit Score™ (HFS), Phase 0™ Readiness Assessment, and ARA™/RAM 2025™ multimodel verification architecture.
-30-
Share this:
Like this:
Related