Signal Index · Vol. 4 · May 2026

The Execution Signal

Vol. 3 named the stack. The stack doesn't deploy itself. The gap between architecture and running agents is where most organizations stall — and the reason has a name: the Knowledge Distance Problem.

Reggie Britt · signal4i.ai · H  O  T

The gap nobody named

Architecture Is Not Deployment

Vol. 3 closed with a precise statement: the IBM i agentic stack exists, all four layers are available today, and what's missing is not technology. What's missing is the posture — the organizational decision to treat this moment as the moment it is.

That framing was right as far as it went. But it left the hardest question unanswered: what actually happens in the space between naming the architecture and running agents in production?

That space has a structure. It is not random. It is not simply "organizational friction" or "change management." It is a specific, measurable gap — and Harvard and Stanford just ran the experiment that named it precisely.

The variable that determines whether AI deployment succeeds or fails at the execution layer is not the model, the prompt, or the compute. It is the distance between the people doing the work and the domain the AI is being asked to operate in.

This is the Knowledge Distance Problem. KD is the binding constraint at the execution layer. It is why 94% of organizations have adopted AI and 6% have bottom-line impact. It is why the architecture gets named and the agents don't get deployed. And for IBM i practitioners, it is the specific variable that places them structurally ahead of nearly every other enterprise technology group — before they've written a single line of agent code.

The Harvard/Stanford finding

The GenAI Wall — Measured Precisely

Researchers from Harvard Business School, Stanford University, and the Stanford Digital Economy Lab ran a controlled field experiment inside a major professional services firm. Three groups — domain insiders, adjacent professionals, and distant outsiders — were given identical tasks with identical AI tools.

0

Performance gap between groups on conceptualization tasks. For structuring, outlining, and ideating, AI equalized performance completely. Every group performed at specialist level. The gap disappeared entirely.

Harvard / Stanford / Stanford Digital Economy Lab · 2026

Then the tasks shifted to execution — writing, building, producing. The results split sharply. Domain insiders and adjacent professionals matched. Distant outsiders consistently underperformed. In many cases they made outputs measurably worse than if they had worked without AI at all.

The researchers named this the GenAI Wall Effect: a threshold defined not by the technology but by the human's distance from the execution domain, beyond which AI cannot bridge the gap.

The mechanism
Why Distance Degrades Output
The distant outsider couldn't evaluate the AI's output because he had no foundation for judging what good looked like. So he removed what he didn't recognize as valuable and replaced it with what felt right to him. He degraded a correct answer because he couldn't see that it was correct. The AI was fine. The knowledge distance was the problem.

This is the execution gap. Not a training deficit. Not a tooling deficit. A proximity deficit. The people closest to the domain can execute with AI. The people furthest from it cannot — and their attempts actively degrade output quality.

What KD looks like in practice

Six Root Causes That Collapse to One

Every standard explanation for why AI deployments fail at the execution layer is a symptom. The root cause is always Knowledge Distance.

Symptom 01 — Inadequate tools
Misdiagnosis at the execution layer
Organizations reach for bigger models or different frameworks when the actual problem is KD. Buying compute to solve a proximity problem doesn't work. The tool gap is a misdiagnosis.
Symptom 02 — High costs
Rework from failed evaluation loops
The actual cost driver is rework — outputs produced by AI that couldn't be validated because the reviewers lacked the domain knowledge to judge them. KD makes evaluation impossible. Evaluation failure makes rework inevitable.
Symptom 03 — Confidence gap
What KD feels like from the inside
Teams don't trust the outputs because they can't evaluate them. The confidence gap isn't about AI capability — it's about whether the human in the loop has the proximity to know if the output is right. Confidence without proximity is dangerous. Skepticism without proximity is paralysis.
Symptom 04 — Data complexity
Invisible without proximity
Data problems that derail AI deployments are only visible to people who understand the domain the data describes. You can't govern what you can't evaluate. KD makes data governance impossible at the execution layer.
94%

of GenAI pilots fail to reach production scale. KD is the root cause operating underneath every reported reason — skills, costs, tools, complexity. Each is a KD gap with a different label on it.

Signal Stack · multiple sources · 2025–2026
The agentic amplifier

KD at Autonomous Scale

KD was always present in AI deployments. Agentic AI makes it dangerous at scale. The distinction is precise.

When AI generates a document or a summary, a human reviews it before anything happens. The KD gap shows up as review time, rework, or a poor output that gets corrected. Manageable.

When AI agents execute decisions autonomously — pricing, routing, approvals, workflow orchestration — the KD gap shows up as compounding errors at autonomous scale, without a human in the evaluation loop to catch the drift.

The governance signal

AI-related security incidents are up 55% year-over-year. The Foundation Model Transparency Index dropped from 58 to 40. The incidents aren't technology failures. They are KD failures operating at agentic speed — systems executing in domains they don't understand, without humans close enough to the domain to catch the drift.

This is the governance gap. And it is the gap Vol. 5 proves cannot be closed sequentially — technology transformation, human fluency, and organizational redesign must run simultaneously, because each one is a precondition for the others to hold at agentic scale.

The IBM i position

Structurally on the Right Side

The Harvard/Stanford finding has a specific application for IBM i practitioners that most are underestimating.

IBM i organizations hold something most enterprises are desperately trying to acquire: 30 to 40 years of encoded domain expertise in production systems. The business logic in RPG programs, the data governance baked into Db2, the workflow intelligence in the job stream and subsystem architecture — all of it is proximity, encoded.

That is the specific variable the experiment identified as the determinant of AI output quality at the execution layer.

Data proximity
40 years of transactional data with referential integrity
Domain-encoded in Db2. The model doesn't need to be trained on what the data means — the practitioner already knows. The agentic layer inherits that proximity through the connection layer.
Process proximity
Business logic as encoded institutional knowledge
The RPG and CL programs are the domain expertise the agentic layer needs to make good decisions. It already exists. It needs to be surfaced and connected via MCP, Mapepire, and the IBM i AI SDK — not rebuilt from scratch.
Platform proximity
Native analogs to agentic execution patterns
IBM i's event-driven architecture, job stream model, and object-level governance are native analogs to the agentic execution patterns enterprises are building from scratch on commodity infrastructure. The distance to agentic is shorter — because the platform anticipated the pattern.
IBM i practitioners don't need to acquire domain knowledge. They are the domain knowledge. The Knowledge Distance Problem — the binding constraint at the AI execution layer — is the specific gap they close by definition.

The execution signal is not that IBM i is "ready for AI." It is that the people running IBM i are structurally positioned on the right side of the variable that determines whether AI deployments succeed or fail — before anyone else in their industry recognized it as a variable.

Vol. 5 picks up from here: why the three tracks required to move from this position to the Human Agentic state — technology, human, organizational — cannot run sequentially, and what happens to organizations that try.

← Vol. 3 · The Architecture Signal
Signal Index · Vol. 1–8 arc
Full archive →