The wall has a location now.

The signals reshaping how organizations deploy AI arrive from outside the room — from the labs, the agentic frontier, the regulators, the markets. Each week I pull a handful from the Signal Stack, sourced and cross-validated, and translate them into what they mean for the people running the systems that matter.

For three years the story has been that some organizations are "ahead" on AI and some are "behind." This week's signals dismantle that frame. The thing that actually separates the organizations getting value from AI from the ones stalling isn't speed of adoption, budget, or tooling. It's something more specific — and for once, it's been measured precisely.

Four signals.

1. Harvard and Stanford found the wall — and put a number on it.

Researchers ran a controlled field experiment inside a major professional-services firm. Three groups — domain insiders, adjacent professionals, and distant outsiders — got identical tasks with identical AI tools. On conceptualization(structuring, outlining, ideating), AI equalized everyone: the performance gap between groups disappeared completely. Then the tasks shifted to execution — writing, building, producing — and the results split sharply. Insiders and adjacent professionals held. Distant outsiders consistently underperformed, in many cases producing output measurably worse than if they'd used no AI at all.

→ The mechanism is the part worth sitting with. 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. 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 wall — and it's defined by the human's distance from the execution domain, not by the technology.

Source: Harvard Business School / Stanford / Stanford Digital Economy Lab controlled experiment

2. Why this is the best news an experienced practitioner has had in years.

Read that finding again from the seat of someone with thirty years in a domain. AI equalizes the conceptualization layer — anyone can now outline, draft, ideate at a high level. But execution quality concentrates with the people closest to the actual work, because they're the only ones who can tell whether the AI's output is right. Deep domain expertise isn't threatened by AI equalizing the easy part. It becomes more valuable, because AI generates more output that requires expert judgment to validate — and validation is exactly what proximity buys.

→ For platform practitioners specifically, this inverts the "you're behind" narrative. Thirty years of depth is maximum proximity to the execution layer, which the research confirms is the precise variable that determines whether AI output gets elevated or degraded. The skills gap that's haunted these platforms for a decade is real — but it cuts the opposite way from how it's usually told. The scarce thing in the AI era is the deep practitioner, and that's exactly what these shops have.

Source: same study, read through the practitioner lens · connects to Signal Stack Knowledge Distance cluster

3. The six reasons AI deployments fail all collapse into one.

Every standard explanation for why AI stalls at the execution layer — inadequate tools, high costs, poor data, change resistance, skills gaps, unclear ROI — turns out to be a symptom. Organizations reach for bigger models or different frameworks when the actual problem is knowledge distance: the people directing the AI are too far from the domain to judge its output. Buying compute to solve a proximity problem doesn't work. It's a misdiagnosis, repeated at scale — which is why 88% of organizations have adopted AI and a single-digit percentage report real operational impact.

→ This reframes what a readiness investment actually is. It's not more tooling. It's closing the distance between the people doing the work and the domain the AI operates in — which means surfacing tacit knowledge, keeping domain experts in the loop at the execution layer, and refusing to hand judgment-heavy work to people who can't evaluate the output. The fix is organizational, not technical.

Source: Knowledge Distance Problem framework · synthesizing the failure-mode literature

4. Meanwhile, the capability ceiling jumped again this week.

Microsoft took an agent that runs complex, multi-step work end-to-end to general availability worldwide — default-distributed inside the productivity suite people already open every morning, priced per task. More than half the Fortune 500 tested it in preview.

→ Hold this against the first three signals and the tension is the whole point. Agentic execution is now everywhere, by default, at the exact layer where knowledge distance is widest — the everyday business user, clicking through approval prompts on work they may be too far from the domain to evaluate. The capability is arriving faster than the proximity to govern it. That's the wall and the wave hitting at the same time.

Source: Microsoft 365 Blog · agentic execution reaches general availability

The pattern

The frame that says some organizations are "ahead" and some are "behind" on AI is the wrong frame. What the research actually shows is that value concentrates where knowledge distance is shortest — with the people closest to the work. The organizations that win won't be the ones that adopted fastest. They'll be the ones that kept their deepest practitioners at the execution layer while the capability scaled around them.

For organizations with decades of domain depth, that's not a disadvantage to overcome. It's the asset the whole AI era is about to make scarce.

That's what this is for. See you next week.

— Reggie Britt

The full signal record — 531 signals, 20 categories, sourced and cross-validated — is public at signal4i.ai. Browse it, draw your own conclusions.