FIELD NOTES
Insights from the field.
Discovery, research and observations from organizational practice.
ANCHOR RESEARCH
CENTRAL ARGUMENT
What it actually takes for AI to become organizational practice
Training can increase individual AI use, but organizational practice only emerges when workflows are redesigned, structural blockers are removed, and verification is built into how work gets done.
The work is not simply encouraging adoption. It is reshaping the conditions around it: processes, access, and repeatable ways of working that others can build on. That is what has to change for AI to move from individual experimentation to embedded, operational practice.
FIELD OBSERVATIONS
The same workflow gets rebuilt, independently, across teams
01
The same workflow often appears in multiple parts of a team or an organization without anyone realizing it. One person builds a prompt, another creates a workaround, a third solves the same problem from scratch, and none of it becomes shared practice.
What discovery work consistently surfaces is not a shortage of good ideas, but the same good ideas built in parallel, invisibly, with no mechanism for transfer.
share workflows across teams
7%
The gap between novice and expert is not prompting skill. It is problem framing​
02
Advanced users do not necessarily have better prompts. They have a different mental model of what makes an interaction with AI successful.Where a novice submits one request and forms a conclusion from the output, an expert breaks the task into bounded steps, iterates across multiple turns, and builds toward a result. The gap is less about prompting skill in the abstract than about task decomposition, knowing how to hand AI a problem in pieces rather than all at once.
interviews, one pattern
90
When the real AI constraint is data infrastructure
03
The limiting factor is often not the model but the conditions around it: missing integrations, broken handoffs, processes that depend on a single person, systems that do not let information move cleanly across a workflow. In one team, five people were independently transferring the same data by hand every week, none of them aware the others had the same problem.
The solution was not an AI build. It was a structural fix that had to happen first.
people doing the same manual data transfer independently
5
Technical fluency predicts adoption. Seniority, tenure or age do not.
04
Across one organization, technical fluency was the clearest predictor of AI adoption, with a correlation of 0.71. The people who moved first were not necessarily the most senior, the youngest, or the longest-tenured. They were the ones most comfortable experimenting, iterating, and recovering when the tool behaved imperfectly.
That changes where to start. Rather than designing AI programs around hierarchy, a better first move is to identify who already has the confidence and working habits to build momentum.
fluency adoption correlation
0.71
Clear guidelines make people bolder, not more restricted
05
In the absence of AI policy, we saw people not experiment more freely. They invented their own rules, and those rules were consistently more conservative than any reasonable policy would require. Across teams handling high-stakes work, the pattern was the same: without clarity on what was permitted, people defaulted to doing less.
The organizations that moved fastest were not the ones that loosened governance. They were the ones that defined it early.
3-zone use police increases AI confidence
3
Why the middle 60% is slow to adopt
06
Early adopters find their own way to the tools. The middle group rarely does without deliberate help. That is not a failure of interest. It is a predictable change problem.
What reaches this group is not more information or more optional training. It is a structured set of conditions: visible proof from peers they trust, reinforcement through managers and formal expectations, AI embedded into routines they already follow, and practical support that makes new behavior easier to repeat.
Without those conditions, individual use can spread, but shared practice rarely does.
requires structured conditions to move
60%
Agent readiness starts before agents
07
The teams with the clearest path to automation shared one characteristic: the work was already structured into repeatable steps, with defined inputs, outputs, and human checkpoints. Where teams were not ready, the problem was usually not the technology. The workflow itself had never been made explicit.
What determines agent readiness is not technical sophistication alone. It is whether the work has been made legible enough for someone else, human or machine, to follow.
characteristic
1
DESIGNING CHANGE
How change takes hold
The middle 60% of any organization does not move through training alone. They move when AI is visibly working in their specific role, when the work around them has been restructured to make it the easier path, and when peers they trust are already using it. That is a design problem, not a learning problem.
The four conditions that shift adoption are culture, conviction, capability, and structure. Training addresses one. This work addresses all four.
ABOUT THESE INSIGHTS
These patterns emerged from direct work inside organizations, across functions and levels of seniority. They reflect what surfaced repeatedly in practice, not isolated observation.