What's so funny 'bout context, knowledge and understanding

Most organisations accept that people learn more through their work than from any training programme. The evidence for that is not really in dispute. The hard part has always been how to design for it. Wiring AI tools into institutional sources does not answer that on its own. What it does give you, for the first time, is a mechanism that can touch all three layers at once: environment, social practice, and the wiring itself. To see why, it helps to be clear about what those layers are.

We frame a learning ecosystem as three interdependent layers. They are not a hierarchy where one outranks another. They shape each other, so any layer can strengthen the others or quietly undercut them.

The environment layer

The Environment Layer is where what your culture actually rewards lives. It covers how feedback gets built into the work, what behaviour earns reward rather than just a mention in the values statement, how decisions get made and made visible, and how mistakes are handled when one happens. Organisations rarely design this layer on purpose, yet it does most of the teaching. If deadlines keep overriding quality, people learn to optimise for speed. If dissent carries a social cost, people stay quiet. None of this needs a training budget, and no programme can counter it on its own, because the environment goes on teaching its lessons long after the workshop ends.

The social layer

The Social Layer is where a team makes sense of things together. It runs on shared experiences that give people common reference points, on structures that let disagreement be productive instead of expensive, and on time that is deliberately protected for talking things through. Teams do not coordinate on everything each person privately knows. They coordinate on what they can reasonably assume the others recognise. Strip the social dimension out and learning gets efficient but shallow: behaviour shifts, judgement does not.

The technical layer

The Technical Layer is the plumbing that makes institutional knowledge easy to reach, and where it makes sense, ties learning evidence to performance. It covers wiring trusted sources into AI tools, organising that knowledge sensibly, and measuring honestly how people engage with the work. This layer serves the other two. It does not lead them. When organisations treat the wiring as mainly a productivity shortcut rather than part of the ecosystem, they get the usual AI result: individual tasks finish faster while the shared sensemaking gets thinner.

The neuroscience matters here, and L&D design rarely uses it. Immordino-Yang, Nasir, Cantor, and Yoshikawa (2022) describe a principle they call sociocultural embeddedness: development and learning happen in relation to the norms, values, and expectations around us across the many contexts we live in, even when we are working alone. Our work is woven through the work of other people, and as we develop we act on those contexts and change them in turn.

For the Environment Layer the implication is structural, not a question of motivation. Knowledge and skill do not sit stored inside a person like a resource waiting to be deployed. They are enacted in the moment, as the person adapts to the situation in front of them. What someone can understand and do at any given moment depends on both their prior experience and the current context, including its social, cognitive, and emotional sides. A person can reason through something complex in a setting where they feel capable and at ease, then fail to reach the same reasoning in a setting they read as unsafe or beside the point.

The same competency, in the same person, varies with context in ways a skills taxonomy never captures. So the Environment Layer is not the backdrop to the learning intervention. It largely decides whether any of the learning shows up in performance at all.

So the technical wiring only matters to the Environment Layer on a condition: the environment has to be set up so that surfaced knowledge feels useful rather than threatening. An organisation that punishes anyone for admitting uncertainty will get nothing from AI tools that put competing decision frameworks on the table. The environment has to already reward the kind of reasoning the Technical Layer is meant to support.

George Siemens' Connectivism framework (2005) argues that in distributed, network-rich environments, learning is less about what an individual stores and more about the quality of the connections they can move along. Knowledge sits spread across people, artefacts, databases, and tools. Competence is the ability to navigate those connections well: spotting which nodes are relevant, reaching them quickly, and putting them to use.

At the Social Layer, Connectivism explains why shared experiences are not inefficiencies to be optimised away. They are how the nodes in the network become legible to each other. Two people can have the same documentation in front of them and still be unable to reason together, because the network between them lacks the shared reference points that make disagreement productive. Working the same cases, arguing out the contested decisions, reflecting together afterwards: these are not social niceties. They are what holds an organisation's intelligence together.

At the Technical Layer, a server that indexes an organisation's decision histories, its frameworks, and the know-how that usually lives in people's heads makes the knowledge network reachable at the moment of need. When someone in their AI tool asks about a problem and gets back organisational knowledge that fits their situation, they are activating a node the organisation has deliberately maintained. How strong that connection is comes down to how well the knowledge was curated, what the Environment Layer rewards, and whether the Social Layer has kept it current and argued over.

Connectivism also shows the risk in wiring-only deployments: a read-only relationship with organisational knowledge never builds the capacity to add to it or revise it. The Technical Layer can serve learning, or it can erode the Social Layer by making individual access so frictionless that the collective work of sensemaking starts to feel optional. That fork is a design decision.

The three layers do not run in sequence. They behave like a complex adaptive system, each one feeding back and changing the others. Immordino-Yang's idea of emergent holism is useful for naming what that produces.

The principle of emergent holism holds that the subsystems of humans and their contexts function not additively but mutually constitutively, giving rise to emergent potentials that are partially unpredictable from the sum of their parts. In organisational terms: collaboration is not an individual trait that can be trained in isolation. It emerges from team dynamics. Psychological safety does not live inside a person. It exists between people. Systems thinking emerges when the environment rewards cross-boundary reasoning, not when someone completes a module about it.

Immordino-Yang et al.

Put that against the three-layer model. Wiring that surfaces good institutional knowledge (Technical Layer), in an environment where that knowledge gets discussed, contested, and updated (Social Layer), inside an organisation that rewards using and adding to institutional knowledge rather than hoarding it (Environment Layer), produces something none of the layers reach on their own. Capability grows in the interplay between the layers. Optimising any single one of them harder does not get you there.

The most useful thing a learning function does is not running the small slice of learning it delivers directly. It is shaping the much larger slice that happens every day through the conditions of work. Solid technical wiring gives that influence a channel that can run closer to real time. Whether an organisation uses it that way is a design question more than a technology one.