Core Thesis: Organizations often face a hidden breakdown when operational metrics report smooth sailing, but ground reality feels otherwise. This breakdown stems from an inability to identify what truly matters. This post delves into signification, the crucial process of transforming weak, unclassified signals into shared meaning. It also explains why aggressive reporting frameworks and AI compression pose a threat to genuine organizational learning.
The incidents each look ordinary. Taken together, something has shifted. A process that still meets its targets is producing worse results. Customers in unrelated categories are behaving the same strange way. A team hits every goal while growing less effective. Someone notices, but cannot yet name what they see. There is no metric, field, category, or agenda slot for it. The manager asks for evidence; the dashboard shows nothing. The next meeting starts in ten minutes. The observation is either forced into a familiar category or it disappears. This is where signification begins.
Signification is how something becomes a thing worth having a word for.
Something in the world changes. Initially, it’s just a variation. Someone notices it matters and distinguishes it from background noise. They relate it to past experiences and test it against new encounters, gradually stabilizing it to communicate. Once stable, it can be recognized again, seen by others, remembered by the organization, and eventually become a category, metric, policy, field, or dashboard line. Organizations depend on this process constantly, often starting later once the sign exists.
Three kinds of knowing
Modern work privileges different forms of knowledge. First, there’s procedural knowledge, which includes knowing how to act, such as which dashboard to open, which process to follow, who owns the service, which field to complete, how to escalate, and how to get something through the organization. Second, there’s situational knowledge, which includes understanding the current system state, what matters now, what’s changing, and the likely consequences. Third, there’s reflexive knowledge, which involves recognizing the observations, distinctions, categories, tools, assumptions, and interpretations that made the situation intelligible. This third capability allows the second to evolve.
Modern organizations heavily invest in the first, automating and instrumenting to achieve the second. The third often receives weaker support, which becomes crucial when existing representation fails to align with the changing world.
A person can become skilled in a system without understanding its underlying principles. Similarly, an organization can be proficient in using a dashboard and responding to metrics, but it may lack the ability to comprehend the reasons behind the metric’s significance, identify its exclusions, or determine if the distinction still warrants guiding action.
The first-order organization
Most organizations already operate based on feedback. For instance, if a service exceeds its latency objective, an engineer intervenes. If a team misses its target, management changes its activities. If costs rise beyond the plan, spending is constrained. If customer churn increases, retention work begins. If employee engagement falls, an organizational response follows. The general pattern is simple: observe → compare → correct → observe.
Cybernetics gave us a language for this. Chris Argyris and Donald Schön later described a related distinction through single-loop and double-loop learning. Single-loop learning adjusts action against established governing variables. Double-loop learning allows those variables themselves to be reconsidered. Heinz von Foerster made a related move in second-order cybernetics by including the observer and the act of observation within the system being examined.
Organizations are filled with first-order loops: telemetry for engineering, usage metrics for products, pipelines for sales, budgets for finance, surveys for HR, and dashboards, forecasts, and objectives for executives. Each level measures, compares, corrects, and measures again. Reflective practices like retrospectives, postmortems, after-action reviews, strategy sessions, research, and intelligence analysis aim to learn from experience, but their effectiveness depends on the framing of their questions.
A postmortem investigation aims to comprehend the underlying reasons behind a deployment failure. However, a more important question is: why was the situation initially perceived as a deployment failure? This question transcends the immediate incident and delves into the category through which the event entered organizational understanding.
Hierarchy doesn’t inherently lead to higher-order observations.
An executive responding to a revenue indicator shares a similar observational relationship with an operator responding to a CPU threshold: detect deviation, compare against expectation, intervene, and observe again.
The altitude changes, but the observation form remains stable.
The compression architecture
Information has to move between organizational levels. Operations report to teams. Teams report to functions. Functions report to business units. Business units report upward again. Every boundary requires compression.
The operational system can contain thousands of observations, but the team lead only needs dozens, the director needs a handful, and the executive dashboard may require only three statuses and a forecast.
Compression enables coordination and also performs semantic processing.
Someone decides what matters. Someone chooses which distinction survives. Someone removes uncertainty. Someone transforms a complicated situation into a status that another level can consume.
At every boundary, information is transformed into the vocabulary and format that the receiving system understands.
The translators understand the next level’s expectations, including which numbers provoke questions, which categories are accepted, and what constitutes a satisfactory explanation. Reporting is shaped by the receiving system.
Information propagation is not learning propagation.
Information dissemination occurs, while the organizational categories used to interpret that information remain static. This is where the significance of tools becomes evident. Every organizational system comprises an ontology.
A ticketing system defines issue types. A CRM defines customers, opportunities, stages, and risks. A monitoring platform defines metrics and thresholds. A project system defines tasks, dependencies, milestones, and completion. An HR platform defines roles, competencies, performance, and engagement. Bowker and Star showed how classification systems become infrastructure: once embedded deeply enough, their categories cease to feel like choices and begin to feel like the world itself. What fits the schema becomes easy to express. What does not fit becomes free text, an attachment, an exception, an anecdote, or something that never enters the system.
Policy performs a similar operation. When X occurs, classify it as Y and perform Z. This is accumulated learning in compressed form. A reflective organization also needs a way to discover when X has stopped signifying Y.
Why compression wins
Compelling economic reasons drive this architectural design.
Compression reduces cognitive load, streamlines processes, and provides tangible outputs. A policy mitigates uncertainty, while a target establishes performance benchmarks. A report fosters accountability.
Reflection, while highly valuable, has a less favorable economic profile. Its costs are incurred immediately, as someone spends time examining something uncertain. There may be no immediate tangible outcome, and the observation might prove insignificant. But, if the work eventually leads to a useful distinction, the benefits may manifest months later and have a widespread impact across the organization. Attribution is weak.
James March described the organizational tension between exploiting existing knowledge and exploring potential new knowledge. Modern management systems heavily favor exploitation, while exploration remains more challenging to quantify, schedule, reward, and defend. Politics further reinforces this tendency.
Ambiguity is especially expensive in low-trust environments. A provisional observation exposes the observer because it contains judgement without the protection of an accepted metric, policy, or category. Leaders often ask for numbers because numbers provide defensibility, apparent objectivity, and a basis for action.
People adapt to navigating local political uncertainty. Weak signals are quickly categorized, turning tentative interpretations into confident statuses. Quantification shields organizations from gaining fluency but hinders maintaining open-ended meanings long enough for maturation. Information overload increases pressure, leading to more observations. This results in more metrics, telemetry, dashboards, analytics, surveys, reports, and documentation, creating a higher demand for compression that reinforces existing categories.
What gets lost
Let’s return to the person who noticed the incidents.
They’re in the early stages of organizational learning. They need to encounter the phenomenon again, recall previous cases, and compare observations with someone who sees the situation differently. Their first explanation may fail. The important distinction may only emerge after several attempts to describe it.
Signification requires sustained attention, memory, ambiguity, dialogue, and permission to revise.
Modern work compresses these conditions, leading to fragmented attention, shortened response times, and meetings optimized for decisions. Reports demand conclusions, performance systems reward visible output, and tools require predefined categories. Management demands evidence before the observer fully understands its significance. Emerging observations are pushed toward certainty while their meaning is still evolving.
When people repeatedly notice things that matter and discover that the organization has no place for those observations, their own work can start to feel less meaningful. Judgement contracts into reporting. Experience is reduced to data entry. Seeing something new matters only after it can be translated into something already known.
The loss of signification becomes a loss of agency.
AI and epistemic smoothing
Generative AI now performs tasks like summarizing documents, classifying incidents, correlating observations, explaining, drafting reports, and compressing organizational information. A hundred reports can be reduced to ten themes, ten themes to three conclusions, and three conclusions to an executive summary.
AI also introduces another form of compression: epistemic smoothing.
Human accounts often exhibit hesitation, disagreement, incomplete explanations, awkward qualifications, uneven confidence, and unresolved contradictions. These features can convey valuable information and indicate the instability of an interpretation. Generative synthesis tends to smooth out these inconsistencies, transforming contradictions into themes, hesitations into assertions, minority interpretations into consensus language, and caveats into smaller details. As a result, an unresolved situation may appear more settled than its underlying structure actually is.
Those traces of uncertainty often mark the places where second-order inquiry should begin.
Once ambiguity has been transformed into a clear explanation, subsequent observers receive fewer cues that warrant further scrutiny of the interpretation itself. Epistemic smoothing is not an inherent property of AI. AI can maintain disagreement, reveal confidence, retain minority interpretations, cluster anomalies, and keep unclassified material visible. The challenge lies in integrating AI into the architecture.
Unfortunately, the vast majority of modern organizations reward concise, fluent, and actionable compression, which AI makes more cost-effective. If designed to preserve provisionality, AI could become the infrastructure for signification.
In such an environment, the most significant human contribution might be the ability to articulate: “Our current vocabulary is insufficient to adequately describe the situation.” This sentence signifies the onset of a new distinction.
Give the unclassified somewhere to live
A diagnosis of this magnitude doesn’t necessitate a massive program.
A small, focused initiative can serve as a starting point. Designate a specific area within the organization to accommodate an observation that doesn’t currently fit into existing categories. During a retrospective review, management meeting, engineering discussion, customer session, or strategy discussion, pose the question: “What are we observing that our current categories fail to adequately explain?”
Give the answer provisional status. Do not require an owner immediately. Do not require a metric. Do not demand a business case or corrective action. Let the observation live long enough for other evidence to accumulate around it.
Some provisional signs will fade out. That is the process working. Others will stabilize. They may eventually become concepts, measures, policies, categories, or new ways of seeing the system.
The important intervention is temporal: the organization gives emerging meaning enough time to survive.
A design rule can be applied to summaries. Instead of automatically resolving disagreements, preserve them. Maintain a visible level of confidence. Allow unresolved categories to remain unresolved. This is a wedge, not a transformation framework. Its purpose is straightforward: preserve enough ambiguity to initiate second-order learning.
The organization that can reconsider itself
Organizations usually describe adaptability through speed: faster feedback, faster decisions, faster deployment, faster analysis, faster correction. Adaptive capacity also depends on whether experience can alter the system through which experience is interpreted. That requires feedback about feedback. It requires categories that can reopen. It requires traces of uncertainty that survive compression. It requires somewhere for a weak distinction to remain visible before it becomes legible to the existing system. And it requires signification: the process through which something not yet understood becomes meaningful enough to recognize, communicate, remember, and eventually reconsider again.
Modern organizations have constructed vast systems for observation. Every level of the organization engages in measurement, recording, reporting, evaluation, and response. Information flows upward, while instructions flow downward. AI will further accelerate this flow. However, the sheer volume of observation does not necessarily equate to reflective capacity. A thousand first-order loops alone do not constitute a second-order organization.
Within the existing system, there must still be a space for individuals to recognize that the current signs are no longer appropriate. Unfortunately, modern work practices are gradually diminishing this space and the practice within it.