You are never in one context
Right now, as you read this, you are simultaneously a professional with deadlines, a person with relationships that need tending, a body that may be hungry or tired, a mind holding at least two unresolved decisions, and a learner trying to absorb new material. Each of these is a context — a frame with its own logic, its own expectations, and its own demands on your attention.
You are not in one of these contexts. You are in all of them, all the time.
This is not a bug in human cognition. It is the default operating condition. The sociologist Bruce Biddle, in his landmark 1986 review of role theory, documented that every person simultaneously occupies multiple social positions — employee, parent, friend, citizen, caretaker — each carrying distinct behavioral expectations that frequently conflict with one another (Biddle, 1986). Role conflict is not an exception. It is the structural reality of being a social creature with obligations in more than one direction.
The problem is not that you hold multiple contexts. The problem is that you rarely identify which one is primary. And without that identification, every context gets a diluted version of your attention — enough to generate anxiety about each, not enough to produce real work in any.
What a context actually is
Before you can stack contexts, you need a precise definition. A context is not a task. It is a frame — a set of assumptions, rules, and interpretive lenses that determine how you process information and what counts as relevant.
Erving Goffman formalized this in Frame Analysis (1974), defining frames as "definitions of the situation built up in accordance with the principles of organization which govern events — at least social ones — and our subjective involvement in them." When you walk into a job interview, you activate a frame: certain behaviors are appropriate, certain information is relevant, certain emotional registers are expected. When you walk into a friend's kitchen, you activate a different frame. The information hasn't changed. You haven't changed. The organizing principles have changed.
Context stacking is what happens when multiple frames are active simultaneously. You are in the job-interview frame (performance, evaluation, impression management) while also in the parent frame (your child is home sick and you are tracking updates on your phone) while also in the financial frame (this job pays 40% more than your current one and your savings are thin). Each frame has its own logic. Each frame wants to be primary. And Goffman identified what happens when frames collide: a frame break — the occurrence of competing framing cues that reduce or eliminate the power of any single frame to organize your attention and behavior.
Frame breaks are not theoretical. They are the lived experience of trying to be fully present in a meeting while your mind runs a subroutine about a difficult conversation you need to have tonight. Neither frame wins. Both degrade.
The cognitive cost of stacked contexts
The degradation is not subjective. It is measurable, and the numbers are worse than most people expect.
Sophie Leroy's 2009 research on attention residue demonstrated the mechanism precisely. When you switch from Task A to Task B, your attention does not cleanly transfer. Cognitive activity about Task A persists — Leroy calls this "residue" — and that residue measurably impairs performance on Task B. Participants who switched tasks mid-stream performed significantly worse than those who completed their first task before moving on. The work produced under attention residue contained more errors, less depth, and required more revision (Leroy, 2009).
But context stacking is worse than task switching, because you are not switching between tasks. You are holding multiple frames simultaneously. Each frame is generating its own stream of relevance assessments, emotional responses, and behavioral expectations. Each one leaves residue on every other one.
John Sweller's Cognitive Load Theory explains why this is catastrophic for performance. Sweller identified that intrinsic load (the actual complexity of what you're doing), extraneous load (noise and irrelevant processing), and germane load (the productive thinking that builds understanding) all draw from the same limited pool of working memory — roughly 3 to 5 slots, per Cowan's (2001) research. When multiple contexts are loaded, they generate massive extraneous load. Your working memory fills with frame-maintenance operations — tracking which context applies right now, suppressing inappropriate responses from other frames, monitoring for signals that require a frame switch — and there is nothing left for the germane processing that the primary task requires (Sweller et al., 2011).
The practical consequence: research compiled by the American Psychological Association shows that task switching can reduce productivity by up to 40%, and Gloria Mark's studies found it takes an average of 23 minutes and 15 seconds to fully regain focus after an interruption. For software developers working with complex code abstractions, the residue can persist for 30 to 60 minutes. Interrupted tasks take twice as long and contain twice as many errors as uninterrupted tasks. This is not a personality flaw. It is how human cognitive architecture responds to stacked contexts.
Intersectionality: when contexts compound rather than add
The cost of context stacking is not always linear. Sometimes stacked contexts interact in ways that produce emergent effects — effects that cannot be predicted or understood by examining any single context in isolation.
Kimberle Crenshaw's concept of intersectionality (1989), while developed in the domain of legal theory and discrimination, reveals a structural principle that applies far beyond its origin. Crenshaw demonstrated that the experience of Black women could not be understood as "racism plus sexism." The combination produced a qualitatively different form of disadvantage — one that was invisible when each axis was examined separately. Antidiscrimination law that addressed race and gender independently failed to address the specific discrimination that occurred at their intersection (Crenshaw, 1989).
The epistemic principle is this: stacked contexts do not merely add. They compound. When you are simultaneously in a deadline-pressure context and a relationship-conflict context, the compound effect is not "deadline stress plus relationship stress." The deadline pressure makes you interpret the relationship conflict as an obstacle to productivity. The relationship conflict makes the deadline feel like evidence that your priorities are wrong. Each context reframes the other, producing a meaning-structure that neither context generates alone.
This is why simply listing your active contexts is not enough. You need to examine how they interact — which ones are amplifying each other, which ones are reframing each other, and which compound effects are operating below your awareness. The person who says "I'm stressed about work and stressed about my marriage" is usually experiencing a third, unnamed stress: the compound effect of two contexts that are each making the other worse.
The primary context protocol
Context stacking is inevitable. The question is not how to eliminate it — you cannot live a single-context life. The question is how to manage it so that one context gets your full attention while the others are explicitly parked.
Step 1: Name every active context. Not tasks. Contexts. "Employee under performance review" is a context. "Parent of a child who is struggling in school" is a context. "Person with a health concern waiting for test results" is a context. These are frames, not to-do items. Write them down. You cannot manage what you haven't named.
Step 2: Identify compound effects. Look at your list and ask: which of these contexts are interacting? Is the health concern making the performance review feel more threatening? Is the parenting context making you resentful of work demands? Name the compound. Write it down. Unnamed compounds run as background processes that consume working memory without producing insight.
Step 3: Designate the primary context. For the next defined period — one hour, one meeting, one deep work block — choose one context as primary. This is the frame through which you will interpret information, make decisions, and allocate attention. Everything else is secondary. Secondary does not mean unimportant. It means "not now."
Step 4: Park the secondary contexts. Each non-primary context needs a specific destination and a specific time. "I will address the school situation at 3pm when I have the counselor's number in front of me." "I will think about the architecture decision during my morning review tomorrow." Masicampo and Baumeister (2011) demonstrated that the Zeigarnik effect — the intrusive cognitive pull of incomplete concerns — is eliminated not by completing the concern but by making a specific plan for when you will address it. The plan must be concrete and the system must be trusted.
Step 5: Protect the primary context. Once designated, the primary context needs boundaries. Close the tabs that belong to other contexts. Silence the notifications from other frames. If you are in a deep work context, the Slack messages from your team-management context can wait. This is not neglect. It is the only way to produce work that is not degraded by residue from every other context you are holding.
How AI handles context stacking — and what it reveals about you
Large language models face a version of this same problem, and their solution is instructive.
Every LLM interaction involves stacked contexts: a system prompt defining the model's role and constraints, a conversation history establishing what has been discussed, retrieved documents providing reference material, and the user's current query. All of these compete for the same finite resource — the context window. As Anthropic's research on context engineering notes, "system prompts, policies, tools, and examples all consume tokens, and if you add more guardrails or examples, you shrink space for user inputs or retrieved evidence."
The parallel to human cognition is direct. Your system prompt is the role you are playing. Your conversation history is the emotional and relational background you carry into every interaction. Your retrieved documents are the relevant facts and concerns loaded into working memory. Your current query is the actual task in front of you. And all of these compete for the same limited capacity.
LLMs exhibit a phenomenon researchers call the "lost in the middle" effect — when the context window is full, the model retrieves best from the beginning and end of its input, and degrades for information buried in the middle. Humans show the same pattern. When your contexts are stacked, you attend best to the most recent demand (recency) and the most emotionally charged demand (salience), and everything in the middle — often including the thing that actually matters most — gets lost.
The AI solution is context engineering: deliberately curating what enters the context window, compressing or removing what is no longer relevant, and ensuring that the highest-signal information occupies the positions where the model attends best. The human equivalent is the protocol above: naming your contexts, identifying which is primary, parking the rest, and protecting the primary context's access to your full cognitive capacity.
There is one critical difference. An LLM processes its entire context window on every inference pass. It cannot choose to attend to one part and ignore another — everything in the window influences the output. You can. You have the ability to consciously designate a primary frame and suppress the others. This is a capability that AI does not have and that most humans never exercise.
The discipline of one context at a time
Context stacking feels sophisticated. Holding many concerns simultaneously feels like the mark of a serious person with a complex life. But the research is unambiguous: cognitive performance degrades in every context when multiple contexts compete for the same limited resources. You are not managing four contexts well. You are managing four contexts poorly and calling it responsibility.
The discipline is not in how many contexts you can hold. The discipline is in how clearly you can identify which one is primary, how decisively you can park the others, and how fully you can give yourself to the one context that matters right now.
L-0175 taught you to provide context when communicating — to give your audience the frame they need to interpret your message. This lesson turns that principle inward. You are the audience. Your competing contexts are the messages. And you need to decide which frame to interpret through before you can do anything well.
The next lesson, L-0177, asks what happens when context is not stacked but stripped away entirely. When information gets separated from the frame that gave it meaning, it does not become neutral. It becomes ambiguous, misleading, or dangerous. Context stacking is the problem of too many frames. Context loss is the problem of no frame at all.
References
- Biddle, B. J. (1986). Recent developments in role theory. Annual Review of Sociology, 12, 67-92.
- Goffman, E. (1974). Frame Analysis: An Essay on the Organization of Experience. Harvard University Press.
- Leroy, S. (2009). Why is it so hard to do my work? The challenge of attention residue when switching between work tasks. Organizational Behavior and Human Decision Processes, 109(2), 168-181.
- Sweller, J., Ayres, P., & Kalyuga, S. (2011). Cognitive Load Theory. Springer.
- Crenshaw, K. (1989). Demarginalizing the intersection of race and sex: A Black feminist critique of anti-discrimination doctrine, feminist theory and antiracist politics. University of Chicago Legal Forum, 1989(1), 139-167.
- Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87-114.
- Masicampo, E. J., & Baumeister, R. F. (2011). Consider it done! Plan making can eliminate the cognitive effects of unfulfilled goals. Journal of Personality and Social Psychology, 101(4), 667-683.