Quality of Chunks
There is a fact about working memory that Miller's 1956 paper established but that subsequent decades of citation have systematically underemphasized. The seven-item limit describes the quantity of chunks working memory can hold. It says nothing about the quality. A slot does not care what is placed inside it; it will hold a single digit or an entire theory of thermodynamics with equal ease, provided the theory has been sufficiently compressed into a single retrievable chunk. The slot is a container. The container has a fixed number. The contents have no fixed size. Consider two developers, each holding seven chunks in working memory as they design a system. Developer A's chunks contain the user requirement, the programming language syntax, the database connection pattern, the authentication library API, the deployment configuration, the testing framework, and the CI/CD pipeline. Developer B's chunks contain the user's actual problem (as distinct from the stated requirement), the system's failure modes and consequences, the architectural trade-offs between consistency and availability, the privacy implications of the data model, the maintenance burden, the organizational context, and the ethical implications of potential misuse. Both have seven chunks. Both are at capacity. Both will produce software. The software will be categorically different.
In the AI Story
The quality difference is not a matter of intelligence. Developer A is not less smart than Developer B. The difference is in what their seven slots contain, and what the slots contain depends on what the environment demands, what the tool handles, and what recoding the developer has completed. Pre-AI software development forced developers to fill slots with implementation concerns because those concerns could not be delegated. Large language models delegate implementation, freeing slots for higher-quality contents — if the developer has higher-quality contents to fill them with.
Quality is relational. The database connection pattern that is a low-quality chunk for an architect designing a system's high-level structure is a high-quality chunk for a developer debugging a connection timeout. Quality depends on context. This means that the same compression event — AI handling implementation — can simultaneously improve chunk quality for some roles and degrade it for others. The architect gains; implementation chunks were distracting her. The junior developer may lose; those same chunks were the learning materials she had not yet finished recoding into architectural understanding.
The contextual nature of chunk quality explains why studies of AI-assisted coding show enormous variance in productivity gains. The variance is not explained by prior skill, tool familiarity, or domain. It is explained by the quality of the chunks developers already possess — and therefore by what fills the slots compression frees. A developer whose pre-AI slots were consumed by syntax gains enormously when those slots fill with user experience and architecture. A developer whose pre-AI slots were already at the architectural level gains marginally because she had already compressed what the AI now compresses.
The stratification this produces is not between AI users and non-users. It is between those whose freed cognitive slots fill with higher-quality chunks and those whose freed slots fill with more quantity of the same. The first group experiences compression as transformative — a step up the ladder of cognitive quality. The second experiences it as merely accelerative — a faster version of the same patterns. Both become more productive. Only one becomes more wise. Miller's framework predicts this with uncomfortable precision: if capacity is fixed and quality is variable, any tool that frees slots amplifies the existing quality of a person's cognitive contents. The compression is an amplifier, and amplifiers do not discriminate.
Origin
The distinction between the quantity and quality of working memory contents is implicit in Miller's original framework but was not systematically developed in the 1956 paper. Later cognitive scientists — particularly K. Anders Ericsson and Walter Kintsch in their work on long-term working memory — extended the analysis to show that experts effectively expand functional capacity by drawing on richer chunks stored in long-term memory.
The application of this quality dimension to AI-mediated work is recent and still developing. Research programs at Stanford, MIT, and the Santa Fe Institute have begun empirically documenting the variance in AI-assisted productivity and tracing it to the chunking vocabularies developers bring to the collaboration.
Key Ideas
Quantity is fixed; quality is variable. Seven slots is the constant. What fits in each slot is the variable along which all cognitive progress occurs.
Quality is relational. A chunk is high- or low-quality relative to the demands of a specific situation. The same chunk can be valuable in one context and irrelevant in another.
Compression amplifies existing quality. AI tools free slots; whatever fills them reflects what the user was already inclined to attend to. High-quality thinkers become higher-quality; low-quality producers become faster low-quality producers.
Stratification by content, not access. The divide in the AI age is not between tool users and non-users. It is between users whose freed slots fill with depth and users whose freed slots fill with volume.
The quality investment. The most important investment in the AI age is not learning to use AI tools; it is improving the quality of the chunks that will fill the slots compression frees.
Appears in the Orange Pill Cycle
Further reading
- K. Anders Ericsson and Walter Kintsch, Long-Term Working Memory, Psychological Review, 1995
- George A. Miller, The Magical Number Seven, Plus or Minus Two, Psychological Review, 1956
- Herbert Simon, Sciences of the Artificial, MIT Press, 3rd edition, 1996
- Edo Segal, The Orange Pill, 2026