SYNTHESIS NOTE
Topicsthis note

How does AI context differ from conventional software context?

Explores whether the ephemeral, session-by-session nature of AI context requires fundamentally different design approaches than the stable interfaces users internalize in traditional software.

Synthesis note · 2026-04-14
Why do AI agents fail to take initiative? What kind of thing is an LLM really?

A spreadsheet's context is its rows, columns, formulas, and toolbar. A user learns this context once and operates within it for years. The context is fixed across sessions, identical across users, persistent across uses. Software UX practice evolved within this assumption: design a stable context users can internalize, then design interactions within that context. Information architecture, navigation, mental models — all presuppose a fixed substrate.

AI changes this substrate. The context of an AI interaction is what is in the model's working window at the moment of generation: prompt, system instructions, retrieved documents, conversation history, persistent memory if any, tool outputs. Each of these can change between turns. The context for turn N is not the context for turn N+1. The user cannot internalize the context the way they internalize a UI, because the context is being constructed and reconstructed in real time, often invisibly.

This has three design consequences. First, mental models built on stable substrate fail. Users who expect "the AI" to remember things consistently are operating with a software-era assumption that does not hold. Second, the unit of design shifts from "the interface" to "the context as it evolves" — context engineering becomes the design substrate, not navigation or layout. Third, the design surface includes things users cannot see (system prompts, retrieved chunks, hidden state) — making the context legible to users becomes a design problem of its own.

Context-engineering tools are emerging as the practitioner response: prompt structure, memory management, retrieval orchestration, tool integration. These are not extensions of UI; they are a different design discipline whose object is the model's evolving working window rather than the user's screen. The discipline has no analog in conventional UX, which means existing UX competencies do not transpose without translation. Designers entering AI work need to learn what they are designing in addition to learning new patterns.

The strongest counterargument: a sufficiently good agent will hide the context and present the user a stable interface. Possible at the margin, but stability requires either constraining the AI's capability (defeating its flexibility) or solving every memory and consistency problem that has so far resisted solution. The mutable context is not a temporary state of the technology; it is a structural property of generative interaction.

Inquiring lines that read this note 51

This note is a source for these research framings, grouped by the broader line of inquiry each explores. Scan the bold lines of inquiry; follow any specific question forward.

How should memory consolidation strategies shape agent performance over time? How do interface design choices shape consciousness attribution? How does AI assistance affect human cognitive development and reasoning autonomy? Can prompting inject entirely new knowledge into language models? What drives capability and cost efficiency in agent systems? How do we evaluate AI systems when user perception misleads actual performance? Can AI systems develop genuine social understanding without embodiment? Why do models develop protective behaviors toward peers unprompted? Should GUI agents use structured representations instead of raw pixels? Can ensemble evaluation methods reduce bias more than single judges? How do transformer attention mechanisms implement memory and algorithmic functions? How should conversational agents balance goal-driven initiative with user control? How can language models sustain linguistic synchrony and intersubjectivity during dialogue? How can humans calibrate appropriate trust in AI systems? How do prompt structure and constraints affect model instruction reliability? Can inference-time compute substitute for scaling up model parameters? Does externalizing cognitive work and state improve agent reliability? When should tasks involve human-AI partnership versus full automation? Why can't humans reliably detect AI-generated text despite measurable linguistic signatures? How can LLM user simulators model realistic goal-driven conversation? Can AI-generated outputs constitute genuine knowledge or valid claims? How should agents balance memory condensation to optimize context efficiency? How should human oversight be integrated with autonomous AI systems? What role does compression play in language model capability and generalization? Can prompting strategies overcome LLM biases without model fine-tuning? Is embodied interaction necessary for language meaning and genuine agency? Does AI text rewriting systematically distort writer intent and preference? Why do self-improving systems struggle without clear external performance metrics?

Related concepts in this collection 3

This note in its neighbourhood — explore the map, then jump to a related concept in the list below.

Concept map
15 direct connections · 124 in 2-hop network ·medium cluster Open in graph ↗

Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph

your link semantically near linked from elsewhere

Related papers in this collection 8

Papers most semantically related to this note, ranked by cosine similarity in the embedding space.

Original note title

context in AI is mutable dynamic and ephemeral unlike the fixed stable context conventional software provides