The Operon Library
The terminology map
Glossary of AI-assisted software engineering
A discipline needs a shared vocabulary. Coinages are marked as ours; adopted terms carry their origin.
5 terms
Operon coinages
Terms the Library introduces — used only where no industry word exists.
- Context Debt
- Lost conversational and decision context that must be re-earned before productive work resumes — the AI-era analogue of technical debt.
- Decision Memory
- Decisions persisted as first-class, queryable records — what was decided, why, and what it superseded.
- Flight Plans
- A goal decomposed into tracked steps that an AI session executes against, with progress detection and checkpoint gates.
- Session ROI
- Value created by an AI session measured against the time and money invested in it.
- The AI Cost Iceberg
- The visible costs of AI-assisted development (subscriptions, tokens) versus the larger hidden mass: context rebuilding, review time, retries, rework.
10 terms
Adopted terms
Established vocabulary, cited to its origin — never renamed.
- Agentic harness
- The tools, context management, and execution environment that turn a language model into a capable coding agent.Origin: Anthropic — Claude Code glossary
- Compaction
- Automatic summarization of a session as the context window approaches its limit, with defined survival rules for what persists.Origin: Anthropic — Claude Code docs
- Context engineering
- Choosing what fills the context window — the configuration of context most likely to generate the desired behavior.Origin: Andrej Karpathy; formalized by Anthropic engineering · 2025-06
- Context rot
- Degradation of model reliability as input context grows — models do not use context uniformly.Origin: Chroma Research · 2025-07
- DORA AI Capabilities Model
- DORA’s 2025 seven-capability model for organizations adopting AI-assisted development; frames AI as an amplifier of existing strengths and dysfunctions.Origin: DORA — State of AI-assisted Software Development 2025 · 2025
- DX Core 4
- A four-dimension developer-experience measurement framework (speed, effectiveness, quality, impact).Origin: DX · 2024
- LLM-as-judge
- Using a model call to score or gate another model’s output, typically with a rubric and a numeric score.Origin: Anthropic engineering — multi-agent research system
- SPACE
- A developer-productivity framework spanning Satisfaction, Performance, Activity, Communication, Efficiency.Origin: Forsgren et al., ACM Queue · 2021
- Spec-driven development
- Inverting code-first workflow: specifications are executable artifacts that generate implementations.Origin: GitHub spec-kit · 2025-09
- Subagent
- An agent spawned in its own isolated context window that returns a summary to its parent session.Origin: Anthropic — Claude Code docs