AI-DLC Glossary
Quick reference for all AI-DLC terminology and concepts.
A
- AHOTLSee in paper
- Autonomous Human-on-the-Loop: human defines criteria and reviews output; AI operates autonomously within boundaries; used for well-defined, programmatically verifiable work
B
- BackpressureSee in paper
- Quality gates that automatically reject work not meeting criteria, providing feedback for iteration
- BoltSee in paper
- Smallest iteration unit in AI-DLC 2026; operates in supervised (HITL), observed (OHOTL), or autonomous (AHOTL) mode
C
- Completion CriteriaSee in paper
- Programmatically verifiable conditions that define when work is successfully done
- Completion PromiseSee in paper
- Signal (e.g., COMPLETE, BLOCKED) that autonomous execution has finished
- Context BudgetSee in paper
- Available attention capacity in AI context window; quality degrades when overloaded
H
- HITLSee in paper
- Human-in-the-Loop: human validates each significant step before AI proceeds; used for novel, high-risk, or foundational work
I
- IntegratorSee in paper
- Final validation hat that runs conditionally based on VCS strategy; validates auto-merged state (trunk) or creates single PR (intent); skipped for unit/bolt strategies
- IntentSee in paper
- High-level statement of purpose with completion criteria that serves as starting point for decomposition
M
- Memory ProviderSee in paper
- Source of persistent context (files, git, tickets, ADRs, runbooks) accessible to AI agents
- Mob ConstructionSee in paper
- Collaborative ritual where multiple teams build Units in parallel with AI assistance
- Mob ElaborationSee in paper
- Collaborative ritual where humans and AI decompose Intent into Units with Completion Criteria
O
- OHOTLSee in paper
- Observed Human-on-the-Loop: human watches AI work in real-time with ability to intervene; synchronous awareness with asynchronous control; used for creative, subjective, or training scenarios
Q
- Quality GateSee in paper
- Automated check (tests, types, lint, security) that provides pass/fail feedback
R
- Ralph Wiggum PatternSee in paper
- Autonomous loop methodology: try, fail, learn, iterate until success criteria met
U
- UnitSee in paper
- Cohesive, independently deployable work element derived from an Intent; named with numerical prefix + slug (e.g., `unit-01-setup-auth`); can declare dependencies via `depends_on` forming a DAG
- Unit DAGSee in paper
- Directed Acyclic Graph of unit dependencies enabling parallel execution (fan-out) and convergence (fan-in)