Skip to content

Glossary

Multi-step model workflows, harness responsibilities, state, memory and reusable workflow patterns.

Designing, building, evaluating, deploying and improving software systems in which AI models perform part of the system work.

Assessing whether AI belongs in the task and whether errors are detectable, tolerable and recoverable.

Boundaries, contracts, integration and structural decisions that make AI systems changeable.

Maintainability, readable diffs and reversible change in human and agent-written code.

The difference between local code help and goal-directed multi-step agent work.

Model configuration, MCP integrations, execution boundaries, approvals and secrets.

Designing the complete information environment available to a model at a particular step.

Managing finite context, persistent project guidance, skills and adaptive plans.

Cost, context, communication, conflict and stopping-condition risks in multi-agent work.

Representing entities, relations, events and lifecycle so AI has meaningful data to work with.

Machine learning using multilayer neural networks, underpinning modern vision, speech, multimodal systems and LLMs.

How orchestrators assign, combine and account for work across agents.

Deciding what belongs in rules or conventional code and what can responsibly use a model.

Build, release and operations practices for observable and recoverable AI software.

Embeddings, Vector Search, Graphs and Semantic Layers

Section titled “Embeddings, Vector Search, Graphs and Semantic Layers”

Retrieval building blocks and their trade-offs for semantic, structured and relationship-heavy information.

Structured tests and qualitative inspection for improving probabilistic AI-system behaviour.

Evidence dimensions for judging agent output, process quality, safety and robustness.

Declaring success based on concrete evidence such as passing checks, inspected diffs or rendered behaviour.

Foundation models, LLMs, multimodal models and model selection as production engineering choices.

Where humans retain authority over risk, irreversible action and release decisions.

Connecting model output to trusted, task-relevant sources or system state.

Controls that constrain AI inputs, outputs or actions and keep human authority meaningful.

Methods that learn patterns from data instead of expressing every rule explicitly.

The practical components that determine what a coding agent can do.

Fan-out/fan-in, pipelines, reviewer patterns and debate patterns.

When separate agents help and when a single agent is the better engineering choice.

Outcome metrics, guardrail metrics, build-measure-learn loops and decision records.

Operational principles for getting useful, reviewable and reversible work from coding agents.

Defining the situation, affected people, desired change and competing interests before selecting an AI solution.

Testing falsifiable product hypotheses with prototypes and vertical slices.

Production Operations, Observability, Cost and Latency

Section titled “Production Operations, Observability, Cost and Latency”

Running AI systems under real deployment, tracing, reliability, cost and performance constraints.

Designing instructions and examples that elicit useful model behaviour.

Requirements engineering, computational thinking and design for AI-supported systems.

Assessing hazards and designing accountable AI systems with meaningful human oversight.

Security, privacy and inclusive design constraints for AI-supported applications.

Specifications, Quality Attributes and Acceptance Criteria

Section titled “Specifications, Quality Attributes and Acceptance Criteria”

Turning desired behaviour, constraints and quality into verifiable conditions.

A reusable task brief structure for bounded, verifiable agent work.

Technical Debt, Documentation and Teamwork

Section titled “Technical Debt, Documentation and Teamwork”

Managing future cost, decision records, collaboration and review in fast-moving AI projects.

Traditional tests for deterministic behaviour and evals for probabilistic behaviour.

Observe, reason, act, inspect, update and stop or continue.

A readable process for directing coding agents from task framing through shipping and observation.

Allowing models to request controlled operations through validated tools and standard interfaces.

Understanding tasks, handoffs, exceptions and user progress before changing a workflow.

Value, Feasibility, Viability and Responsibility

Section titled “Value, Feasibility, Viability and Responsibility”

Four tests for deciding whether a shaped AI intervention deserves further investment.