Agentic Periodic Table

Prompt Engineering

Structuring input text to shape model output through precise instruction.

Zero / Few-shot

Providing zero or a handful of examples to guide model behaviour in-context.

Context Window

Using the active prompt window as the sole memory store for a single exchange.

Function Call

Invoking a single named tool or API from a model response.

Solo Agent

One model, one task, no inter-agent communication.

Format Validation

Checking that output matches an expected schema or format before use.

Prompt Guardrails

Hard-coded constraints in the system prompt that limit unsafe or off-topic output.

RAG

Retrieval-augmented generation: fetching relevant chunks and injecting them into context before reasoning.

Chain-of-Thought

Prompting the model to reason step-by-step before producing a final answer.

Scratchpad

A writable working-memory buffer the agent uses to track intermediate state across steps.

Tool Chaining

Sequentially calling multiple tools where each output feeds the next input.

Sequential Handoff

Passing a task linearly from one agent or step to the next.

Self-Critique

The model reviews and scores its own output before returning it.

Checkpoint Approval

Human approval gate inserted at defined points in an agentic workflow.

Multi-source Retrieval

Querying and fusing results from multiple retrieval sources before reasoning.

Tree-of-Thought / ReAct

Exploring multiple reasoning branches and interleaving thought with tool use to solve complex tasks.

Vector Store

Persisting and retrieving semantic chunks via embedding similarity search.

Parallel Tool Exec.

Dispatching multiple tool calls simultaneously and merging results.

Supervisor / Worker

An orchestrator agent delegates sub-tasks to specialised worker agents.

Cross-agent Critique

A dedicated critic agent reviews the output of a peer agent.

Interrupt & Escalation

The agent pauses and routes a decision to a human when confidence is low or risk is high.

Active Sensing

The agent proactively queries its environment for information rather than waiting for input.

Reflection + Replan

The agent evaluates its own trajectory and revises its plan mid-execution.

Episodic Memory

Storing and recalling structured records of past interactions or task runs.

Autonomous Action Loop

The agent executes a self-directed action loop with intermediate observation and re-planning.

Peer-to-Peer Network

Agents communicate directly with one another without a central orchestrator.

Automated Eval Loop

Continuous automated scoring of agent outputs against defined success metrics.

Audit Trail / Observ.

Logging every agent decision, tool call, and output for traceability and review.

Ambient Awareness

Continuous passive monitoring of the environment to detect relevant signals autonomously.

Meta-cognition

The agent reasons about its own capabilities, limitations, and reasoning process.

Persistent World Model

A continuously updated structured model of the environment that persists across sessions.

World-affecting Exec.

The agent takes actions with real-world side-effects: deploying code, sending messages, moving funds.

Swarm Coordination

Many agents coordinate via emergent behaviour and shared signals with no central authority.

Self-improving Loop

The agent iteratively updates its own prompts, tools, or weights based on performance feedback.

Constitutional Alignment

Encoding explicit principles the agent uses to self-evaluate and refuse misaligned instructions.