AI Agent Prompting Techniques

FOUNDATIONAL TECHNIQUES

Basic

RBP

Role-Based Prompting

Assign specific persona for specialized, consistent outputs with appropriate tone and expertise

Basic

FS

Few-Shot Learning

Provide examples in prompt to guide behavior and demonstrate desired output patterns

Basic

SYS

System Prompt

Define agent behavior, capabilities, and constraints at the system level

Basic

ZS

Zero-Shot

Rely solely on instructions without examples, testing model's general capabilities

REASONING TECHNIQUES

Core

CoT

Chain-of-Thought

Generate step-by-step reasoning for complex math, logic, and sequential problems

Core

CoT-Z

Zero-Shot CoT

Trigger reasoning with "Let's think step by step" without providing examples

Core

CoT-F

Few-Shot CoT

Show examples with complete reasoning paths to guide problem-solving approach

Advanced

SC

Self-Consistency

Sample multiple reasoning paths and select most frequent answer for robustness

Advanced

L2M

Least-to-Most

Break complex problems into progressively simpler sub-problems

ACTION-TAKING TECHNIQUES

Core

ReAct

Reason + Act

Interleave reasoning (thought) with action (tool use) in iterative loop

Component

TC

Tool Calling

Execute external functions and APIs as part of ReAct framework to interact with environment

QUALITY CONTROL TECHNIQUES

Essential

GC

Gate Checks

Programmatic validation between workflow steps to catch errors early

Essential

VAL

Validation

Verify outputs match expected format, constraints, and business rules

Core

LLM-J

LLM-as-Judge

Use LLM to evaluate other LLM outputs for quality and correctness

Essential

SO

Structured Output

Enforce specific formats (JSON, Pydantic) for reliable, parseable responses

WORKFLOW TECHNIQUES

Core

PC

Prompt Chaining

Link multiple LLM calls programmatically to break down complex workflows

Core

FL

Feedback Loops

Iteratively refine outputs using evaluation results for self-improvement

Advanced

PR

Prompt Refinement

Systematically optimize prompts based on performance data and testing

Pattern

RT

Routing

Application pattern to direct tasks to appropriate workflows based on input classification

PRODUCTION TECHNIQUES

Critical

EVAL

Evaluation

Measure agent performance using accuracy, precision, recall, and custom metrics

Critical

MON

Monitoring

Track latency, costs, usage patterns, and performance in production

Critical

TRC

Traces

Log all inputs, outputs, and intermediate steps for debugging and analysis

Essential

OBS

Observability

Understand actual vs expected behavior through comprehensive logging

The Prompt Engineering Periodic Table