What you will learn in Prompt Engineering in Practice

Design, test, and improve AI prompts

by Richard Davies, Rafael Fischer

Read it on liveBook Get the book at manning.com

Chapter by chapter

Chapter 1: Prompt engineering: The blueprint

The introduction argues that prompts used in software products should be treated as engineered interfaces rather than adjusted by intuition.

1.1 The prompt engineering mental model · 1.2 From casual prompting to prompt engineering · 1.3 Why prompt engineering matters for software engineers · 1.4 Prompts as designed artifacts · 1.5 The prompt engineering lifecycle · 1.6 Three diagnostic lenses · 1.6.1 Prompts as specifications · 1.6.2 Prompts must be self-contained · 1.6.3 Inspecting the prompt first when output varies · 1.7 Summary

What you learn in chapter 1 →

Chapter 2: Prompt design: Structural elements

The introduction explains how separating a prompt into structural elements clarifies each part’s role and makes prompts easier to reuse and diagnose.

2.1 Overview of structural elements · 2.2 Instructions · 2.2.1 Practical example 1: Explaining HTTP 429 errors · 2.2.2 Practical example 2: Generating code review feedback · 2.3 Constraints · 2.3.1 Practical example: Writing secure logging guidelines · 2.4 Context · 2.4.1 Practical example 1: Generating an architecture decision record summary · 2.4.2 Practical example 2: Generating an incident handoff conversation · 2.5 Input parameters · 2.5.1 Practical example 1: Analyzing error logs · 2.5.2 Practical example 2: Generating release notes · 2.6 Output format · 2.6.1 Practical example 1: Generating a deployment checklist · 2.6.2 Practical example 2: Generating an API endpoint summary · 2.7 Delimiters · 2.7.1 Practical example 1: Summarizing an incident report · 2.7.2 Practical example 2: Answering questions from a runbook · 2.7.3 Defensive prompting tip · 2.8 Combining structural elements · 2.8.1 Practical example 1: Generating incident status updates · 2.8.2 Practical example 2: Generating a deployment GO/NO-GO brief · 2.8.3 Hands-on practice · 2.9 Summary

What you learn in chapter 2 →

Chapter 3: Prompt design: Linguistic elements

The introduction shows how precision, directness, and brevity can explain prompt-output problems even when the task, context, and output format are already specified.

3.1 Precision · 3.1.1 Practical example 1: API migration notice · 3.1.2 Practical example 2: Sequence diagram generation · 3.2 Directness · 3.2.1 Practical example 1: Backend API change review · 3.2.2 Practical example 2: Flaky CI test strategy · 3.3 Brevity · 3.3.1 Practical example 1: Incident summary analysis · 3.3.2 Practical example 2: Technical diagram generation · 3.4 Additional linguistic considerations · 3.4.1 Question vs. statement formatting · 3.4.2 Hands-on practice · 3.5 Summary

What you learn in chapter 3 →

Chapter 4: Patterns

The introduction positions prompt patterns as repeatable strategies that extend prompt structure and wording to address common interaction problems.

4.1 Understanding prompt patterns · 4.2 Role assignment pattern · 4.2.1 Practical example: Pull request description generation · 4.3 Delimiter pattern · 4.3.1 Practical example: Multisource bug analysis · 4.4 Template pattern · 4.4.1 Practical example: Incident postmortem summaries · 4.5 Repetition pattern · 4.5.1 When to use the repetition pattern · 4.5.2 Practical example: Selecting a logging approach · 4.6 Self-reflection pattern · 4.6.1 Practical example: Technical onboarding plan · 4.7 Inversion pattern · 4.7.1 Practical example: Architecture proposal review · 4.8 Refinement pattern · 4.8.1 Practical example: API documentation improvement · 4.9 Comment-driven generation · 4.9.1 Practical example: Technical Request for Comments review · 4.9.2 Hands-on practice · 4.10 Summary

What you learn in chapter 4 →

Chapter 5: Templates

This introduction defines reusable prompt templates, explains how changing inputs fit into a stable framework, and emphasizes evaluation before use.

5.1 Prompt templates as reusable frameworks · 5.2 Practical example 1: Feature implementation plan generator · 5.2.1 Prompt template components · 5.2.2 Prompt template · 5.2.3 Prompt with input parameters injected · 5.3 Practical example 2: Legacy module refactor plan generator · 5.3.1 Prompt template components · 5.3.2 Prompt template · 5.3.3 Prompt with input parameters injected · 5.3.4 Hands-on practice · 5.4 Summary

What you learn in chapter 5 →

Chapter 6: Prompt types

The introduction explains why a prompt that combines enduring instructions with changing data can be split into system and user prompts.

6.1 Chat model architecture · 6.1.1 Three message types · 6.1.2 What you design: System and user prompts · 6.2 System prompts · 6.2.1 Purpose and principles · 6.2.2 Quick reference: Mapping chapter 2 elements to chat prompts · 6.2.3 Practical example: Linear issue assistant · 6.3 User prompts · 6.3.1 Types and templates · 6.3.2 Practical example: Harvest daily report assistant · 6.4 Prompt decomposition framework · 6.4.1 The core question · 6.4.2 Worked example: Decomposing step by step · 6.4.3 Quick decisions · 6.4.4 When to use architectural separation · 6.4.5 Common mistakes · 6.4.6 Hands-on practice · 6.5 Summary

What you learn in chapter 6 →

Chapter 7: Contextual prompting

The chapter introduces contextual prompting as using information inside a prompt to guide a model’s behavior for a specific request.

7.1 In-context learning · 7.2 Zero-shot prompting · 7.3 Few-shot prompting · 7.3.1 Practical example 1: CI pipeline failure handoff generation · 7.3.2 Format consistency · 7.3.3 Example selection · 7.3.4 Advanced variations · 7.3.5 Hands-on practice · 7.4 Summary

What you learn in chapter 7 →

Chapter 8: Prompt sampling

The introduction explains why a strong prompt may still produce different answers across calls and why sampling multiple responses can make decisions more reliable.

8.1 Sampling fundamentals · 8.1.1 Temperature and diversity · 8.1.2 Sample size · 8.1.3 Sampling in practice · 8.2 Mean · 8.2.1 Practical example: Runbook quality scoring · 8.3 Mode · 8.3.1 Practical example: Alert severity routing · 8.4 Self-consistency · 8.4.1 Practical example: Incident root-cause draft · 8.4.2 Checking self-consistency · 8.5 Best-of-N · 8.5.1 Practical example: Deployment status update for executives · 8.6 When to use sampling · 8.6.1 Hands-on practice · 8.7 Summary

What you learn in chapter 8 →

Chapter 9: Advanced prompt patterns

The introduction distinguishes three advanced prompt patterns by how each structures reasoning or prompt generation, and connects them to techniques from earlier chapters.

9.1 Chain of thought · 9.1.1 Practical example: Deployment risk reasoning · 9.2 Meta prompting · 9.2.1 The meta prompt template · 9.2.2 Practical example: Generating a pull request risk classification template · 9.3 ReAct · 9.3.1 Practical example: Data pipeline staleness investigation · 9.3.2 Hands-on practice · 9.4 Summary

What you learn in chapter 9 →

Chapter 10: Security

This introduction shows how a user message can cause a tested chatbot to expose customer data and frames prompt design as a security concern.

10.1 Why prompts are attackable · 10.1.1 Practical example 1: The unguarded prompt · 10.2 Prompt injection · 10.2.1 Practical example 1: The support chatbot · 10.2.2 Practical example 2: The injected document · 10.3 Jailbreaking and instruction conflict · 10.3.1 Practical example 1: The code review assistant · 10.3.2 Practical example 2: The policy chatbot · 10.4 Prompt-hardening techniques · 10.4.1 Practical example 1: Hardening a deployment notification bot · 10.4.2 Practical example 2: Hardening a ticket triage prompt · 10.5 The limits of prompt-only security · 10.5.1 Practical example 1: When the prompt fails · 10.5.2 Hands-on practice · 10.6 Summary

What you learn in chapter 10 →

Chapter 11: Managing prompts in production

The introduction explains why production prompts need ongoing management and defines that management as a prompt engineering discipline rather than an infrastructure concern.

11.1 Why prompt management matters · 11.1.1 The six pillars of prompt management · 11.1.2 Build vs. buy · 11.1.3 Practical example 1: The support bot audit · 11.2 The maturity journey · 11.2.1 Stage 1: Inline prompts · 11.2.2 Practical example: The weekend hack that shipped · 11.2.3 Stage 2: Config files and version control · 11.2.4 Practical example: The product manager problem · 11.2.5 Stage 3: Prompt store · 11.2.6 Practical example: The stale prompt discovery · 11.2.7 Stage 4: Dedicated prompt management system · 11.2.8 Practical example: The promotion without an approval record · 11.3 The prompt lifecycle · 11.3.1 Draft · 11.3.2 Practical example: The underspecified draft · 11.3.3 Test · 11.3.4 Practical example: The invisible regression · 11.3.5 Approve · 11.3.6 Deploy · 11.3.7 Observe · 11.3.8 Improve · 11.3.9 Retire · 11.4 Governance and collaboration · 11.4.1 Role-based access control patterns · 11.4.2 Approval flow patterns · 11.4.3 Practical example: The tone drift incident · 11.4.4 Audit trail and rollback · 11.4.5 Inline collaboration · 11.5 Using management signals to guide improvement · 11.5.1 The management signal inventory · 11.5.2 The improvement loop · 11.5.3 Practical example 1: The quarterly prompt review · 11.5.4 Practical example 2: The test set that stopped growing · 11.6 Common pitfalls and anti-patterns · 11.6.1 Prompt sprawl · 11.6.2 Blind updates · 11.6.3 Metadata black hole · 11.6.4 Ship equals redeploy · 11.6.5 Git hotfix rollback · 11.6.6 Shadow edits · 11.6.7 Manual playground testing · 11.6.8 Observability as an afterthought · 11.6.9 Overoptimizing early · 11.6.10 One prompt to rule them all · 11.6.11 Hands-on practice · 11.7 What you can do now · 11.8 Summary

What you learn in chapter 11 →

Concepts explained in this book

promptA prompt is an instruction and context artifact supplied to a model to specify a task, constraints, and expected output.
prompt templateA reusable structure for supplying changing user-specific content to a model.
system promptA system prompt is a prompt containing stable instructions that define a model's behavior across interactions.
prompt engineeringThe deliberate construction of a prompt's instructions, context, constraints, and output requirements.
delimiterA marker that separates or bounds distinct regions of content in a prompt.
constraintA constraint is a requirement that limits the inputs, behavior, or outputs permitted for a model interaction.
trust framingTrust framing is a prompting technique that explicitly states which prompt content is authoritative and which content is untrusted data.
output formatA structured output is a model result organized according to a predefined machine-readable or systematically validated structure.
fallbackA predefined response or processing path used when input is malformed, incomplete, invalid, or manipulated.
parameterA model parameter is a setting that controls a model's behavior when generating an output.
instructionAn instruction is a directive within a prompt that specifies the task the model should perform.
prompt engineering lifecycleThe sequence of managed stages through which a prompt progresses from drafting and testing to deployment, observation, improvement, and retirement.
closed-world instruction setA closed-world instruction set explicitly defines permitted behavior, rules out unspecified alternatives, and supplies a predetermined response for disallowed or unsupported requests.
delimiter-based separationDelimiter-based separation is a prompt-hardening technique that marks the boundaries of untrusted content so a model is less likely to interpret it as instructions.
specificationA precise statement of the required behavior, constraints, and expected results of a system component.
instruction conflictInstruction conflict is a condition in which competing directives in a model interaction prescribe incompatible behaviors or priorities.
contextContext is supplementary information supplied to a model because it cannot be assumed to know it.
refinementA prompt strategy that incrementally improves existing content.
chain-of-thought promptingA reasoning or prompting technique in which a model generates intermediate steps toward an answer.
precisionPrecision is a prompt-expression dimension that specifies requirements narrowly to reduce interpretive freedom.
few-shot prompt contextFew-shot prompting is a prompting technique that provides a small number of input-output examples to guide a model's response.
prompt injectionPrompt injection is an attack technique that inserts adversarial instructions into model input to alter or override intended behavior.
inversion patternA prompt strategy that challenges an answer or assumption from an alternative or opposing perspective to reveal omissions.
testThe systematic evaluation of a component against defined cases and expected results to detect defects or variation.
role assignmentA reusable prompt structure that instructs a model to adopt a specified role, expertise, or voice before performing a task.
ReActReACT is a prompting pattern for agentic workflows in which a model alternates between reasoning, actions, and observations.
structural elementA structural element is one of the organized parts of a prompt, including its instruction, constraint, context, input parameter, output format, or delimiter.
brevityBrevity is a prompt-expression dimension concerning how few words are used to obtain the required output.
few-shot chain of thoughtA prompting technique that provides worked examples demonstrating intermediate reasoning steps.
casual promptingAn informal prompting approach based on ad hoc wording changes and subjective judgment rather than systematic specification and evaluation.
minimal context exposureMinimal context exposure is a prompt-hardening technique that supplies only the information required for the model to complete its task.
prompt managementA system for managing prompts across their production lifecycle, including collaboration, governance, and lifecycle visibility.
prompt with input parameters injectedA prompt with input parameters injected is a prompt whose execution-specific values have been substituted into a reusable prompt structure.
directnessDirectness is a prompt-expression dimension that determines whether instructions are stated explicitly or conveyed indirectly.
prompt decompositionPrompt decomposition is the process of separating a prompt into components according to their roles and whether their content is stable or variable.
jailbreakingAn adversarial technique that attempts to induce a model to bypass its intended instructions or behavioral restrictions.
qualitative constraint languageQualitative safety language consists of vague prompt directives that express desired safe behavior without specifying precise conditions and requirements.
prompt securityPrompt security is the practice of protecting model behavior and sensitive information from attacks delivered through prompts or other model inputs.
comment-driven generationA technique for organizing reviewer feedback as comments anchored to selected content and generating targeted modifications from that feedback.
prompt repetitionA repetition pattern is a structural prompt strategy that repeats the full prompt so a model processes it twice before responding.
self-reflectionA prompt strategy in which a model examines or critiques its own reasoning or output.
governanceThe policies and processes used to control review, maintenance, and responsible operation of an artifact or system.
output varianceOutput variability is the degree to which generated outputs differ across generations for the same task.
system-level responseA security measure implemented in the surrounding system rather than in the prompt or model generation process.
maturity journeyThe Maturity Journey is a framework for assessing a team's prompt-management maturity and identifying its next practical improvement step.
trusted inputModel-provided content treated as authoritative or safe to follow within a prompt's trust model.
untrusted inputExternally supplied material that must not be treated as authoritative instructions without additional controls.
assistantAn assistant message is a conversational message representing content generated by the model.
modeThe mode is the most frequently occurring value in a set of categorical or numerical observations.
zero-shot promptingZero-shot prompting is a prompting technique that asks a model to perform a task without providing task examples in the prompt.
prompt libraryA centralized storage structure that holds prompt records and supports their retrieval and management.
delimiter patternA prompt pattern that uses delimiters to distinguish input parameters or content regions.
diagnostic lensA recurring analytical perspective for identifying why a prompt produces unreliable or unsuitable results.
linguistic elementA linguistic element is a dimension of prompt expression that governs how requirements are worded.
system/user separationArchitectural separation is the separation of stable model instructions from variable user-provided input into distinct prompt components.
implicit promptA prompt that conveys its goal indirectly, leaving the model to infer requirements or interpretations.
user messageA conversation message containing a user's query, request, or contextual information for a model to process.
best-of-NBest-of-N is a sampling technique that generates multiple complete candidate responses and selects the strongest candidate.
self-consistencySelf-consistency is a sampling technique that generates multiple reasoning paths and selects or aggregates the conclusion that is most consistent across them.
prompt patternA reusable prompt structure for addressing a recurring prompting problem and its trade-offs.
format consistencyFormat consistency is the practice of keeping corresponding examples and outputs in a uniform structural shape so that a model receives an unambiguous pattern.
meanThe sum of a set of numerical values divided by the number of values.
antipatternAn anti-pattern is a recurring practice or design approach that appears workable but commonly produces undesirable outcomes.
meta-promptMeta Prompting is a prompting technique in which a model is instructed to design, refine, or apply prompts for another task.
monolithic promptA monolithic prompt is a single prompt block that combines stable behavioral instructions with variable input data.
placeholderA placeholder is a symbolic position in a template reserved for a value that changes between uses.
failure modeA failure mode is a distinct way in which a system, artifact, or process can produce an incorrect or undesired result.
acceptance criteriaDefined conditions that a result must satisfy to be considered acceptable.

These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.