The Patterns That Emerge Across Diverse Professional Contexts
The five walkthroughs in this module describe professionals whose work is genuinely different from one another in almost every material respect. A strategy consultant working across multiple client engagements operates in a fundamentally different environment from a litigation paralegal managing discovery productions, a claims analyst processing insurance losses, a financial analyst producing management reporting, or an operations manager overseeing a logistics facility. The tools they use, the documents they produce, the regulatory frameworks they operate under, the consequences of errors in their work, and the professional standards they are accountable to are all distinct.
Despite this genuine diversity, the same structural patterns appear in every walkthrough. These patterns were not imposed on the walkthroughs by design. They emerged from the application of the same analytical framework to five different professional contexts, and their appearance across all five is evidence that they reflect something stable and fundamental about how AI assistance integrates effectively into professional knowledge work rather than being idiosyncratic features of specific roles or industries.
Understanding why these patterns appear consistently, and why departing from them reliably produces worse outcomes, is the most transferable lesson the walkthroughs offer. A professional who understands the logic behind the common patterns can apply that logic to their own context with confidence, even when their specific role differs from all five walkthroughs.
Common Pattern One: Selectivity Is a Feature, Not a Limitation
None of the five professionals described in this module use AI for everything. Sarah does not use AI to build presentation slides. Marcus does not use AI to make privilege determinations or draft legal conclusions. Priya does not use AI to assess photographic damage evidence. David does not use AI to construct or audit financial model logic. Jennifer does not use AI to manage vendor disputes or handle sensitive personnel matters.
These exclusions are not provisional: they are not things these professionals intend to automate once AI capability improves, or temporary workarounds for current technical limitations. They are deliberate boundaries that reflect a considered judgment about where AI assistance adds value and where it does not, based on a clear understanding of the specific properties of each excluded task.
The tasks excluded from all five AI practices share a specific characteristic: they are tasks where the value of the output depends critically on properties that the professional brings and that AI tools cannot supply. These properties fall into three categories.
The first is relationship and contextual judgment. The calibration of a client communication to the specific dynamics of a long-standing relationship, the handling of a sensitive personnel conversation with the full awareness of an employee's history and circumstances, the management of a vendor dispute with knowledge of the commercial and personal dynamics involved: these require a kind of situated human judgment that depends on knowing the specific people involved in ways that AI tools cannot replicate regardless of how comprehensive the context documents are. Context documents can brief an AI tool on the facts of a situation. They cannot give it the interpretive sensitivity that comes from sustained personal engagement with the individuals involved.
The second is professional accountability. Legal conclusions, coverage determinations, financial model audits, and strategic recommendations are not merely analytical outputs. They are professional judgments on which other parties will rely and for which the professional bears personal and institutional accountability. The professional's responsibility for these judgments is not transferable to an AI tool, and a professional who treats AI outputs as a substitute for their own judgment in these areas has not reduced their accountability: they have failed to discharge it. The walkthroughs reflect this through the consistent pattern of AI assistance producing first drafts and supporting analysis while human review and approval remains the non-negotiable final step for any output on which professional accountability is engaged.
The third is physical and visual assessment. The evaluation of photographic damage evidence in an insurance claim, the assessment of a physical site in real estate or construction work, the observation-based validation of an operational process in a warehouse: these require direct physical engagement with the subject of the assessment that AI tools, at their current stage of development, cannot perform reliably. The claims analyst who reviews damage photographs manually is not failing to leverage available technology. She is applying the appropriate tool for the task.
The practical lesson of this pattern is that selectivity is itself a professional competence. A professional who has thought carefully about where AI assistance adds value and where it does not, and who has made explicit decisions about the boundary between AI-assisted and manual work, is in a fundamentally better position than one who uses AI indiscriminately across all tasks or who avoids it indiscriminately out of caution. The discrimination between the two kinds of tasks is what makes an AI practice genuinely productive rather than merely busy.
Common Pattern Two: Context Documents Are the Foundation of AI Usefulness
Every walkthrough in this module rests on a set of context documents that are specific to the professional's role, their clients or counterparties, their current engagements, and the terminology and history of their work. Without these documents, the AI tools used in each walkthrough would produce outputs that are generically competent but professionally useless: responses that accurately reflect general knowledge of the relevant domain but that fail to address the specific situation, the specific constraints, the specific relationships, and the specific history that distinguish useful professional advice from generic information.
The reason this pattern appears consistently across five different professional contexts is that it reflects a property of AI tools that does not vary between roles: AI tools can only respond to what they are given. A model with excellent general capability and no access to the specific context of a situation will produce a general response. The same model with access to a current, accurate client background document, a precise project scope, a documented decision log, and a glossary of the specific terminology in use will produce a response that is directly applicable to the specific situation. The quality difference between these two responses is not a function of the model's capability. It is a function of the information made available to the model.
The implication for building an AI practice is that investment in context documents produces a higher return per hour invested than almost any other element of the practice. A professional who spends three hours in the first week of a new engagement writing a comprehensive client background document, a precise scope document, and a structured key findings log will recover that investment through improved AI output quality over the first few AI sessions of the engagement, and the return continues to compound for as long as the documents are maintained. A professional who skips this investment because it feels like a distraction from the actual work will receive consistently mediocre AI outputs throughout the engagement, because the AI tool will never have the information it needs to produce anything better.
The maintenance of context documents, addressed in the weekly knowledge base habit from Module 4.1, is as important as their initial creation. A context document that was accurate three months ago and has not been updated since is not a neutral resource: it actively misleads the AI tool by providing a description of the situation that no longer reflects reality. The AI tool produces outputs based on the context it has been given, and if that context is wrong, the outputs will be wrong in ways that may not be immediately apparent to the professional receiving them.
Common Pattern Three: Verification Is an Unconditional Professional Obligation
In every walkthrough, without exception, the professional verifies AI outputs before incorporating them into their work. The verification standard varies between task types and between roles, reflecting the different accuracy requirements and different consequences of error in each context. The paralegal verifies every legal citation in a primary source database because the consequences of relying on a fabricated citation in a legal filing are severe. The financial analyst verifies every numerical figure in an AI narrative against the source data because inaccuracy in financial communications undermines the credibility of the finance function. The claims analyst verifies every policy provision reference in a coverage analysis against the actual policy document because an incorrect coverage determination has direct financial and regulatory consequences.
The consistency of this verification pattern across five roles reflects a principle that holds regardless of the specific professional context: AI tools produce outputs through a probabilistic process that does not guarantee accuracy, and the professional who relies on those outputs without independent verification has accepted responsibility for the errors they contain. This is not a temporary state of affairs that will change when AI models become more capable. Current AI models at the frontier of capability make errors. Future models will also make errors, likely less frequently and in different ways, but they will make them. The professional's verification obligation is a permanent feature of the relationship between AI assistance and professional responsibility, not a transitional accommodation to current technological limitations.
The verification discipline described in the walkthroughs is not onerous. It is proportionate to the task: a cursory consistency check for a low-stakes internal draft, a more systematic figure-by-figure comparison for a financial narrative, a full primary source check for every citation in a legal document. The investment in verification is calibrated to the cost of the error it is designed to prevent, not to a blanket standard applied regardless of risk. Developing this calibrated approach to verification is itself a professional skill that improves with practice.
Common Pattern Four: High-Judgment Work Requires Human Deliberation
The tasks that remain entirely manual in all five walkthroughs share a structural property: their output is not primarily a document or a calculation but a professional judgment. A coverage determination is a judgment about how a policy applies to a specific set of facts, based on the professional's understanding of the policy language, the facts of the loss, the applicable regulatory framework, and the organisation's coverage guidelines. A legal conclusion is a judgment about how the law applies to a specific situation, based on the lawyer's understanding of the applicable authorities, the specific facts of the matter, and the professional and strategic context. A strategic recommendation is a judgment about what action a specific organisation should take, based on analysis of the available options, the organisation's specific capabilities and constraints, and the strategic context in which the decision will be made.
AI tools can assist with the analytical work that informs these judgments. They can summarise relevant information, identify applicable precedents, produce preliminary analyses of the options, and draft communications that convey the judgment once it has been made. What they cannot do is make the judgment itself in a way that carries the professional's accountability or reflects the full range of considerations that the professional brings to bear.
The consistent exclusion of high-judgment work from AI practice in the walkthroughs is therefore not a limitation of the AI practice but a definition of it. The AI practice is designed to reduce the time and effort invested in the preparatory, drafting, and administrative work that surrounds the core professional judgments, freeing the professional's time and cognitive capacity for the judgments themselves. An AI practice that attempted to substitute for professional judgment would not be a more ambitious version of the practice described in the walkthroughs. It would be a categorically different thing, with different properties, different risks, and a different relationship to professional responsibility.
Common Pattern Five: Build Incrementally, Deepen Deliberately
None of the five professionals in the walkthroughs arrived at their AI practice fully formed on day one. Each began with one or two workflows, used them consistently, developed an understanding of where the AI tool performed reliably and where it required more careful verification, and expanded the practice on the basis of that accumulated experience. The twelve-week build plan at the end of this section formalises this incremental approach, but the logic behind it is visible in every walkthrough.
The reason for incremental building is not caution for its own sake. It is that the specific patterns of AI assistance that deliver genuine value in a professional context are not predictable in advance: they are discovered through use. A professional who builds an integration for a task they expect to be high-value and finds that the AI outputs require more editing than the drafting they replaced has not failed to plan adequately. They have discovered something important about how AI assistance interacts with that specific task in their specific context, and they can adjust the practice accordingly. This discovery process is only productive if the practice is built in small enough increments that individual discoveries can be acted on before the complexity of the practice makes them difficult to trace.
The incremental approach also reflects the maintenance reality of AI tool integrations. An integration built before a professional fully understands the task it is intended to support is an integration whose configuration will need to be substantially revised once that understanding develops. Building integrations on the basis of specific, well-understood use cases, developed through a period of manual AI use that establishes what the integration needs to do, produces configurations that are more stable, more reliable, and less demanding to maintain.
Adapting the Walkthroughs to Roles Not Explicitly Described
The five roles in the walkthroughs cover a significant portion of the professional services landscape, but professional practice is more diverse than any five examples can fully represent. The following guidance addresses how to adapt the walkthroughs to roles not explicitly described, using both the specific examples provided and the underlying analytical logic that applies across all adaptations.
The Adaptation Method
The correct approach to adapting a walkthrough to a different role is not to identify the walkthrough that feels most similar and adopt it wholesale. It is to identify which elements of which walkthroughs are most directly applicable to the specific role, understand why those elements apply, and use the underlying logic to design the elements that need to be developed specifically for the role.
The adaptation process follows a structured sequence. First, identify the primary deliverable types of the role: the documents, analyses, communications, and other outputs that constitute the core professional work. Map these deliverable types to the task categories addressed in the model selection framework from Module 4.2 to identify which AI capabilities are most relevant. Second, identify the highest-frequency, highest-time-cost tasks within those deliverable types: the tasks that consume the most time and that the professional performs most often. These are the priority candidates for AI-assisted workflows. Third, identify the sensitivity profile of the information involved in each candidate task, using the three-tier framework from Module 4.1 to determine the appropriate data handling arrangements. Fourth, design the knowledge base structure that best supports the identified workflows, using the organising principles from Module 4.1 and the examples from the walkthroughs as structural references. Fifth, select the integration approach appropriate to each workflow, using the decision framework from Module 4.3 and the sequential build discipline from Section 1 of that module.
Residential and Commercial Real Estate Agents
A residential real estate agent's work combines high-volume client relationship management, property-specific research and documentation, and market analysis with a communication-intensive sales process. The Management Consultant walkthrough provides the most relevant structural template: the client-primary folder structure applies directly, with individual property files replacing project files within the client structure, and context documents focused on buyer preferences, search criteria, and relationship history replacing the engagement scope documents of the consulting context.
The specific AI workflows most valuable for real estate agents centre on property research synthesis, the drafting of property descriptions and marketing materials, client communication management, and the preparation of offer and negotiation correspondence. The verification requirements for property descriptions and market data are equivalent to those for consulting deliverables: factual claims about property specifications and market conditions must be confirmed against authoritative sources before they are communicated to clients or used in marketing materials.
A commercial real estate professional managing investment transactions or lease negotiations will find additional relevance in the Financial Analyst walkthrough, particularly for the workflows addressing financial analysis narrative and scenario comparison, since commercial real estate analysis involves financial modelling of investment returns and lease economics that generates the same need for AI-assisted narrative generation alongside manually verified financial calculations.
Accountants and Finance Professionals in Public Practice
An accountant working in public practice, whether in audit, tax, or advisory services, will find the Financial Analyst walkthrough most directly applicable for the quantitative and narrative work associated with client financial analysis and reporting. The specific adaptations required reflect the client-facing structure of public practice work: the knowledge base should adopt a client-primary organisation similar to the Management Consultant rather than the period-primary structure of the in-house FP&A function, and the context documents should include client business model overviews, engagement scope documents, and prior-period context notes analogous to those in the consulting and FP&A walkthroughs.
The compliance-specific dimensions of tax and audit work impose verification requirements more analogous to those in the Paralegal walkthrough than those in the FP&A walkthrough: regulatory references, statutory provisions, and technical accounting standards cited in professional reports must be independently verified against primary sources, and AI tools should not be relied upon as the primary source for regulatory or technical guidance without confirmation that the specific provisions cited are current and accurate for the applicable jurisdiction and period.
Human Resources Professionals
An HR coordinator or HR business partner will find the Operations Manager walkthrough most structurally applicable, given the shared emphasis on process documentation, FAQ management, and the systematic reduction of the supervisory burden of answering recurring questions. The SOP creation workflow applies directly to HR process documentation: recruitment processes, onboarding procedures, performance review processes, and disciplinary procedures are all candidates for the same observation-draft-validate-publish cycle used for operational SOPs.
The specific constraint that differentiates HR AI practice from operational management AI practice is the sensitivity of the information involved in HR work. Personnel matters, performance concerns, disciplinary processes, and compensation information all fall within the confidential tier of the sensitivity framework and require careful attention to data handling terms before AI tools are used in connection with them. The HR professional's AI practice should begin with the process documentation and general communication workflows that involve no personal employee data and expand toward the more sensitive areas only once the appropriate data handling arrangements have been established and verified.
Marketing and Communications Professionals
A marketing coordinator or communications professional will find the Management Consultant walkthrough most applicable for the project and deliverable management dimensions of their work: the context document approach to client or brand briefs, the deliverable development workflow, and the research synthesis process all translate directly to marketing content development. The Operations Manager walkthrough is additionally relevant for the process documentation of marketing workflows, content calendars, and campaign execution procedures that characterise the operational side of marketing management.
The specific AI applications most valuable in marketing work include content drafting and editing, audience and message research synthesis, campaign performance narrative, and the production of briefs and creative direction documents. The verification requirements for marketing AI outputs centre less on factual accuracy in the legal or financial sense and more on brand consistency, accuracy of product and service claims, and compliance with advertising standards requirements that vary between sectors and jurisdictions.
The Twelve-Week Build Plan
The twelve-week build plan provides a structured sequence for establishing a functional AI practice from no prior AI integration. The sequence is designed to allow each stage of the practice to be thoroughly understood before the next stage adds complexity, preventing the problems of premature comprehensiveness that are addressed in Section 7 of Module 4.3. The plan is a guide rather than a rigid schedule: the pace of progress through each stage should reflect the professional's developing understanding of their specific context rather than a fixed timetable.
Week One: Knowledge Base Foundation
The first week's work establishes the informational foundation without which AI assistance cannot be reliably useful. The specific tasks are the construction of the folder structure appropriate to the role, using the organising principles from Module 4.1 and the relevant walkthrough as a structural reference; the application of a consistent file naming convention to all new documents created from this point forward, with a selective retrospective application to the files most likely to be referenced in early AI interactions; and the writing of one substantive context document for the most active current engagement, client, matter, or project.
The choice of which context document to write first should be guided by two criteria: the depth and currency of the contextual knowledge it will provide to the AI tool, and the frequency with which the relevant engagement or project will be the subject of early AI interactions. The client background document for the most active client engagement, the case summary document for the most active current matter, or the policy coverage summary for the most common policy type currently in the pipeline: whichever document will be used most immediately in the first AI interactions is the right starting point.
The knowledge base establishment work in week one requires a one-time investment of focused effort that should be scheduled explicitly rather than attempted in the margins of a normal working week. A professional who attempts to build the folder structure and write a context document while simultaneously managing their regular workload will produce a structure and a document that are incomplete, inconsistent, or both. Allocating a half-day or a full day specifically to this work at the start of the programme produces a foundation that will serve the practice well for months.
Week Two: First Workflow with a Single Model
The second week introduces a single AI model and a single, specific task type for which that model will be used. The choice of model should apply the four-dimension framework from Module 4.2 to the specific task type selected. The choice of task type should be the highest-frequency, highest-time-cost task from the professional's current work that falls within the appropriate sensitivity tier for the data handling terms of the chosen model.
The purpose of limiting week two to a single model and a single task type is to allow the professional to develop a clear and specific understanding of how the AI tool performs in this specific context: what prompting approaches produce the most useful outputs, what the consistent limitations and failure modes are, what verification steps are required to catch the errors the tool reliably makes, and how the outputs need to be edited to meet professional standards. This understanding is the practical foundation of AI competence in the specific professional context, and it cannot be developed across multiple models and multiple task types simultaneously without creating confusion between the behaviour of different models and the characteristics of different task types.
The first week of working with the AI model and the first task type will produce outputs of variable quality that require substantial editing. This is expected and appropriate: the variability is the learning data from which the professional develops the prompting precision and verification practice that improves output quality over subsequent interactions. The professional who abandons the AI tool after a week of variable outputs has ended the practice before the learning phase has produced its benefit.
Week Three: First Integration
The third week adds one technical integration between the AI tool and an existing platform, selected using the five-question framework from Section 6 of Module 4.3 and tested against non-sensitive material before any connection to live professional data is established. The integration chosen should be the one that most directly reduces the friction of the task type being worked on in week two: if the week two task involves regular reference to documents in a cloud file system, a file system integration is the logical first integration candidate.
The testing discipline for the week three integration is non-negotiable: the integration is configured and tested against realistic but non-sensitive data before it is given access to any live professional material. The test should cover the normal expected inputs for the integration, the edge cases that represent the limits of those inputs, and at least one case that represents an input the integration might encounter that was not anticipated in the original configuration. Failures identified in testing are corrected before the integration is connected to live data.
Week Four: First Complete Workflow Documentation
The fourth week consolidates the practice developed over the first three weeks into a documented workflow: a written record of the specific steps the professional follows for the task type identified in week two, incorporating the model, the integration, the context documents, the prompt structures, and the verification steps that experience over the preceding weeks has established as the reliable approach to this task. The workflow documentation is written for the professional's own reference, not for external audiences, and its level of detail should be sufficient to allow the professional to reproduce the workflow consistently after a period away from it.
Documenting the first workflow explicitly serves two purposes. First, it confirms that the professional's understanding of the workflow is sufficiently complete and specific to be written down, which is a reliable indicator that the practice has reached a sufficient level of maturity to be expanded. A workflow that cannot be documented specifically enough to be reproduced from the documentation is a workflow that is still partially improvised and will produce inconsistent outputs. Second, the documented workflow serves as a reference that can be reviewed and improved as experience accumulates, creating a record of the practice's development that is valuable when the quarterly integration review described in Section 6 of Module 4.3 is conducted.
Months Two and Three: Expansion and Honest Evaluation
The second month builds on the foundation of the first month by adding a second complete workflow: a new task type, a new integration, or a combination of both, developed through the same sequence of manual use, model selection, integration building, and workflow documentation used in the first month. The context documents developed in month one should be reviewed and updated at the start of month two to reflect any changes in the professional's situation during the first month of practice.
The third month introduces the practice of honest self-evaluation: a deliberate assessment of the AI practice as a whole, examining whether the time saving and quality improvement described in the after states of the relevant walkthroughs are being achieved, whether the integrations are performing reliably and delivering the value that justified their construction, and whether the verification practices are being maintained consistently under the pressure of normal working conditions. This evaluation is not primarily about measuring success. It is about identifying the specific elements of the practice that are working as intended and those that require adjustment, and making those adjustments before the practice is expanded further.
The evaluation should produce three outputs: a list of the integrations and workflows that are delivering genuine value and should be maintained, a list of the elements of the practice that require improvement and a specific plan for improving them, and a decision about whether the practice is ready to be expanded with additional workflows or whether consolidation and improvement of the existing practice should precede further expansion.
Recognising When to Seek External Support
The AI practice described in this module is designed to be buildable by individual professionals working within the constraints of their normal working environment. Most of the setup, configuration, and maintenance work can be performed by a non-specialist professional with the guidance provided in Modules 4.1 through 4.3. There are, however, specific categories of situations where the professional's own resources are not sufficient and where seeking support from colleagues with relevant expertise is the appropriate response rather than a sign of insufficient capability.
When integration setup is taking too long. If the configuration of a technical integration between an AI tool and an existing platform is taking more than two hours of effort without reaching a working state, the problem is unlikely to be resolved by continued independent effort and is more likely to be a technical issue that requires IT expertise to diagnose. Common causes include permission configuration issues that require IT administrator access, API authentication problems that require credentials held by the IT function, firewall or network security restrictions that prevent the connection from being established, and compatibility issues between the AI tool's integration requirements and the version or configuration of the platform being connected. The professional's time is better invested in continuing to use the manual workflow while IT support resolves the technical barrier than in extended independent troubleshooting.
When AI output quality is consistently poor despite varied prompting. If AI outputs for a specific task type are consistently unhelpful or inaccurate despite repeated attempts at different prompting approaches, the most common cause is insufficient or incorrect context rather than a limitation of the model's capability. Before investing further effort in prompting refinement, the professional should review the context documents they are providing with the relevant prompts and assess whether those documents are current, accurate, and sufficiently specific to the task at hand. An AI tool producing generic outputs when prompted for client-specific analysis is almost always operating without adequate context, and improving the context documents will produce more improvement in output quality than any amount of prompt refinement.
Managing AI practice is consuming more time than it saves. If the maintenance of integrations, the management of permissions, the troubleshooting of inconsistent outputs, and the configuration work associated with the AI practice is consuming a significant portion of the professional's time without being offset by equivalent savings in the work the practice was intended to support, the practice has become over-engineered for the actual use case. The appropriate response is simplification: removing the integrations that are generating the most maintenance burden, reverting the relevant tasks to manual workflows until a more reliable integration approach can be developed, and rebuilding the practice from the manual foundation with a more selective and better-tested integration strategy.
When compliance or security concerns arise. The data handling, privacy, and regulatory considerations addressed in Module 4.1 and throughout Module 4.3 create situations where professional judgment alone is insufficient to determine the appropriate course of action. When a question arises about whether a specific category of information can appropriately be submitted to the AI tool in use, whether the data handling terms of the applicable service tier satisfy the regulatory requirements for a specific type of data, whether the use of AI assistance in a specific context creates privilege or confidentiality concerns, or whether a proposed integration would be consistent with the organisation's information security policies, the appropriate action is to seek guidance from the organisation's legal counsel, data protection officer, or IT security function before proceeding, not after. The cost of delaying an AI practice activity while compliance questions are resolved is a minor inconvenience. The cost of proceeding without resolution and subsequently discovering that the activity created a regulatory or security problem is substantially greater, and in some professional contexts carries consequences that extend to the professional's own standing and accountability.
The Practice as a Reflection of Professional Maturity
The cross-role lessons and the build plan presented in this section converge on a perspective about AI-assisted professional work that is worth stating explicitly. The highest expression of AI competence in professional practice is not the most comprehensive or the most technically sophisticated AI integration. It is the most thoughtful and most deliberate practice: one that knows precisely where AI assistance adds value, maintains the knowledge base infrastructure that makes that assistance reliable, applies the verification discipline that keeps professional standards intact, and preserves the human judgment that professional accountability requires.
The five professionals described in the walkthroughs are not defined by their AI tools. They are defined by the quality of their professional judgment, the depth of their domain knowledge, the strength of their relationships with clients and colleagues, and the standards they hold their work to. The AI practice serves all of these properties by recovering the time and reducing the cognitive burden of the work that does not require judgment, knowledge, relationships, or professional standards, so that the professional's best capacity is available for the work that does.
This is the purpose of Stage 4 as a whole: to equip professionals not with a collection of tools and techniques but with the understanding needed to build and maintain a personal AI practice that is genuinely theirs, genuinely serves their work, and genuinely reflects the professional standards that define what good work in their field means.
A Critical Note on Session Management
Professionals must maintain strict boundaries between conversational topics by initiating a new chat session for each distinct task. Mixing unrelated subjects within a single continuous chat degrades the model's contextual focus and introduces reasoning errors. A dedicated session for each specific workflow ensures the AI tool applies only the explicitly relevant context to the immediate problem. Furthermore, isolating tasks into individual chat threads significantly improves the long-term searchability and navigability of the user's historical chat records.
Please note that while this principle of context hygiene applies universally across all AI workflows, it lacked a structurally appropriate placement within the preceding module sections and is therefore appended here as a foundational best practice.