The Decay Problem
A personal knowledge base is not a static asset. It is a living system that reflects the current state of your professional work, and like any living system, it requires regular attention to remain healthy. The files you organised carefully at the start of an engagement accumulate companions: documents saved in haste, renamed in shorthand, stored on the desktop because there was no time to file them properly. The context documents you wrote at the beginning of a project drift further from reality with each week that passes without an update. The decision log that was current in January becomes an incomplete record by March and a potentially misleading one by June.
This drift is not a consequence of carelessness. It is a structural feature of busy professional work. The moments when documents are created are also the moments of highest pressure, when the work itself demands full attention and the administrative act of proper filing feels like a secondary concern. Over time, the gap between how the knowledge base was designed and how it actually looks widens, and the AI tools that depend on that knowledge base begin to produce outputs that reflect the accumulated disorder rather than the intended organisation.
The compounding nature of this problem deserves emphasis. A knowledge base that has drifted for three months is not simply three months out of date. It is, in important respects, worse than no knowledge base at all. When an AI tool has access to no context, it operates in acknowledged uncertainty and produces responses that are explicitly general. When an AI tool has access to a context document that was accurate three months ago but no longer reflects the current situation, it produces responses that are confidently specific but wrong. It references a key contact who has left the engagement. It treats a decision that was subsequently reversed as though it still stands. It frames the current phase of the work using the objectives of a previous phase. The professional receiving these outputs must identify the errors, trace them back to the outdated context document, and correct the AI's orientation before useful work can proceed. This costs more time than the maintenance that would have prevented it.
The Case for a Fixed Weekly Practice
The solution to the decay problem is not a periodic large-scale audit of the knowledge base. Large-scale audits are time-consuming, psychologically costly, and tend to be postponed until the disorder has become serious enough to justify the effort, at which point the effort required is substantial. They treat maintenance as an exceptional activity rather than a routine one, which means the knowledge base spends most of its life in a state of partial disrepair.
The more effective approach is a fixed, brief, weekly practice. Fifteen minutes at a consistent point in the working week, applied to a defined set of maintenance tasks, is sufficient to keep a well-structured knowledge base current. The consistency of the practice is more important than any individual instance of it. A knowledge base maintained with fifteen minutes of attention every Friday afternoon for a year will be in substantially better condition than one that receives two hours of attention every three months.
The psychological mechanism behind this is simple. Fifteen minutes of maintenance at the end of each week addresses the accumulation of that week only. The volume of disorder to be managed is small, the tasks are familiar, and the effort required is low. Two hours of maintenance after three months of drift requires confronting a much larger accumulation, making decisions about a greater number of files, and reconstructing context that has partially faded from memory. The weekly practice prevents the problem from reaching the scale at which it becomes genuinely burdensome.
The choice of when to conduct the weekly practice matters in a secondary way. The end of the working week is the natural moment for this activity because the work of the week is complete and can be assessed as a whole. Files that were created or modified during the week are still fresh in memory. Decisions that were made can be recorded accurately in the decision log. Context changes are recent enough to be recalled without difficulty. The knowledge base, maintained at this point in the weekly cycle, reflects the current state of work with minimal reconstruction effort.
The Four Tasks of Weekly Maintenance
The weekly maintenance practice consists of four defined tasks. Each task addresses a specific type of disorder that accumulates during the working week. Together, they address the full range of maintenance needs for a well-structured knowledge base.
Filing displaced documents. During the working week, documents accumulate in locations that are outside the folder structure: the desktop, the downloads folder, email attachments that were opened but not saved, draft files created in temporary locations. The first maintenance task is to collect these documents and move them to their correct locations within the folder structure, applying the naming convention described in Section 2 to any that were saved without proper names. This task takes the longest of the four when the week has been busy, but its scope is bounded by the week's work and rarely requires more than five minutes of the allotted time.
Updating context documents. During the working week, things change. New people join an engagement. Key contacts on the client side are replaced. The scope of a project is adjusted following a review meeting. A decision is made that supersedes an earlier one. A term enters the working vocabulary of the team that has not been recorded in the glossary. The second maintenance task is to identify any changes of this kind that occurred during the week and update the relevant context documents to reflect them. This task requires judgment rather than mechanical filing, because it involves assessing which changes are significant enough to record. The guiding principle is to record any change that would affect the AI tool's responses if it were not documented: any change to who the key people are, what the current scope is, what has been decided, or what terminology is being used.
Archiving completed work. As work moves from active to complete, the documents relating to it should be moved to the Archive subfolder described in Section 3. The third maintenance task is to assess whether any work has been completed during the week and to archive the documents associated with it. A closed phase of a project, a resolved claim, a filed court document, a delivered and approved report: all of these represent completed work whose associated documents should be moved out of the active folder and into Archive. This keeps the active working set focused on current work and ensures that AI tools searching for relevant material are not surfacing completed work alongside active material as though both were equally current.
Correcting names applied in haste. During the working week, under deadline pressure or simply in the routine pace of work, files are often saved with names that do not follow the naming convention. Draft documents get saved as Draft or New Document. Meeting notes get saved by date alone. Revised versions accumulate suffixes that are meaningful in the moment but inconsistent with the established convention. The fourth maintenance task is to identify any files saved during the week with non-standard names and rename them according to the convention. This is the briefest of the four tasks when the weekly practice is consistent, because the number of files to be corrected is limited to a single week's worth of work.
Reading the Signals That Maintenance Has Slipped
Even with a consistent weekly practice, there will be periods when the practice is interrupted: periods of unusually high workload, illness, travel, or the particular intensity that accompanies a deadline or a crisis. When maintenance slips, the knowledge base begins to show signs of disorder that are recognisable once you know what to look for. These signals are worth understanding because they allow you to identify the need for corrective action before the disorder becomes serious.
You are spending more than two minutes locating a specific file. In a well-maintained knowledge base with consistent naming and a clear folder structure, any specific document should be locatable within a few seconds. A search by client name, project name, document type, or date should return the correct file immediately. When finding a specific file requires extended searching, browsing through multiple folders, or opening documents to determine their contents, this indicates that naming conventions have not been applied consistently, that the folder structure has been bypassed in favour of saving to convenient default locations, or that both problems have accumulated together.
AI tools are referencing outdated information. When AI outputs reference people who are no longer involved in an engagement, describe the scope of work in terms that were accurate in an earlier phase, treat reversed decisions as current, or use terminology that the team has since revised or abandoned, the cause is almost always context documents that have not been updated. This signal is particularly important because it indicates that the knowledge base is actively producing incorrect AI outputs rather than simply failing to produce correct ones.
Individual folders have grown beyond a manageable size. When any single folder contains more than approximately twenty files without subfolders, navigation becomes cumbersome and AI tool searches become less precise. This is a signal that either the folder structure needs additional subcategories, that archiving has not been kept current, or that naming conventions have not been applied consistently enough to make the folder's contents distinguishable at a glance.
Context documents have not been reviewed in more than a month. For any active engagement, matter, or project, a context document that has not been reviewed in more than a month is unlikely to be fully current. Professional work changes continuously, and a month is a sufficient period for significant changes in personnel, scope, decisions, and terminology to have accumulated. A context document review need not result in changes to every session, but the act of reviewing it confirms that its contents still reflect reality and ensures that any changes that have been overlooked are captured before they produce incorrect AI outputs.
Maintenance as Professional Discipline
It is worth addressing a question that reasonable professionals sometimes raise about the weekly maintenance practice: whether the time it requires is justified by the benefit it produces. The calculation is straightforward.
Fifteen minutes of maintenance per week amounts to approximately twelve hours over a working year. Against that cost, the benefit is a knowledge base that reliably supports AI-assisted work throughout the year, that does not require periodic large-scale reconstruction, and that does not generate the kind of confidently incorrect AI outputs that require time to identify, diagnose, and correct. The cost of a single significant error traced to an outdated context document, whether that error is a client communication sent with incorrect relationship context, a coverage analysis based on a superseded policy position, or a financial narrative that misrepresents the current state of the business, is likely to exceed twelve hours of remediation.
The deeper point is that knowledge base maintenance is not separable from professional competence in AI-assisted work. A professional who maintains an excellent knowledge base is a more effective user of AI tools than one who does not, in the same way that a professional who maintains an organised and current file system is a more effective practitioner than one who does not. The maintenance practice is not an addition to professional work. It is a component of it.