The Knowledge Half-Life: Why Your Wiki Is Already Out of Date and a Company Brain Never Is

Open your company wiki and pick a page at random. Check the last-edited date. If it says anything older than a few months, you are looking at a description of a company that no longer exists: a tool that has since been swapped out, a process that has since changed, an approval step that a reorg quietly deleted, an exception that only made sense while a now-departed colleague was handling it. The page is not wrong because someone was careless. It is wrong because it stopped observing the work the moment it was saved.

This is the knowledge half-life, and it is the quiet reason most corporate knowledge projects fail. Every document, SOP, SharePoint site, and Confluence space starts decaying the instant it is written, because nobody is paid to keep it current and it has no way to watch what the team actually does. You can hire a knowledge manager, run a documentation sprint, mandate quarterly reviews, and the decay curve barely bends, because you are fighting physics with willpower.

A company knowledge base built on AI works the other way round. Instead of a store of documents that ages, a Company Brain is a living memory that AI employees build and use every day, correct through feedback, and connect to your live systems. It does not need a review cycle to stay current, because staying current is what it does when the work runs through it. This guide is for the Geschaeftsfuehrer, operations lead, or IT director who is tired of pouring effort into documentation that is stale before it is read, and wants the mechanism that actually keeps company knowledge fresh.

The Knowledge Half-Life Is a Real Thing, Not a Metaphor

The idea has a measured history. The economist Fritz Machlup coined the phrase half-life of knowledge in the 1960s, borrowing from radioactive decay to describe how long it takes for half of what is known in a field to be superseded or shown wrong¹. The rate is not a constant. It depends on how fast the field moves.

• The decay rate varies by field - it is a property of a discipline at a moment in time, driven by the pace of change, not a fixed number².
• Engineering knowledge halved faster over time - estimated at around five years in the early 1990s, falling to roughly two and a half years a decade later, and shorter still in IT².
• Medical knowledge decays in two to three years - fast enough that some of what a student learns is outdated by graduation².
• Company knowledge decays faster than a whole field - a specific process, price, contact, or exception can be wrong within weeks, because a single reorg or tool change invalidates it instantly.
• The document does not know it is wrong - unlike a person, a static page never notices the world changed, so it keeps confidently stating the old answer.

This reframes the whole knowledge-management debate. The question is not how do we write better documents, but how do we stop depending on artefacts that decay.

Why Every Wiki Decays the Moment It Is Written

Blaming people for not updating the wiki is the wrong diagnosis. The decay is structural. A document store has four built-in flaws that no amount of discipline removes.

1. It captures a snapshot, not a process - a page records how the work was done on one day; the work keeps changing and the page does not move with it.
2. Updating it is always the losing task - documentation is unpaid, unmeasured overhead that competes with real work, and real work wins every time.
3. It never observes the work - a wiki has no way to see that a step was skipped or a tool was replaced, so it cannot notice it has gone wrong.
4. Nobody owns freshness - authorship is diffuse, so when reality changes there is no single person whose job is to correct the record.
5. Contradictions accumulate silently - three versions of a policy sit in three places, and the reader has no way to know which one is current.

The 2026 wave of AI tooling made this failure impossible to ignore, because feeding decayed documents to an AI produces confident wrong answers at scale. An analyst covering SharePoint’s 25th anniversary put it bluntly.

A document store is not a neutral asset waiting to be searched. Once it decays, it is an active source of wrong answers, and adding AI on top amplifies the mistake rather than fixing it.

For a deeper look at why the raw material itself is the bottleneck, see our companion piece on the unstructured data problem behind every AI project.

What Stale Knowledge Actually Costs

The cost of decayed knowledge does not show up as a line item, which is exactly why it goes unmanaged. It hides in time, errors, and lost expertise. The numbers, once you add them up, are large.

• A fifth of the week goes to searching - McKinsey found knowledge workers spend around 20 percent of the working week hunting for internal information or the right colleague to ask¹⁰.
• Hours a week are lost to re-creation - workers lose roughly 5.3 hours a week waiting for information from colleagues or recreating knowledge that already existed somewhere¹².
• Poor knowledge sharing has a price tag - Panopto and IDC research puts it at about 4.5 million dollars a year per 1,000 employees, and around 47 million a year for a large enterprise^11,13.
• Poor information management costs per head - IDC estimates roughly 5,700 dollars per worker per year in wasted effort from not finding the right information¹⁵.
• Most expertise is undocumented - an estimated 42 percent of role-specific knowledge exists only in the head of the person currently doing the job¹².
• Wrong answers cost more than slow ones - a stale page that sends someone down the wrong path costs rework, not just search time, and erodes trust in the whole knowledge base.

A worked example makes the leak concrete. Consider a 500-person mid-sized company.

These figures overlap and should not simply be summed, but the direction is unambiguous: stale, scattered knowledge is one of the largest unmanaged costs in a mid-sized company. Our breakdown of what having no Company Brain really costs works these numbers through in detail.

Why This Broke Into the Open in 2026

The knowledge half-life is not new, but 2026 forced it onto the agenda. Three things happened at once, and they turned a background nuisance into a strategic problem.

1. SharePoint turned 25 and Microsoft changed its story - the platform that defined the document-store era now serves more than a billion users a year and takes in around two billion files a day, and Microsoft used the anniversary to reframe knowledge as something that must be activated, not merely stored³.
2. Microsoft shipped a context layer - Work IQ became generally available in June 2026 as a layer that gives agents continuously updated business context rather than static files, an explicit admission that a folder of documents is not enough^4,23.
3. AI made stale knowledge dangerous, not just wasteful - feeding decayed documents to an assistant produces confident wrong answers at scale, so the cost of staleness jumped from slow searches to bad decisions.
4. Agents need current context to act - Gartner projects 40 percent of enterprise applications will feature task-specific AI agents in 2026, and agents that act on stale knowledge cause real errors, not just bad search results²¹.
5. Demographics keep draining tacit knowledge - in Germany, Bitkom and Fraunhofer warn that retirements are a leading cause of knowledge loss, with age-related departures cited by 42 percent of firms as a driver, against a backdrop of more than 100,000 unfilled IT roles^17,18.

Microsoft frames the destination as software that reasons over current context rather than files.

The strategic point holds whether or not you run Microsoft: the future of company knowledge is a memory that stays current, and the wiki era is ending.

Document Store vs Living Company Memory

The difference between a wiki and a Company Brain is not a feature list. It is a difference in what the thing fundamentally is. One is a passive archive you read from. The other is an active memory the work writes to.

The core distinction

• A document store holds what someone wrote - it is only as current as the last manual edit, and only useful if a person finds and trusts the right page.
• A Company Brain holds what the company does - it is written to by the ongoing work, so it reflects current practice without a documentation step.
• A store is read; a memory is used - people consult a wiki occasionally, but AI employees use the Company Brain on every task, which is what keeps it exercised and current.
• A store forgets; a memory learns - a wiki cannot improve on its own, while a Company Brain gets more accurate with every correction fed back into it.
• A store is a silo; a memory is connected - it draws from email, chat, CRM, and ERP as live sources rather than a snapshot someone pasted in.

For the tightly related case of a specific platform, our piece on moving beyond a SharePoint knowledge base covers the migration question directly.

How a Company Brain Stays Current

A Company Brain does not beat decay by being written more carefully. It beats decay by never being finished. Three mechanisms keep it fresh, and all three are automatic side effects of using it, not extra chores.

The three freshness mechanisms

1. It observes the work - because AI employees perform the tasks, the current way of doing them is captured as a by-product; there is no separate documentation step to skip.
2. It learns from feedback - every correction a person makes updates the shared memory, so a mistake fixed once does not recur, and the memory tracks how the work actually evolves.
3. It reads from live systems - prices, stock, policies, and records come from the source of truth in your CRM, ERP, and files, not a figure someone typed into a page last quarter.

Microsoft describes its own context layer in exactly these terms: continuously processed rather than statically stored.

Our deep dive on the feedback loop that makes AI employees better every week covers the learning mechanics in full.

The Feedback Loop That Beats the Decay Curve

A wiki has one curve: down. From the day it is written, its accuracy falls as the world changes around it. A Company Brain has the opposite curve, because feedback pushes accuracy up faster than change pushes it down. The two curves are the heart of the argument.

• Decay is continuous, feedback is corrective - the world keeps changing a document toward wrong, while feedback keeps pulling the memory back toward right, and use makes the correction frequent.
• Every task is a maintenance event - because the memory is used on real work daily, it is checked and corrected daily, instead of reviewed rarely.
• Corrections are shared, not siloed - when one person fixes an error, the whole system benefits, unlike a person who learns something the wiki never records.
• Contradictions get resolved - conflicting versions surface as disagreements the loop can settle, rather than sitting quietly in three folders.
• The advantage compounds - a memory that improves weekly pulls further ahead of a store that decays weekly, and the gap is a moat competitors cannot copy by buying software.

This is also why bolting AI onto a stale store disappoints. Without the correction loop, the AI inherits the decay.

Knowledge That Survives Turnover

The cruelest form of decay is not slow drift; it is the cliff edge when an experienced person leaves and takes the undocumented knowledge with them. A wiki was supposed to prevent this and does not, because the hardest knowledge was never written down in the first place.

• The valuable knowledge is tacit - the how-we-actually-handle-this lives in people, is highly experiential, and is hard to document in any detail²⁰.
• Most role knowledge is in one head - an estimated 42 percent of role-specific expertise is known only by the person currently doing the job¹².
• Retirement is a leading cause of loss - German firms name age-related departures as a top driver of knowledge loss, a pressure that only grows as boomers retire¹⁷.
• Documentation sprints miss the exceptions - people write down the happy path and forget the edge cases that are the real value, so the wiki captures the least important part.
• A Company Brain captures knowledge in use - because the memory is built from doing the work, the exceptions get captured as they are handled, not in a one-off writing exercise.

This is the strongest argument for starting before you need to. Our guide on capturing what retiring staff know before they leave works through the timing.

How to Move From a Static Wiki to a Company Brain

You do not fix a decaying wiki with a bigger documentation project; you make knowledge a by-product of the work. The move is practical and starts in one department, not with a company-wide migration.

Phase 1: Pick the function and connect the systems (Weeks 1-4)

1. Week 1: Choose a high-decay, high-volume function - where knowledge changes often and searches are frequent, such as customer service, order handling, or internal IT support.
2. Week 2: Find where the current truth actually lives - identify the live systems, the trusted people, and the wiki pages that are still accurate, so the memory starts grounded.
3. Week 3: Connect an AI employee to the live sources - email, chat, CRM, ERP, and file stores, so answers come from the source of truth, not a snapshot.
4. Week 4: Seed the memory and set feedback rules - load the good existing documents as context and agree how corrections are captured, so the loop works from day one.

Phase 2: Run it in the work and let it learn (Weeks 5-8)

5. Week 5-6: Run in parallel with the team - the AI employee handles routine work alongside people, who correct it, and the Company Brain captures the current way of working.
6. Week 7: Resolve the contradictions it surfaces - where the memory finds conflicting versions of a process, settle them once, and retire the stale pages.
7. Week 8: Measure freshness and accuracy - track how often answers are correct, how fast corrections stick, and how much search time drops against a baseline.

Phase 3: Shift the source of truth and expand (Weeks 9-12)

9. Week 9: Make the memory the reference, not the wiki - point the team to the Company Brain for current answers and keep the wiki only for archival reference.
10. Week 10-11: Capture the exceptions in use - as edge cases arise, they are handled and fed back, so the hardest knowledge accrues where it is used.
11. Week 12: Report and pick the next function - present the freshness and time-saved gains, then repeat the cycle in the next high-decay department.

The change-management side matters as much as the technical side. Our guide on onboarding your team when AI employees join covers making the shift stick.

How Superkind Fits

Superkind builds AI employees for the Mittelstand that carry routine work and, in doing so, build a Company Brain: a living memory of how your company actually operates. The point is not to replace your wiki with a shinier wiki. It is to make knowledge a by-product of the work, so it stays current without anyone maintaining it by hand.

• Company Brain, not a document store - the memory is built and used by AI employees every day, so it reflects current practice rather than an old snapshot.
• Connects to your existing systems - email, Teams, SharePoint, CRM, and ERP feed one live memory instead of a dozen decaying silos, with no rip-and-replace.
• Learns from daily feedback - every correction updates the memory, so an error fixed once does not repeat and accuracy compounds week over week.
• Grounded in live data - answers use the current figure from the source system, not a number typed into a page months ago.
• Captures tacit knowledge in use - the exceptions and judgement calls get recorded as they are handled, before the expert who knows them leaves.
• Process-first discovery - we map how your team actually works before building, so the memory fits your workflows rather than a generic template.
• Department by department - we prove one high-decay function, measure the freshness gain, then expand, so risk stays contained.
• Compliant by design - data stays in your infrastructure, access is controlled, and the memory is observable for DSGVO and EU AI Act record-keeping.

To see how the same memory keeps its value as it grows, read our piece on a Company Brain that stays under your control.

Decision Framework: Is Your Knowledge Decaying Faster Than You Think?

Not every company needs to rip out its wiki tomorrow. Use these signals to judge how badly the knowledge half-life is already costing you and where to start.

The goal was always right. What changed is the mechanism: sharing knowledge now means a memory that stays current, not a store that decays.

Frequently Asked Questions

Henri Jung

Co-founder of Superkind, where he helps SMEs and enterprises deploy custom AI employees that actually fit how their teams work. Henri is passionate about closing the gap between what AI can do and the value it creates in real companies. He believes the Mittelstand has everything it needs to lead in AI - it just needs the right approach: knowledge that stays current instead of decaying, and a company memory that survives turnover rather than walking out the door.