article
The institutional knowledge paradox
Why investing in AI knowledge retrieval cannot replace the people, ownership, and dedicated time needed to create reliable institutional knowledge.
We are eliminating the people who create institutional knowledge, then spending fortunes on AI systems that promise to retrieve it.
It’s refreshing to see companies like Supabase investing in a proper Knowledge Base.
Perhaps, once the AI and SaaS hype settles down, companies will rediscover the value of in-house technical writers and other specialized roles.
We are currently living through a reductionist era in software engineering. Technical writers, Scrum Masters, project managers and several other specialized professionals are increasingly considered dispensable.
The prevailing assumption is simple: Someone else on the team can do it. They just need to stretch a little.
But when specialized responsibilities are distributed among people whose primary deliverables lie elsewhere, standards inevitably suffer. Ownership becomes diluted, and prioritization disappears.
Knowledge bases are a classic casualty.
Why would a developer spend time documenting an existing implementation when another feature, another sprint commitment, or another piece of AI-generated code is waiting to be delivered?
There was a practical advantage to having dedicated technical writing teams that is often overlooked.
The period immediately following a major release could serve as a natural hiatus from relentless feature development.
While technical writers worked on documenting the newly delivered functionality, developers could turn their attention to less urgent but important engineering work: static code analysis findings, code cleanup, minor improvements, technical debt and implementation reviews.
It was a relatively relaxed period when smaller, lower-priority tasks could finally receive attention without the pressure of another major delivery.
Documentation and engineering housekeeping could progress in parallel, each handled by people with the appropriate expertise.
Today, with leaner teams and continuous delivery expectations, such breathing spaces are increasingly disappearing. Documentation gets postponed, technical debt accumulates, and everyone moves straight to the next commitment.
Even a somewhat outdated knowledge base can provide a useful starting point for new employees, developers and customers. Without one, institutional knowledge survives through informal conversations, repeated explanations and the memories of a shrinking group of experienced employees.
And now comes the irony.
Companies are enthusiastically investing in AI assistants, RAG pipelines and enterprise knowledge agents while simultaneously reducing investment in the people and processes that create reliable knowledge in the first place.
AI can retrieve, summarize and reorganize information. But it cannot reliably reconstruct years of undocumented architectural decisions, operational exceptions and business context out of thin air.
Perhaps the next great innovation in enterprise software will be something remarkably old-fashioned:
A well-maintained knowledge base, written by people who understand the system, with clear ownership and dedicated time to keep it useful.
Not everything that looks like overhead is waste. Sometimes, it is the very infrastructure that keeps an organization functional.
#SoftwareEngineering #KnowledgeManagement #TechnicalWriting #AI #EngineeringLeadership