Yet another post on documentation. I have written quite a few articles about the importance and process of documentation, both for work and private matters. This post explores why documentation remains crucial, especially in the age of AI.
Documentation as Cognitive Extension 🔗
First and foremost, documenting serves as a form of a “cognitive offloading”. Our brains have limited working memory capacity. We simply cannot memorize everything and retrieve it accurately much later when needed. Documentation helps to overcome this limitation by externalizing the information. We then only need to remember a metadata of what it is about and where to find it when necessary.
Documentation as Communication Infrastructure 🔗
Moreover, documentation functions as a communication tool. As a Product manager, structured documentation helps me log product development and be on the same page with my stakeholders. This became especially critical during the shift to remote working where synchronous communication was no more possible, documentation emerged as the key for effective communication and collaboration.
The LLM Knowledge Paradox 🔗
The Large Language Models (LLMs) possess enormous knowledge. They are trained on terabytes of text scraped (hopefully only) from publicly available sources such as websites, books, articles, code repositories, forums, and social media platforms. Through this training, they acquire broad general knowledge about language, facts, and common reasoning patterns.
However, there is a big caveat. The knowledge of LLM is confined to what exists in the data it is trained on. It does not include private, proprietary, or real-time information, unless specifically incorporated via external tools or via fine-tuning.
Bridging the AI Knowledge Gap 🔗
If we want to leverage AI for specific use cases, we need to supplement it with external knowledge. Fortunately, several methods now exist for this purpose:
- Retrieval-Augmented Generation (RAG)
- Multi-Context Processing (MCP)
- Expanded context window capabilities
- Fine-tuning on domain-specific data. Most crucially, these approaches require high quality documentation to feed the LLM with relevant context.
The Enterprise AI Transformation and Trust 🔗
According to diverse studies (example: McKinsey Survey on the state of AI), more than 70% of the enterprises have adopted Generative AI in some capacity. Companies are integrating AI components to all of their products to support business optimizations, customer support, product offerings, R&Ds and many more business critical functions.
With such a big bet on AI, “trust” becomes paramount. How do we make sure that AI is serving the company with the best of its skills and knowledge? While AI’s technical capabilities has proven itself to be more than well-qualified, its knowledge base requires careful cultivation.
For an AI language model to provide accurate and complete information, it needs access to relevant organizational knowledge, including:
- Brand guidelines
- Product requirement documents
- SharePoint files
- Spreadsheets
- Codebases
- Recent company news
- Historical customer cases
- Financial reports
Documentation Advantage 🔗
With access to up-to-date and domain-specific documentation, AI can:
- Answer questions more accurately and reliably
- Reduce the likelihood of generating incorrect information
- Understand and reflect company language, style, and tone—making interactions more authentic than robotic
- Provide better customer support by being grounded in company knowledge.
Conclusion 🔗
Communication remains the cornerstone of success. Reflecting on problems, understanding challenges, and developing solutions all stem from effective communication. Documentation enables knowledge transfer across time and between people and machines without loss in quality.
The pressing question now becomes: how can organizations build efficient documentation workflows that serve both human and AI consumers of information?