Big day for wikis. Google and Factory both had wiki drops today. Open Knowledge Format (OKF) from Google and AutoWiki from Factory.
Cognition already had DeepWiki, and Karpathy started an LLM wiki fad recently. The problem is that no one has ever asked for—or liked—a work wiki, in my humble experience. Maybe agents will like wikis more?
There's nuance between wiki <> internal documentation <> knowledge base.
A wiki is a crowdsourced knowledge base. Great for Wikipedia, not so great for companies. Within a company, crowdsourced wikis just don't work.
Despite the nuances on the merits of a wiki vs. a knowledge base, it's all about context for agents. And people are excited about documentation again—which is great.
When my career started in 2010, the prevailing narrative was that documentation was futile.
“The UI should be intuitive! The code should be self-explanatory!”
Then companies like Stripe and Vercel showed how great docs can elevate brand and developer experience to drive PLG. Many others quickly followed. Great docs became a competitive advantage and a badge of honor.
Stripe doesn’t make money on devs until their integration is functioning and payments are flowing through Stripe. Shaving off every minute of integration time at scale through docs yields more revenue that starts compounding faster.
Every company has some version of this incentive. The faster customers can onboard, the faster you get revenue. The easier customers can evolve their integration, the easier you can upsell and expand with that customer.
The last few years I saw this cycle again. Docs were kinda dead. The narrative post-Claude / ChatGPT was that docs were irrelevant. “LLMs can just read the code and provide answers on demand!”
But now codebases are larger and messier than ever. And they’re growing exponentially. It’s clear that context is everything for deploying agents—for accuracy, speed, token efficiency, and cost.
Documentation is the best representation of context for agents. The same rumbles I heard from companies in 2018 wanting high-quality docs are picking up again. Companies need great docs more than ever—for context-starved agents in addition to humans.
Now the components are less about design, UX, and information architecture as in 2018, and more about completeness, accuracy, and agent experience.
Completeness, accuracy, and agent experience yield advantages in speed, efficiency, and cost. Time to sharpen the pencils again, because the alpha is in the docs.