The Source of Truth Problem
CSMs troubleshooting an issue, SEs fielding a customer question about the product, or really anyone at the company who doesn't know every product and how it works all hit the same wall: there is no centralized source of truth.
Andy points to the GTM Help channel as the clearest symptom. The same questions get asked over and over, not because there's no answer, but because there's no clearly right answer. People have no good way to know whether what they found is correct, whether a Confluence page is accurate, or whether a random Jira ticket is still up to date.
Before AI, the only real option was for everyone to write a doc and then search through all of them hoping to land on the best match. The problem stayed unsolved because there was never a good way to centralize every source.
Reverse-Engineering the Truth
What changed is that LLMs are now very good at synthesizing sources. Instead of asking people to hunt for the best-matching doc, the Product Knowledge Base takes all the docs and all the sources and reverse-engineers what the truth actually is.
From there it can do the harder thing: work out the right way to answer a given question, rather than just surfacing the closest document.
From GitHub to Confluence
The first iteration lived in GitHub, essentially a regular file system, and many people will remember that version. The team has since moved most of that knowledge into Confluence, because it lets humans supervise it.
Today there is an entire Confluence space holding almost all of Abnormal's product knowledge. It's organized by product, so Email Security, for example, has its own set of pages beneath it. Everything is structured to be both human-readable and AI-retrievable, optimized for both at once.
That dual design is the point. The future Andy sees is one where humans and AI both read and edit the same knowledge base.
Launching a Product Instantly
A recent example shows why that matters. Abnormal just released a new product umbrella called Identity Attack Protection, which Tanisha has been working on. With a Black Hat demo coming up and plenty of people still unsure how it works, she texted Andy to ask if it could go into the Product Knowledge Base.
She sent over some materials, and Andy generated all the pages instantly. From there she can come in and edit them directly. The source of truth everyone relies on can now change in minutes, instead of going through a long process to get information out to reps.
Answers Approved by Experts
The first way the knowledge base speeds things up is a simple Cowork skill that lets anyone ask it questions. It hasn't been shared widely yet, but the intent is to put it in front of reps: they open GTM Help, ask a question, and the skill pulls straight from the Product Knowledge Base to give an answer that's already approved by subject matter experts.
Freshness is what makes it work. When an expert like Tanisha changes a page, the answer updates instantly the next time someone asks. There's no lag between updating the knowledge and reps getting the new answer.
Pages That Write Themselves
The second accelerator flips contribution around. Instead of manually editing pages, someone can hand the work to AI. Andy built an AI-facing guide that tells the model how to contribute to the Product Knowledge Base, covering citation, accuracy guardrails, and organization.
With that in place, a contributor can give a simple prompt, something as light as "can you fill in this page," and the skill follows the guide to produce a page that's ready for GTM reps to use. Andy did exactly this with Tanisha to fill in a feature on the Identity Attack Protection page.
That speed matters because the company is shipping so many things that no one can keep track of all of them. The Product Knowledge Base is how internal knowledge keeps pace with how fast Abnormal is shipping.
What's Next
Andy is excited to release the Product Knowledge Base more broadly. The plan is to make it available in Cowork, through a Slack bot, and across other services, so it reaches all of Abnormal's enablement surfaces rather than living in one place.
The bigger idea is a knowledge base that humans and AI maintain together. Experts stay in control of what's true, AI handles the drafting and the retrieval, and reps get current, cited answers wherever they already work.
Andy called this update number 1.5, which is a fitting label. The foundation, a supervised and synthesized source of truth, is in place, and the surfaces that put it in front of people are still expanding.
