Latest / Future of Work Tech with Fexingo: Remote Tools, AI Productivity, and Workplace Software / How AI Is Automating Your Internal Knowledge Discovery
Transcript
- Lucas: Luna, I want to talk about something that's quietly become one of the most debated features inside enterprise software: the shift from 'pull' to 'push' in how we find information at work. Luna: You mean instead of me going to search for a document, the document comes to me? Lucas: Exactly. And it's not just search getting smarter. We're seeing a whole category of AI tools that monitor what you're working on — your calendar, your chat threads, the documents you have open — and then proactively surface relevant internal knowledge. Like, you're in a meeting about a product launch, and suddenly a notification pops up linking to last quarter's post-mortem and the updated pricing sheet. Luna: That sounds incredibly useful. Also potentially creepy. Lucas: Right. That's the tension. But the use case is compelling. I was talking to a product manager at a large financial services firm — I won't name them because they're not public about this yet — and they rolled out a tool from a startup called Mem, which is basically an AI knowledge base that watches your Slack activity. The PM said new analysts used to spend their first two weeks just trying to find which internal documents actually mattered. After they deployed this, the time to find relevant documents dropped by over 60 percent. Luna: Sixty percent? That's enormous. But how does it decide what's relevant? Is it reading my private messages? Lucas: That's the million-dollar question. The way these tools generally work is they index all your company's internal documents — wikis, Google Docs, Notion pages, chat archives — and build a knowledge graph. Then they use a combination of your calendar events, your recent chat activity, and the document you're currently viewing to infer context. They don't read the full contents of every private message — at least the responsible ones don't — but they do look at channel names, message topics, and which documents people are linking to. Luna: So it's more about signals than content. Still, that's a lot of data being aggregated. Lucas: It is. And companies are aware of the privacy concerns. Microsoft's Viva Topics, for example, gives admins control over what sources are indexed, and it doesn't surface content from private chats. But the more aggressive tools — like the one that financial firm used — do tap into Slack DMs. The trade-off is effectiveness. The PM told me that the most valuable suggestions came from patterns in private conversations, because that's where people actually discussed which documents were critical. Luna: So either you get the full benefit and accept the surveillance, or you get a watered-down version that's less useful. Lucas: That's the spectrum right now. But I think there's a third path, which is that employees might be willing to opt in if they see clear value. Like, if the tool explicitly shows you what it's using to make its suggestions — 'We suggested this document because you're in a meeting called Q4 Budget Review' — that transparency might build trust. Luna: That's a smart design principle. Show your work, literally. Lucas: Yeah. And some companies are already doing that. There's a tool called Guru that offers a 'context card' that explains why a piece of knowledge was surfaced. Another one, Slab, uses AI to tag documents with 'read this if you're working on X' based on keywords from your active projects. Luna: This also changes something deeper, which is how we think about institutional memory. If the AI is constantly resurfacing old documents, does that mean we stop remembering things ourselves? Lucas: That's a great point. There's a concept in cognitive science called 'transactive memory' — the idea that we offload remembering to other people or tools. Your partner remembers your aunt's birthday, you remember the Wi-Fi password. With these tools, the AI becomes a kind of external memory. The risk is that if the tool goes down or you switch jobs, you lose access to that memory. But the upside is that institutional knowledge doesn't disappear when someone leaves. Luna: Speaking of upsides, I was reading a case study about a company that used a tool called Coda Brain. They're a mid-size tech firm, about 500 employees, and they had a problem where salespeople kept asking engineering the same questions about product capabilities. They set up Coda Brain to scan their internal docs and chat history, and it started proactively sending sales reps a 'knowledge brief' before each customer call — relevant case studies, technical specs, even recent bug fixes that might matter. Their win rate went up 12 percent in six months. Lucas: That's a really concrete example. And it gets at something I think is the real killer app here: reducing the friction of 'I know this exists somewhere but I can't find it.' That's the death of a thousand cuts for knowledge workers. Luna: Yeah, it's like having a really good librarian who follows you around and hands you the right book before you even ask. Lucas: And that librarian learns from everyone. If someone in marketing finds a useful document and shares it in a channel, the AI can note that connection and surface it for others working on similar tasks. It's a network effect for knowledge. Luna: This is exactly the kind of conversation where I'm glad we keep this show ad-free. It lets us dig into the nuance without having to plug a sponsor who sells a knowledge management tool. Lucas: Totally. And it's not easy to run a podcast with no ads. If you find this kind of deep dive useful, and you want to support keeping it that way, we do have a link: buy me a coffee dot com slash fexingo. No pressure, just a way to help if you're able. Luna: It really does make a difference. And now, back to that librarian — one thing I'm curious about: how do these tools handle conflicting information? Like, if one doc says the pricing is X and another says Y? Lucas: That's the hard part. Most of them don't resolve conflicts — they just surface both and let you decide. But some are starting to use confidence scoring. If a document has been edited recently or viewed by more people, it gets ranked higher. There's also a startup called Tettra that uses human-verified 'best answers' on top of AI discovery. So you have a mix of algorithmic and manual curation. Luna: So it's not a magic bullet. You still need someone to maintain the knowledge base. Lucas: Right. The AI can suggest, but humans have to validate. That's actually a lesson I've heard from several companies that tried to go fully automated. Without a human in the loop, you get noise. With a human curator, you get a really powerful tool. Luna: What about the onboarding use case? You mentioned new analysts earlier. I've heard of companies using knowledge discovery to create personalized onboarding paths based on the new hire's role and team. Lucas: Yes. That's actually one of the most popular implementations right now. Instead of giving new employees a giant wiki to read, the AI pushes them the top five documents each day, based on what their mentor is working on and what their team has been discussing. One tech company I spoke to said it cut ramp-up time for new engineers by about 40 percent. They used a tool called Stack Overflow for Teams, which now has an AI that recommends questions and answers based on the user's job title and recent searches. Luna: Forty percent is huge. That's weeks of productivity saved. Lucas: Exactly. And the cost of implementing these tools is relatively low. Many are add-ons to existing platforms like Slack, Teams, or Confluence. The pricing is usually per-user, starting around five to ten dollars a month. For a company of a thousand employees, that's maybe fifty to a hundred thousand a year. If it saves even a week of lost productivity per employee, the ROI is massive. Luna: Let's talk about the competitive landscape. You mentioned Mem, Guru, Slab, Coda Brain, Tettra, Stack Overflow for Teams. Who else is in this space? Lucas: A lot of players. Microsoft has Viva Topics, which integrates with Teams and SharePoint. Google has an internal tool called Google Cymbal that's not publicly available but is used internally. There's also Notion AI, which recently added a 'Q&A' feature that can answer questions based on your workspace. And a newer entrant called Knoetic, which focuses on HR knowledge — surfacing policies and benefits info to employees. Luna: So it's becoming table stakes. If you're a knowledge management platform, you need an AI layer. Lucas: Absolutely. And I think the next frontier is multimodal knowledge discovery — pulling from video transcripts, voice calls, even screenshots. Imagine an AI that watches a recorded demo and then surfaces that demo to a sales rep who's about to talk to a similar prospect. Luna: That's powerful. But also makes the privacy conversation even more intense. Now it's not just text, it's my voice and face. Lucas: Yeah. And that's why I think the companies that win will be the ones that get the transparency right. Clear opt-in, clear explanations of what's being used, and easy ways to turn it off. The technology is impressive, but trust is the real bottleneck. Luna: Well said. I think we've only scratched the surface here. Let's come back to this in a few months and see how the landscape has evolved. Lucas: Definitely. And if anyone listening has tried one of these tools internally, we'd love to hear your experience.