Doug Camplejohn
(00:01)
Hello everyone, this is Doug Camplejohn, and welcome to this week’s episode of Revenue Renegades. I’m thrilled to be joined by Tamar Yehoshua, the former President of R&D at Glean, who is now advising the company. She was previously the Chief Product Officer at Slack, which is where we met, and spent time in venture at IVP. Tamar, welcome to the show.
Tamar Yehoshua
(00:27)
Thanks so much for having me. It’s great to see you again.
Doug Camplejohn
(00:30)
It’s great to see you too. You’ve had this fascinating journey—not only big roles at Slack, Google, and more—but also a deep dive into AI while in venture at IVP. I remember you really exploring every corner of the space. I’d love to start with what you took away from that time. And what led you to ultimately join Glean?
Tamar Yehoshua
(01:00)
It was an amazing time to be in venture. I left Slack and became a venture partner at IVP just as AI was taking off—ChatGPT had launched, and everyone was trying to figure out how to leverage OpenAI’s APIs.
There were many startups at the time—so-called “wrappers” around OpenAI—that popped up and fizzled out. Only a few infrastructure players looked promising. Over six months, I met with a ton of startups and started to spot who might have staying power and who was just hype. What became incredibly clear: AI is real, and it’s a game-changer.
Having worked on Google Search before Slack, I’d seen the tectonic shift when Transformers came out and how much of the stack we moved over to ML. Still, no one anticipated how powerful generative AI would become.
I knew Arvind Jain, Glean’s co-founder and CEO, from our days at Google Search. He was a core infrastructure engineer, and we crossed paths during a “code yellow” to improve search performance. Later, he co-founded Rubrik, and eventually started Glean because he saw enterprise search was fundamentally broken.
When I was at IVP, the Slack Fund had been one of Glean’s early investors, so I stayed in touch with Arvind. IVP got involved during their Series C or D, and I was part of the deal team. That meant doing deep diligence—customer interviews, metrics, team assessments—and what stood out to me was just how much customers loved the product. It felt like the early days of Slack again. So once we invested, I joined.
Doug Camplejohn
(03:59)
That’s fantastic. How did IVP react when you decided to leave and join Glean?
Tamar Yehoshua
(04:05)
They were incredibly supportive. Honestly, that’s the goal with venture partners—either keep investing or join a portfolio company. So they were thrilled I did.
Doug Camplejohn
(04:20)
I’ve heard people call Glean the “ChatGPT for organizational knowledge.” But these days, you’re seeing more MCP interfaces into systems. So why isn’t ChatGPT alone sufficient?
Tamar Yehoshua
(04:40)
Great question. Glean started as enterprise search—ingesting knowledge across all your SaaS apps: Salesforce, Jira, Slack, Google Docs, Microsoft, and more.
The big difference versus previous attempts is that Glean doesn’t rely on federated search, where you’re limited by the quality of each app’s own search. Instead, we built a complete index—like Google for your company—fully optimized with access controls and ranking signals.
In the enterprise, those ranking signals include things like your role, your org structure, recent usage, and team proximity. If you ask for an onboarding document, Glean will suggest the ones used by others in your role or team, not just the most globally popular ones. And we’re doing that securely—even with customers having billions of documents.
Once the foundation was built, we introduced Assistant, powered by RAG—Retrieval-Augmented Generation—where the “retrieval” is essentially search. Since we’d already built a best-in-class enterprise search engine, layering on LLMs gave us highly relevant responses.
Ask it for an onboarding plan—it gets you docs, team intros, even company videos. Because we integrate internal and external data, the assistant can be deeply contextual.
Doug Camplejohn
(08:13)
So it’s clear that search, relevance, and permissioning are foundational—and not something ChatGPT alone can replicate.
Tamar Yehoshua
(08:20)
Exactly. We support MCP, and you can call Glean through it. But federating MCP connections without deeper integration won’t yield high-quality results.
And security matters. If someone searches for something sensitive—like a layoff list—you better be sure permissions are right. One of the first questions CISOs ask us is: “Are you telling me my employees can now find everything they have access to? I’m not sure the documents are correctly secured.” It’s a wake-up call. That’s why we built sensitive content detection tools—because app-level permissions often fail.
Doug Camplejohn
(10:26)
So why didn’t Google solve enterprise search?
Tamar Yehoshua
(10:43)
They didn’t care enough. For a moment, I owned enterprise search at Google. But the company operates on a consumer scale—unless it’s a billion-dollar business, it doesn’t move the needle. Building enterprise-grade permissions, deployments, and compliance frameworks—HIPAA, FINRA, SSO—requires deep focus. That just wasn’t Google’s DNA. Cloud is finally changing that, but Glean made that commitment from day one.
Doug Camplejohn
(11:47)
Why did earlier enterprise search players fail?
Tamar Yehoshua
(11:55)
Two main reasons: First, they were too early. SaaS APIs weren’t mature. Second, the infrastructure investment was too high. Many tried to launch with a basic search engine, putting off the hard parts like indexing and ranking. The quality wasn’t there.
Doug Camplejohn
(12:33)
Let’s talk about agents. What is Glean’s definition of an “agent”?
Tamar Yehoshua
(12:44)
Everyone defines it differently, but for us, an agent is something that chains together multiple LLM calls and tools to take real action. It’s not just answering questions—it’s doing things.
Say you need a customer briefing. The agent will combine stock info, internal notes, feature requests from Jira, incidents from Zendesk, and Slack conversations, then summarize it all and email it to you. You can trigger this via a prompt or define it as a workflow. We also support external tools—via our custom APIs or MCP.
Doug Camplejohn
(17:30)
Sounds like we’re in the co-pilot phase, not autopilot. How far are we from full autonomy?
Tamar Yehoshua
(18:05)
Closer than we think. Trust builds over time. Eventually, when an agent’s been correct ten times in a row, users will say “just go ahead.” But use case matters—a mass email is low stakes; a VIP customer proposal is not.
Doug Camplejohn
(21:18)
How is managing AI products different from traditional SaaS?
Tamar Yehoshua
(21:18)
Non-determinism is the biggest challenge. CIOs expect the same answer every time. With AI, that’s not guaranteed. So expectation-setting becomes critical. You need new eval frameworks, too—it’s not about QA anymore, it’s more like debugging how reasoning happens.
Doug Camplejohn
(23:10)
Can you share examples of how Glean PMs use Glean internally?
Tamar Yehoshua
(23:13)
Sure. One PM built an agent to analyze Gong calls from the past year to identify API feature requests and the customers behind them—in 10 minutes. That would’ve taken quarters before. We also use agents to write help documentation and release notes—automating away a lot of grunt work so PMs can focus on strategy.
Doug Camplejohn
(25:10)
Any favorite tools for prototyping or prompt testing?
Tamar Yehoshua
(25:10)
Prototyping has been transformative. Tools like V0 from Vercel, Lovable, and Figma Make are popular with our teams. Also, engineers love AI coding tools—Cursor, Codium, Copilot. They use Glean for incident management, documentation, code comprehension. It’s deeply embedded in both our dev and customer workflows.
Doug Camplejohn
(26:29)
Setting expectations is critical for UX. What techniques work best?
Tamar Yehoshua
(27:16)
People either freeze at a blank box or think the AI can read their mind. Prompt sharing helped—it let power users create and distribute high-quality prompts. Just seeing what others were doing increased engagement.
With latency, even a fake progress bar helps. I learned this at Google Search—expectation setting makes people more patient. If they believe something is happening, they’ll wait.
Doug Camplejohn
(32:48)
How do you evaluate quality?
Tamar Yehoshua
(32:59)
We built internal tools modeled after Google Search eval tools. Side-by-side comparisons, debug views, signal tracing—it’s all in there. We also use LLMs as judges and rate responses across dimensions like grounding, completeness, and correctness. I wrote a blog about it called The AI Evaluator—though we’ve updated the framework significantly since.
Doug Camplejohn
(36:01)
Zooming out—how do you build your roadmap in today’s fast-paced AI world?
Tamar Yehoshua
(36:26)
Balancing short-term needs with long-term vision is always tough—especially now. We follow a loose 70/20/10 split, but AI has blown that up. We focus on integrating external innovations (like Bedrock, LangChain, CrewAI) instead of reinventing everything. Flexibility is key. If we’d stuck rigidly to our roadmap, we would’ve missed launching support for major frameworks that didn’t even exist when we planned the quarter.
Doug Camplejohn
(39:00)
Internal alignment must get harder as teams grow.
Tamar Yehoshua
(39:39)
Absolutely. Over-communication is essential—especially to first-line managers. They’re the ones fielding questions from ICs, and they need to be bought in. If they don’t understand the “why,” it all breaks down.
Doug Camplejohn
(41:08)
You once said “curiosity beats certainty.” Is that yours?
Tamar Yehoshua
(41:19)
Not originally, but I say it often. Being curious—trying to understand strong reactions and assumptions—is crucial. It helps you uncover insights or false premises.
Doug Camplejohn
(42:15)
Let’s end on a personal note. What do you like to do for fun?
Tamar Yehoshua
(42:25)
Now that my kids are grown, I actually have time! I work out with my daughter—we train with the same coach, and it’s brutal but fun.
Doug Camplejohn
(42:57)
Bonding through shared pain—I love it. What’s something people might be surprised to learn about you?
Tamar Yehoshua
(43:07)
I danced most of my life. I got back into it when my daughter started ballet. We’d even perform together—super fun. She’s now a serious dancer, doing salsa, and I love watching her.
Doug Camplejohn
(44:27)
Outside enterprise software, what’s one product you love?
Tamar Yehoshua
(44:27)
Aside from Glean—because it’s changed how I work—recently, Granola has been huge. I’m a big note-taker, and it solves that awkwardness of typing during meetings. Plus, its summaries are structured and reliable.
Doug Camplejohn
(45:53)
Totally agree. So last question—how can folks stay in touch?
Tamar Yehoshua
(46:11)
Best way is through LinkedIn—whether it’s about Glean or something else, I’ll route you to the right person.
Doug Camplejohn
(46:23)
Thanks so much for joining, Tamar. This was a fantastic conversation!
Tamar Yehoshua
(46:30)
Thank you, Doug.
