Person
2 episodes · 27 quoted claims
Arvind Jain said AI will capture revenue now spent on services, an industry he said is 25 times larger than the software industry.
“it's actually going to grab a lot of revenue that actually today is in services industry, which is 25 times larger than software industry.”
Arvind Jain said a common Glean agent turns meeting recordings into action items and updates Salesforce notes automatically.
“you take these meeting recordings, you figure out, like, you know what you talked to the customer, uh, what were the action items? And then the agent goes, updates the notes, you know, in Salesforce with that.”
Jain said Glean's work on fine-tuning its own models for specific product use cases did not pan out, so the company used existing models instead.
“some of our fine tuning work, building, building models, uh, for specific use case within our product. Like, you know, didn't, didn't really pan out for us.”
Arvind Jain says Glean started in early 2019, before large language models and the term generative AI were widely discussed.
“Glean started in early 2019. You know, in those days, nobody was talking about large language models, and the term generative AI did not exist to my knowledge at the time.”
Jain cites research studies saying people spend about one third of their working time trying to find things.
“there have been a lot of research studies that have been done. Uh, people, uh, would spend one 3rd of their working time just trying to find things.”
Jain says Google released BERT-based models openly in late 2018.
“Google had actually put, uh, BERT based models in open domain in late 2018, and they were able to use them.”
Jain says Glean was probably the first company to use transformer models in the enterprise.
“we were probably the 1st company to use transformers in the enterprise, you know, in in that day”
Jain says transformer models in 2019 could match content by meaning but could not generate text or reason as models do today.
“In 2019, the transformers had no capability to generate information, they had no, like, none of the reasoning capabilities that models have today.”
Glean uses out-of-the-box large language models, such as GPT, to synthesize answers from retrieved information.
“we don't actually use our own models, you know, we use out of the box models like, you know, GPT or Um.”
Glean uses a fine-tuned small language model for spell checking user queries.
“for spell checking, um, we, we have a fine tuned version of a language model that will do it for us.”
Glean builds source-system permissions into its indexing so it retrieves only documents the user may access.
“Index indexing technology is the permission for baked in right into our, you know, uh, index indexing system.”
Glean trains its custom models only on content it considers safe, to avoid leaking information through embeddings.
“we typically would train them on, um, content that we think is safe to train on.”
Jain says he started Rubrik in early 2014, focused on protecting business data against threats such as ransomware.
“we started Rubrik, uh, that was, um, early 2014, and we picked a problem which we knew every business faced.”
Jain says enterprise search had a history of failures over two to three decades, including Google's attempts.
“in the two or three decades, two or three decades of enterprise search history, including like products from Google. Like, you know, every time, you know, Google tried to actually build an enterprise search product, it was always, you know, always failed.”
Jain says the shift to SaaS made finding information harder but also made enterprise data readable, helping make Glean viable.
“One of them was SaaS, the the whole SaaS transformation, um, which made the problems so much more worse. But also it actually made the problem tractable.”
Jain says Google's culture, where engineers had great autonomy and resources, strongly shaped his management style at Glean.
“Google was a really special company, like it was very different from.”
Jain says software engineers may increasingly become code reviewers rather than code writers.
“software engineers, you know, you hear that in the future we're gonna become more more code reviewers as opposed to, like, you know, people who write code.”
Jain uses a Glean agent to generate up-to-date tutorials on internal systems by checking design docs against recent Jira tickets and code commits.
“I actually ask Glean to actually, gen, you know, give, generate a latest, you know, up to date tutorial on that topic for me”
Jain says Glean had no churn dashboards until some customers left in a recent quarter.
“we had some churn that happened last quarter. And then, like, you know, suddenly got worried, like, okay, what's happening?”
A Glean finance employee used Glean to build a customer risk report drawing on Salesforce, shared Slack channels and product usage dashboards.
“look in Salesforce to get out the list of all the customers.”
Jain says roughly 95 percent of enterprise data is unstructured.
“95 of your enterprise data is unstructured, uh, and this, you know, is full of insights.”
Glean builds golden question-and-answer sets for each customer, often from real conversations such as Slack threads, to test changes.
“we build these golden sets, you know, like for each one of our customers.”
Glean uses large language models as judges to compare new answers with the golden responses.
“you also use AI to sort of, uh, cause, you know, like these are subjective answers. Then you have to actually also use LLMs as judge”
Glean checks each line of a model's answer against the input it supplied, using the match for citations.
“line by line, you know, we'll take every, um, you know, the response of the model.”
When Glean cannot find support for a line in its source material, it suppresses the line or omits the citation.
“when we don't see it, we either suppress it or we actually would. Don't show, don't show the citations.”
Jain says agents asking humans for information is a plausible future use, alongside humans asking agents questions.
“agents are actually pinging. You know, people to retrieve some information on them, from them”
Jain says he does not shrink his team because of AI and instead wants each person to do ten times more work.
“I don't actually shrink my team size. Um, uh, I sure do grow it, but I want every to be able to do ten times more work”