StarZero All shows Try the podcast clipper →

Apr 24, 2025 · 52m · mad

Glean’s Breakthrough: CEO Arvind Jain on Scaling AI Agents & Search

Arvind Jain · 36m spoken Matt Turck · 12m spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

In this episode of The MAD Podcast, host Matt Turck interviews Arvind Jain, Founder and CEO of Glean, to discuss the evolution of enterprise AI search, RAG, and AI agents. Jain shares strategic insights on AI model selection, open-source adoption, building defensible enterprise tech stacks, and the future of AI-assisted workplace productivity.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Matt holds 24.8% of the talking time here. How this is scored →

Matt as informed peer 4.0 Guest teaching 3.5 Guest disagreement 1.4 Matt pushing back 2.1
05100:0015:0030:0045:001:14–10:34 · Matt as informed peer 5/10 The Macro AI Model Landscape The host displays strong domain knowledge by bringing up specific recent releases like O3, O4 Mini, Gemini 2.5 Flash, and new model startups like Thinking of Machines and SSI. The guest collaboratively explains enterprise model adoption phases, outlining why closed models dominate early before companies transition to distilled open-source models for cost and control.10:34–19:25 · Matt as informed peer 3/10 Defining AI Agents and Their Current Reality The guest mildly pushes back against adding yet another agent definition to the industry noise before breaking down the evolution from RPA to supervised AI agents. He educates the host on how full autonomy is rare in enterprise settings, emphasizing that agents currently deliver 90% efficiency gains through human-in-the-loop review.19:25–26:37 · Matt as informed peer 6/10 The Origins and Evolution of Glean The host shows deep industry memory by listing historical enterprise search pioneers like FAST, Endeca, Autonomy, and Verity, noting HP's $8 billion write-off. The guest gently corrects the host's premise about grand market foresight, explaining Glean was born out of raw personal frustration with fragmented SaaS data at Rubrik.26:37–33:58 · Matt as informed peer 4/10 Glean's Agentic Era and Platform Strategy The host cites specific figures including 50 million automated agent actions and asks why Glean pursued a horizontal strategy instead of building vertical per-function products. The guest explains how rebranding apps to agents aligned with market terminology and how existing search connectors naturally enabled horizontal workflows.33:58–43:51 · Matt as informed peer 5/10 Technical Deep Dive: Enterprise Search Architecture and Agent Design The host asks probing technical questions regarding search permissions, document decay, model routing, and why Glean chose tool-use architecture over computer-use. The guest politely clarifies that computer-use isn't inherently rejected but prioritized behind immediate customer demand for tool integration.43:51–45:58 · Matt as informed peer 3/10 Data Flywheels and Personalization in Glean The host inquires about data flywheels and usage-based intelligence accumulation. The guest educates the host on how implicit human signals—such as answering questions in Slack with links or peer endorsements—feed authority scores and expert mapping in Glean.45:58–48:33 · Matt as informed peer 3/10 AI Prompt Studio, Prompt Libraries, and Agents The host pushes on defensibility and how Glean builds a moat as competition heats up. The guest rejects traditional consumer network-effect framings like eBay for enterprise software, arguing that moats in B2B come solely from speed of execution and solving immediate customer problems.48:33–51:38 · Matt as informed peer 3/10 The Future of Work and Enterprise AI Architecture The host asks for the guest's long-term vision regarding centralized enterprise AI versus departmental agents. The guest paints a clear architectural vision of horizontal data and search layers supporting thousands of specialized functional AI assistants.1:14–10:34 · Guest teaching 3/10 The Macro AI Model Landscape The host displays strong domain knowledge by bringing up specific recent releases like O3, O4 Mini, Gemini 2.5 Flash, and new model startups like Thinking of Machines and SSI. The guest collaboratively explains enterprise model adoption phases, outlining why closed models dominate early before companies transition to distilled open-source models for cost and control.10:34–19:25 · Guest teaching 4/10 Defining AI Agents and Their Current Reality The guest mildly pushes back against adding yet another agent definition to the industry noise before breaking down the evolution from RPA to supervised AI agents. He educates the host on how full autonomy is rare in enterprise settings, emphasizing that agents currently deliver 90% efficiency gains through human-in-the-loop review.19:25–26:37 · Guest teaching 3/10 The Origins and Evolution of Glean The host shows deep industry memory by listing historical enterprise search pioneers like FAST, Endeca, Autonomy, and Verity, noting HP's $8 billion write-off. The guest gently corrects the host's premise about grand market foresight, explaining Glean was born out of raw personal frustration with fragmented SaaS data at Rubrik.26:37–33:58 · Guest teaching 3/10 Glean's Agentic Era and Platform Strategy The host cites specific figures including 50 million automated agent actions and asks why Glean pursued a horizontal strategy instead of building vertical per-function products. The guest explains how rebranding apps to agents aligned with market terminology and how existing search connectors naturally enabled horizontal workflows.33:58–43:51 · Guest teaching 4/10 Technical Deep Dive: Enterprise Search Architecture and Agent Design The host asks probing technical questions regarding search permissions, document decay, model routing, and why Glean chose tool-use architecture over computer-use. The guest politely clarifies that computer-use isn't inherently rejected but prioritized behind immediate customer demand for tool integration.43:51–45:58 · Guest teaching 4/10 Data Flywheels and Personalization in Glean The host inquires about data flywheels and usage-based intelligence accumulation. The guest educates the host on how implicit human signals—such as answering questions in Slack with links or peer endorsements—feed authority scores and expert mapping in Glean.45:58–48:33 · Guest teaching 4/10 AI Prompt Studio, Prompt Libraries, and Agents The host pushes on defensibility and how Glean builds a moat as competition heats up. The guest rejects traditional consumer network-effect framings like eBay for enterprise software, arguing that moats in B2B come solely from speed of execution and solving immediate customer problems.48:33–51:38 · Guest teaching 3/10 The Future of Work and Enterprise AI Architecture The host asks for the guest's long-term vision regarding centralized enterprise AI versus departmental agents. The guest paints a clear architectural vision of horizontal data and search layers supporting thousands of specialized functional AI assistants.1:14–10:34 · Guest disagreement 1/10 The Macro AI Model Landscape The host displays strong domain knowledge by bringing up specific recent releases like O3, O4 Mini, Gemini 2.5 Flash, and new model startups like Thinking of Machines and SSI. The guest collaboratively explains enterprise model adoption phases, outlining why closed models dominate early before companies transition to distilled open-source models for cost and control.10:34–19:25 · Guest disagreement 2/10 Defining AI Agents and Their Current Reality The guest mildly pushes back against adding yet another agent definition to the industry noise before breaking down the evolution from RPA to supervised AI agents. He educates the host on how full autonomy is rare in enterprise settings, emphasizing that agents currently deliver 90% efficiency gains through human-in-the-loop review.19:25–26:37 · Guest disagreement 2/10 The Origins and Evolution of Glean The host shows deep industry memory by listing historical enterprise search pioneers like FAST, Endeca, Autonomy, and Verity, noting HP's $8 billion write-off. The guest gently corrects the host's premise about grand market foresight, explaining Glean was born out of raw personal frustration with fragmented SaaS data at Rubrik.26:37–33:58 · Guest disagreement 1/10 Glean's Agentic Era and Platform Strategy The host cites specific figures including 50 million automated agent actions and asks why Glean pursued a horizontal strategy instead of building vertical per-function products. The guest explains how rebranding apps to agents aligned with market terminology and how existing search connectors naturally enabled horizontal workflows.33:58–43:51 · Guest disagreement 2/10 Technical Deep Dive: Enterprise Search Architecture and Agent Design The host asks probing technical questions regarding search permissions, document decay, model routing, and why Glean chose tool-use architecture over computer-use. The guest politely clarifies that computer-use isn't inherently rejected but prioritized behind immediate customer demand for tool integration.43:51–45:58 · Guest disagreement 0/10 Data Flywheels and Personalization in Glean The host inquires about data flywheels and usage-based intelligence accumulation. The guest educates the host on how implicit human signals—such as answering questions in Slack with links or peer endorsements—feed authority scores and expert mapping in Glean.45:58–48:33 · Guest disagreement 3/10 AI Prompt Studio, Prompt Libraries, and Agents The host pushes on defensibility and how Glean builds a moat as competition heats up. The guest rejects traditional consumer network-effect framings like eBay for enterprise software, arguing that moats in B2B come solely from speed of execution and solving immediate customer problems.48:33–51:38 · Guest disagreement 0/10 The Future of Work and Enterprise AI Architecture The host asks for the guest's long-term vision regarding centralized enterprise AI versus departmental agents. The guest paints a clear architectural vision of horizontal data and search layers supporting thousands of specialized functional AI assistants.1:14–10:34 · Matt pushing back 2/10 The Macro AI Model Landscape The host displays strong domain knowledge by bringing up specific recent releases like O3, O4 Mini, Gemini 2.5 Flash, and new model startups like Thinking of Machines and SSI. The guest collaboratively explains enterprise model adoption phases, outlining why closed models dominate early before companies transition to distilled open-source models for cost and control.10:34–19:25 · Matt pushing back 2/10 Defining AI Agents and Their Current Reality The guest mildly pushes back against adding yet another agent definition to the industry noise before breaking down the evolution from RPA to supervised AI agents. He educates the host on how full autonomy is rare in enterprise settings, emphasizing that agents currently deliver 90% efficiency gains through human-in-the-loop review.19:25–26:37 · Matt pushing back 3/10 The Origins and Evolution of Glean The host shows deep industry memory by listing historical enterprise search pioneers like FAST, Endeca, Autonomy, and Verity, noting HP's $8 billion write-off. The guest gently corrects the host's premise about grand market foresight, explaining Glean was born out of raw personal frustration with fragmented SaaS data at Rubrik.26:37–33:58 · Matt pushing back 2/10 Glean's Agentic Era and Platform Strategy The host cites specific figures including 50 million automated agent actions and asks why Glean pursued a horizontal strategy instead of building vertical per-function products. The guest explains how rebranding apps to agents aligned with market terminology and how existing search connectors naturally enabled horizontal workflows.33:58–43:51 · Matt pushing back 3/10 Technical Deep Dive: Enterprise Search Architecture and Agent Design The host asks probing technical questions regarding search permissions, document decay, model routing, and why Glean chose tool-use architecture over computer-use. The guest politely clarifies that computer-use isn't inherently rejected but prioritized behind immediate customer demand for tool integration.43:51–45:58 · Matt pushing back 1/10 Data Flywheels and Personalization in Glean The host inquires about data flywheels and usage-based intelligence accumulation. The guest educates the host on how implicit human signals—such as answering questions in Slack with links or peer endorsements—feed authority scores and expert mapping in Glean.45:58–48:33 · Matt pushing back 3/10 AI Prompt Studio, Prompt Libraries, and Agents The host pushes on defensibility and how Glean builds a moat as competition heats up. The guest rejects traditional consumer network-effect framings like eBay for enterprise software, arguing that moats in B2B come solely from speed of execution and solving immediate customer problems.48:33–51:38 · Matt pushing back 1/10 The Future of Work and Enterprise AI Architecture The host asks for the guest's long-term vision regarding centralized enterprise AI versus departmental agents. The guest paints a clear architectural vision of horizontal data and search layers supporting thousands of specialized functional AI assistants.

speaking balance: gold is Matt, purple is the guest (3 minute bins)

0:00 · Matt 77.1% · guest 22.9%0:00 · Matt 77.1% · guest 22.9%3:00 · Matt 11.1% · guest 88.9%3:00 · Matt 11.1% · guest 88.9%6:00 · Matt 38.1% · guest 61.9%6:00 · Matt 38.1% · guest 61.9%9:00 · Matt 27.5% · guest 72.5%9:00 · Matt 27.5% · guest 72.5%12:00 · Matt 9.8% · guest 90.2%12:00 · Matt 9.8% · guest 90.2%15:00 · Matt 9.6% · guest 90.4%15:00 · Matt 9.6% · guest 90.4%18:00 · Matt 44.9% · guest 55.1%18:00 · Matt 44.9% · guest 55.1%21:00 · Matt 2.7% · guest 97.3%21:00 · Matt 2.7% · guest 97.3%24:00 · Matt 32.2% · guest 67.8%24:00 · Matt 32.2% · guest 67.8%27:00 · Matt 13.9% · guest 86.1%27:00 · Matt 13.9% · guest 86.1%30:00 · Matt 15.7% · guest 84.3%30:00 · Matt 15.7% · guest 84.3%33:00 · Matt 36.2% · guest 63.8%33:00 · Matt 36.2% · guest 63.8%36:00 · Matt 3.6% · guest 96.4%36:00 · Matt 3.6% · guest 96.4%39:00 · Matt 35.9% · guest 64.1%39:00 · Matt 35.9% · guest 64.1%42:00 · Matt 28% · guest 72%42:00 · Matt 28% · guest 72%45:00 · Matt 18.2% · guest 81.8%45:00 · Matt 18.2% · guest 81.8%48:00 · Matt 12.8% · guest 87.2%48:00 · Matt 12.8% · guest 87.2%51:00 · Matt 39.2% · guest 60.8%51:00 · Matt 39.2% · guest 60.8%
Sharpest disagreement ▶ 11:07 Guest dismisses need for another agent definition

When prompted by the host for an agent definition, the guest playfully pushes back by stating he doesn't know if the world is looking for one more definition.

 Hardest push from Matt ▶ 39:18 Host challenges practical reality of LLM agnosticism

The host presses past marketing terminology to ask what being LLM agnostic really means in practice when new models exhibit vastly different behaviors and require constant re-evaluation.

 Biggest teaching moment ▶ 19:25 Guest reframes Glean origins from market foresight to personal pain

When the host frames Glean's 2019 founding around industry cycles and transformer timing, the guest clarifies that starting the company stemmed from personal frustration with fragmented data at Rubrik rather than grand market insight.

 Matt holds his own ▶ 19:25 Host details legacy enterprise search graveyard

The host showcases deep domain knowledge by listing classic 2000s enterprise search companies like FAST, Endeca, Autonomy, and Verity, citing HP's $8 billion write-off to establish the challenge Glean faced.
the scores for every segment, with the reasoning behind each
ChapterTopicMatt as informed peerGuest teachingGuest disagreementMatt pushing backWhy
The Macro AI Model Landscape 5312 The host displays strong domain knowledge by bringing up specific recent releases like O3, O4 Mini, Gemini 2.5 Flash, and new model startups like Thinking of Machines and SSI. The guest collaboratively explains enterprise model adoption phases, outlining why closed models dominate early before companies transition to distilled open-source models for cost and control.
Defining AI Agents and Their Current Reality 3422 The guest mildly pushes back against adding yet another agent definition to the industry noise before breaking down the evolution from RPA to supervised AI agents. He educates the host on how full autonomy is rare in enterprise settings, emphasizing that agents currently deliver 90% efficiency gains through human-in-the-loop review.
The Origins and Evolution of Glean 6323 The host shows deep industry memory by listing historical enterprise search pioneers like FAST, Endeca, Autonomy, and Verity, noting HP's $8 billion write-off. The guest gently corrects the host's premise about grand market foresight, explaining Glean was born out of raw personal frustration with fragmented SaaS data at Rubrik.
Glean's Agentic Era and Platform Strategy 4312 The host cites specific figures including 50 million automated agent actions and asks why Glean pursued a horizontal strategy instead of building vertical per-function products. The guest explains how rebranding apps to agents aligned with market terminology and how existing search connectors naturally enabled horizontal workflows.
Technical Deep Dive: Enterprise Search Architecture and Agent Design 5423 The host asks probing technical questions regarding search permissions, document decay, model routing, and why Glean chose tool-use architecture over computer-use. The guest politely clarifies that computer-use isn't inherently rejected but prioritized behind immediate customer demand for tool integration.
Data Flywheels and Personalization in Glean 3401 The host inquires about data flywheels and usage-based intelligence accumulation. The guest educates the host on how implicit human signals—such as answering questions in Slack with links or peer endorsements—feed authority scores and expert mapping in Glean.
AI Prompt Studio, Prompt Libraries, and Agents 3433 The host pushes on defensibility and how Glean builds a moat as competition heats up. The guest rejects traditional consumer network-effect framings like eBay for enterprise software, arguing that moats in B2B come solely from speed of execution and solving immediate customer problems.
The Future of Work and Enterprise AI Architecture 3301 The host asks for the guest's long-term vision regarding centralized enterprise AI versus departmental agents. The guest paints a clear architectural vision of horizontal data and search layers supporting thousands of specialized functional AI assistants.

Statements from this episode (7)

Assertion Supported
Arvind Jain: Most enterprise AI use cases rely on closed models
“Today, the, for most customer use cases the companies are using the closed, closed models at the moment”
Arvind Jain Apr 24, 2025 ▶ 4:17
Assertion Not checkable as stated
Arvind Jain: Enterprise customers rarely run AI agents fully unsupervised
“We barely see any application where customers are running agents in a fully unattended unsupervised, you know, setting.”
Arvind Jain Apr 24, 2025 ▶ 18:17
Insight
Arvind Jain: AI agents are shifting from RAG to process automation
“Agents are now getting a lot more powerful. They are, You know, they're getting they're sort of shifting from sort of basic two step rack kind of application flow where you take a task, you find some information, and then you make AI work on it to generate the…”
Arvind Jain Apr 24, 2025 ▶ 30:19
Disclosure
Arvind Jain: Glean evaluates new AI models within 30 minutes
“We need to have a way within, like, you know, 30 minutes to know how well it's going to do on our product, right? I mean, and you can do those things. You can actually build the right, you know, evaluation frameworks, things like that.”
Arvind Jain Apr 24, 2025 ▶ 40:43
Insight
Arvind Jain: Enterprise software cannot build eBay-style network effects
“In enterprises, you know, it's sort of hard to explore. It's hard to actually you know, have, like, Network effects like, you know, the kind that you have with eBay. Like, you know, enterprise software doesn't work that way.”
Arvind Jain Apr 24, 2025 ▶ 48:08
Prediction Not checkable as stated
Arvind Jain: Large enterprises will deploy thousands of vertical AI agents
“Now on top of that, you will have agents which will proliferate and they're going to be they're going to be like thousands of these agents inside, inside a large enterprise and they are all functional and they are, you know, vertical in nature.”
Arvind Jain Apr 24, 2025 ▶ 49:43
Prediction Not checkable as stated
Arvind Jain: Every worker will have a team of AI assistants
“And the future that I see with AI is that every person who works is going to have this amazing team of assistants, coworkers, and coaches around them that is going to actually make them a lot more effective.”
Arvind Jain Apr 24, 2025 ▶ 50:33
Made with StarZero

Turn any episode into a week of clips.

This entire site, over 400 conversations transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more