Skip to content

Podcast

20VC- Why OpenAI and Anthropic Won't Win the App Layer | Why Teams Will Get Bigger Not Smaller in a World of AI | Why AI Removes Incumbents Advantage of Bundling | China vs America- Who Wins the AI War With Arvind Jain, Co-Founder @ Glean

Startup Funding | The Pitch

Source ↗ ← All highlights
  • Glean As An AI Coworker
    • Glean evolved from enterprise search into an AI coworker that unifies models like ChatGPT and Gemini into one experience.
    • It connects to a company’s context across hundreds of systems to deliver answers grounded in corporate data rather than generic model output. (Time 0:05:51)
  • Agents Accumulate Institutional Knowledge
    • Enterprises fear operational dependence if frontier model providers own agents that accumulate institutional learning.
    • If the agent stores decades of process knowledge and the enterprise doesn’t control it, the company loses its compounding operational IP. (Time 0:06:47)
  • Use Open Source To Control AI Cost
    • Prioritize open source models primarily to cut inferencing costs and avoid runaway token spend.
    • Enterprises now commonly prefer open models for economics, shifting adoption once open models reach frontier parity. (Time 0:09:08)
  • Treat Model Labs As Assets Not Enemies
    • Stop obsessing about frontier labs as existential competitors and treat them as ecosystem enablers.
    • Build on top of model providers and open source to deliver differentiated context and product capabilities. (Time 0:13:15)
  • Models Are Commoditized For Most Workflows
    • Most enterprise use cases (90%+) can’t be fully solved by generic models; selection and orchestration matter.
    • Glean optimizes by choosing the right model per task and falling back to open source when quality and cost align. (Time 0:14:11)
  • Consumption Breaks Bundling Advantage
    • Bundling (e.g., Microsoft) is powerful, but consumption-based pricing undermines bundling advantage.
    • When customers pay per unit of work, they will mix best-of-breed tools and pay where value is delivered. (Time 0:18:31)
  • ROI Is Strongest In Narrow Repeatable Tasks
    • AI shows clear ROI in narrow workflows like customer support where productivity metrics are measurable.
    • Broad areas such as engineering show mixed ROI because coding speed rose but shipping velocity didn’t clearly increase. (Time 0:21:17)
  • Context Is The Key To Efficient AI
    • AI performance depends on pre-attaching high-quality context; without it models brute-force assemble context, burning tokens and time.
    • Investing in contextual plumbing around models dramatically reduces latency and token consumption. (Time 0:24:51)
  • Expensive Triage Agent Example
    • Glean built a triage engineering agent that automated 95% of on-call alerts but cost about $1M per month to run.
    • The team questioned whether the agent was more efficient than humans given its extreme inferencing expense. (Time 0:30:40)
  • This Moment Is A Land Grab
    • AI market is a land grab where early deployment grants durable advantages; delaying adoption makes future entry much harder.
    • Arvind warns that acquiring customers now avoids a 10x harder motion later. (Time 0:40:07)
  • Composite Roles Replace Narrow Specialists
    • Composite generalist roles will rise as AI automates specialized tasks, merging product, design, and engineering responsibilities.
    • Arvind expects fewer pure specialists and more multi-skilled people who leverage AI to cover multiple functions. (Time 0:40:29)
  • Too Much Early Capital Harms Startups
    • Avoid excessive early-stage capital that inflates unsustainable pay and hiring practices.
    • Overabundant funding leads startups to pay unrealistic compensations that undermine long-term company viability. (Time 0:48:34)