Keep pulling the thread on Google Search.
Arvind Jain believes the step-function improvement in AI agent capabilities occurred with the release of advanced reasoning models like Anthropic's Claude Opus and OpenAI's GPT-5 class models.
For high-stakes, customer-facing tasks like bank transfers, Cresta recommends that the core logic should be hard-coded rather than relying on LLM-driven autonomy.
Glean recommends that any agent designed to mutate enterprise state, such as modifying a database, should be implemented as a "human in the loop" agent.
Some telephony providers charge customers extra fees for third-party AI vendors like Cresta to access the same audio stream that the customer has already paid for.
Glean's strategy is to remain a horizontal AI platform for businesses, providing an "enterprise brain" that can power various vertical products via APIs.
Arvind Jain believes the next major adoption driver for Glean will be shifting from a reactive, user-query-driven model to a proactive one where Glean anticipates user needs and offers help.
Glean is working towards a future capability where its proactive AI companion can perform half of a knowledge worker's tasks.
Glean's founding goal was to create a "Google-like" search product for the workplace.
Glean used Transformer models in its initial product from its inception.
The adoption of Transformer models significantly improved the performance of intent detection in Cresta's platform.
Glean built the first version of its agentic product approximately two years ago using OpenAI's GPT-4 model.
Cresta deployed its first voice AI agent for complex troubleshooting issues with a design partner customer sometime in the last year.