Grit

The Grittiest Conversations of 2025: AI, Business & Beyond

December 29, 2025

Key Takeaways Copied to clipboard!

  • Remaining politically neutral allows figures like David Rubenstein to effectively engage with and educate members across the political spectrum on non-partisan issues. 
  • The current AI wave is fundamentally exciting because it shifts software's role from performing tasks to maximizing customer outcomes, potentially eliminating vast amounts of preparatory work previously required in roles like sales. 
  • The next significant gains in AI model performance will rely on high-quality, domain-specific human expertise (post-training data) rather than just increased compute or general internet data, necessitating expert networks like Handshake's. 

Segments

Rubenstein on Political Neutrality
Copied to clipboard!
(00:00:30)
  • Key Takeaway: David Rubenstein avoids political donations to prevent being blamed for politicians’ mistakes and to maintain access for non-partisan civic roles.
  • Summary: Remaining neutral allows Rubenstein to work across administrations and bring Democrats and Republicans together for educational dinners at the Library of Congress. He cites concerns that political donations could lead to accusations of buying access, especially concerning Carlyle’s government contracts. Rubenstein noted he was the first person fired by a president (Trump) and then succeeded by him (Biden) in a board role.
HubSpot CEO on AI Value
Copied to clipboard!
(00:04:50)
  • Key Takeaway: AI’s true value lies in shifting focus from performing tasks to maximizing customer outcomes, fundamentally rewriting professional roles.
  • Summary: The jobs of marketers, salespeople, and support staff are being rewritten, allowing professionals to achieve better results with their time. Historically, salespeople spent only about 25-27% of their time directly in front of customers, with the rest dedicated to preparation and administrative work that AI can now automate. Companies must bet on transformative cycles like AI, as failing to invest when a trend is real can lead to leaving hundreds of millions of dollars on the table.
Mailchimp Founder on Tech Cycles
Copied to clipboard!
(00:11:26)
  • Key Takeaway: The acceleration of technology cycles, particularly driven by AI, makes running a nimble tech company increasingly difficult, leading to relief post-exit.
  • Summary: Ben Chestnut felt compelled to reinvent Mailchimp every three years due to changing customer needs, limiting commitments to one-year timelines to maintain agility. He observed this acceleration increasing over time, making it harder to keep up as the company grew larger. Having invested early in big data and AI since 2008, he is relieved not to be a CEO navigating the current rapid AI disruption.
Harvey on Legal AI Strategy
Copied to clipboard!
(00:13:08)
  • Key Takeaway: Harvey focuses on the ‘meatiest’ segment of legal AI—top law firms—because that work is context-specific and requires building trust to survive future model advances.
  • Summary: The most impactful entry point for legal AI is targeting big law firms and enterprise legal teams, where trust and brand references are critical. Building a business must account for what models can do in 10 years, focusing on messy, context-specific work where accuracy is paramount. Their strategy involves deep enterprise engagement rather than immediate self-serve models, which are more susceptible to cannibalization by future general models.
Handshake on Expert Data for AI
Copied to clipboard!
(00:18:20)
  • Key Takeaway: Frontier AI models require domain experts to train, validate, and correct complex reasoning, moving beyond generalists who were sufficient for earlier internet data.
  • Summary: AI gains on the post-training side are fueled by human data, and as models absorb the general internet corpus, the need shifts to specialized experts in fields like law, medicine, and STEM. Experts are paid high rates (averaging over $100/hour) to correct reasoning errors in step-by-step instructions and provide ground truth answers for complex problems. Handshake builds community experiences to retain these experts to produce high-quality, domain-specific data for AI labs.
Cohere CEO on Transformer Architecture
Copied to clipboard!
(00:23:40)
  • Key Takeaway: The core insight of the Transformer paper was its efficiency and suitability for scaling across many GPUs, which catalyzed the current AI progress.
  • Summary: Aidan Gomez believes that even if Google had kept the Transformer proprietary, the underlying ideas were in the ether, and a similar architecture would have emerged within 12 to 18 months. The key insight was designing an architecture that was extremely well-suited to scaling training across multiple GPUs, which became crucial as model sizes grew exponentially. The community, not just Google, carried the work forward by implementing and testing the architecture across various frameworks.
Cloudflare on Web Business Model Risk
Copied to clipboard!
(00:28:09)
  • Key Takeaway: AI crawlers threaten the web’s existing business model by consuming content without sending traffic or compensating creators, necessitating new payment rails.
  • Summary: AI adoption is progressing five times faster than mobile, and this technology puts the web’s content creator compensation model at risk because users consume answers directly from LLMs instead of visiting original sources. Cloudflare customers face rising infrastructure costs from increased bot crawling while simultaneously seeing traffic decline as eyeballs stay within AI chatbots. New protocols, like the X402 code, are emerging to facilitate micro-payments, allowing creators to charge AI crawlers for content usage.
Snap CEO on AR and AI
Copied to clipboard!
(00:35:53)
  • Key Takeaway: AI accelerates the utility of augmented reality glasses by enabling them to function as a portable, collaborative workstation, shifting computing from single-player to shared experiences.
  • Summary: For AR glasses to be adopted, the world must immediately become better upon wearing them, requiring significant functionality beyond simple recording. AI will accelerate glasses adoption by operating the computer for the user, allowing them to observe and monitor AI rather than constantly operating the device. This shift enables users to multitask with an infinite workstation, overcoming the single-app limitation of phones and bringing desktop-level productivity on the go.