Episode Summary
Executive Summary: The episode centers on Google Gemini’s Deep Research, which Jason and Alex praise as a major leap in consumer AI for web-based research, planning, citations, and iterative follow-up. Google PM Arush Selvin explains how it decomposes prompts into research plans, runs parallel web searches, self-checks outputs, and keeps sources visible. The discussion then broadens to product naming, monetization, hallucination control, future agents/reminders, AI-generated media, and a TikTok ban update.
Main Topics: Google Gemini Deep Research demo and product design: Arush Selvin explains that Deep Research acts as a personal AI research agent: it turns a query into a research plan, searches the web in parallel, reasons over results, and revises its own draft before presenting a sourced report. AI product velocity and Google’s shift in shipping culture: Jason argues Google’s AI product velocity has improved substantially, moving from cautious releases to faster iteration and broader consumer-facing experimentation. Monetization and compute economics of AI research: The conversation examines whether Deep Research is expensive to run, how Google may absorb costs for market share, and whether subscriptions, ads, or hybrid models will dominate. Hallucinations, citations, and trust: Arush emphasizes that Deep Research is source-forward with inline citations and exportable docs to help users verify information, acknowledging hallucinations remain an unsolved industry problem. Overlap with NotebookLM and product convergence: Jason questions why Google has multiple adjacent research tools, while Arush says products may converge over time and borrow the best features from one another. Future agents, reminders, and workflow automation: The hosts connect Deep Research to OpenAI’s scheduled tasks feature and predict a near future where AI monitors ongoing topics, updates research automatically, and supports recurring work. TikTok ban, app migration, and geopolitical control: The episode closes with a discussion of the looming TikTok ban, user migration to other apps, and concerns about Chinese state influence and data access.
Key Arguments: Deep Research is valuable because it lets one episode or one research session cover what previously took multiple prep cycles. The product’s core advantage is not just answering prompts but planning, searching, and self-correcting with visible sources. Google’s improved shipping speed reflects a broader cultural change toward releasing useful AI products before they are perfect. Deep Research’s business model is currently subscription-based, but the hosts speculate that ads or hybrid monetization could emerge. Hallucinations are not solved, so source transparency and linkable citations are essential to user trust. NotebookLM and Deep Research are adjacent tools that should likely converge rather than remain siloed. AI agents that monitor topics and trigger updates could become the next major layer of research workflows. TikTok’s fate is framed as a geopolitical and legal issue, not just a product shutdown; users and app traffic will likely migrate elsewhere.
Data Points: Google AI product rating from Jason: A - Jason gives Deep Research a top grade after being initially skeptical of Google’s AI execution. Approximate Deep Research runtime: 5 minutes - Arush says the system really does take the full time while it researches the web. User subscription price: $20/month - Deep Research is available through the Gemini Pro subscription tier. Jason’s monthly AI spend increase: $20 to $40 - He says subscribing to Gemini doubled his monthly AI spend. Estimated human research time replaced: 4–8 hours - Jason says a human researcher would take this long to produce comparable work. Battery demo output: A table plus market overview - Alex shows Deep Research producing structured analysis on battery technology. Research thread reuse: All outputs remain in context - Arush explains follow-up research and Q&A stay within the same thread. TikTok ban date discussed: January 19, 2025 - The hosts discuss the app store removal / shutdown timeline if the Supreme Court doesn’t intervene. Synthesia raise: $180 million at a $2.1 billion post-money valuation - Jason uses this as a context point for AI video generation startup momentum. Prior Synthesia round: $90 million at a $1 billion post-money valuation - Jason references the company’s earlier funding round for comparison. AI investment count: 8 deals - Jason notes eight AI-related investments over $100 million in the first half of January 2025.
Pivotal Quotes: "Deep research is designed to be your personal AI research agent." — Arush Selvin: Defines the product’s core value proposition during the demo explanation. "What it does is it allows us to cover more topics better and then move up the stack to having a deeper, richer discussion about it." — Jason: Explains why the product matters for podcast research and prep. "We’ve been really surprised and like very pleasantly surprised at how it’s being received." — Arush Selvin: Arush describes early user reception to the product.
Implications: Deep Research points toward AI research workflows that are faster, more iterative, and citation-backed. For media, VC, and knowledge work, the moat may shift from gathering information to curating, verifying, and adding human judgment, humor, and context.
About This Week in Startups
Jason Calacanis covers startups, tech, markets, media, and all the hottest topics in business and technology. He also interviews the world’s greatest founders, operators, investors, and innovators.