# Citrini Research Breakdown: Agents, "Ghost GDP", Consumer Spend | Figma Earnings Beat Source: https://www.youtube.com/watch?v=mBE_9vGJBUM Recap page: https://rapidrecap.app/video/mBE_9vGJBUM Generated: 2026-02-26T16:00:33.57+00:00 --- ## Quick Overview The discussion concludes that while AI acceleration is rapidly making existing B2B software dated and causes panic in certain stocks like cybersecurity firms due to lofty valuations, the greatest opportunity lies in building new, specialized, intelligence-led applications and competitive agents that integrate AI capabilities, rather than relying on foundation models directly or incumbents simply 'sprinkling AI dust'. **Key Points:** - Anthropic's security review feature triggered a $20 billion wipeout in cybersecurity stocks, but the capabilities largely already existed in tools like Replet, indicating market panic over features that were months old. - When companies are priced for perfection, like CrowdStrike trading at 16x revenues despite 22% revenue growth projections, 'anything less than perfection will be a kick in the nuts,' leading to substantial corrections. - The speaker predicts the creation of '100,000 deca millionaires out of these AI leaders,' noting that Nvidia already accounts for 20,000. - The consensus suggests intelligence will infuse all software, delivered either by foundation models, incumbents integrating AI, or new companies building on top of foundation models (scenarios three and four); enterprises are unlikely to build proprietary systems directly on Claude. - Publicly traded B2B companies lack competitive agents, with Palunteer cited as the only one demonstrating revenue acceleration from agents, suggesting most incumbents are just 'sprinkling AI dust' and face disruption. - Implementing effective agents requires overcoming significant institutional momentum, custom training, onboarding, and data cleansing, which most current B2B companies are ill-equipped to handle, leaving the door open for startups. - The 'Ghost GDP' concept suggests productivity gains from AI accrue to fewer people, potentially shrinking the consumer wallet base and softening overall consumer spend, although historically, productivity gains benefit long-term GDP. **Context:** The 20VC podcast episode analyzes recent market events, focusing on the fallout from Citrini research regarding a potential 2028 global intelligence crisis, the impact of Anthropic's new security feature on cybersecurity stocks, and the valuation challenges faced by high-growth B2B software companies. Key figures discuss the micro-level impact of agents on existing software categories like DoorDash and Netflix versus the macro implications of productivity gains leading to wealth concentration, as exemplified by the massive liquidity events surrounding AI leaders like OpenAI and Anthropic. ## Detailed Analysis The discussion begins by analyzing the market overreaction to Anthropic's security release, noting that the touted capabilities for code auditing already existed on platforms like Replet, suggesting the $20 billion drop in cybersecurity stocks was driven by overvaluation rather than a fundamental technological shock, though the pace of improvement remains staggering. A major theme is the risk associated with being 'priced for perfection'; high-multiple stocks like CrowdStrike suffer significant drawdowns from minor setbacks because their valuation demands flawlessness, contrasting with cheaper stocks trading at lower multiples like six to eight times 2026 EBITDA. Regarding the future of B2B software, the speakers argue that intelligence-led applications are the only ones that will grow quickly, while non-intelligent applications face stagnation or decline, emphasizing that incumbents are being disrupted in real-time by the agentic layer, even if they publicly claim to be integrating AI. Implementing these agents, however, is difficult, requiring extensive custom work, data cleansing, and specialized engineers, which explains why startups built on foundation models are seeing explosive revenue while public companies struggle to show material impact from their own agent efforts. On macroeconomics, the conversation addresses the Citrini 'Ghost GDP' hypothesis—that productivity gains accrue to fewer people, thereby reducing overall consumer spending—which the speakers agree is a genuine short-term risk if displacement is too rapid, but ultimately argue that historical precedent favors productivity gains as the long-term engine of wealth creation. ### Anthropic Security Impact - Security review feature wiped out $20 billion from cybersecurity stocks like Cloudflare and CrowdStrike - Capabilities like total static code review already existed in environments like Replet - The market panic reflects high valuations rather than entirely new technology. ### Valuation and Market Pricing - CrowdStrike traded at 16x revenues with 31% cash revenue growth projected, illustrating the risk of being priced for perfection - The correction is a logical consequence of increased tail risk when valuations are extraordinary high. ### The Agentic Layer Threat - Only Palunteer among public B2B companies shows revenue acceleration from competitive agents - Most incumbents are 'sprinkling AI dust' and face terminal decline if they do not own the agents in their category. ### Barriers to Agent Adoption - Implementing agents requires massive custom work, training, onboarding, and data cleansing, which existing B2B companies are institutionally resistant to - Hyper-niche startups succeed because they have a small, manageable set of tasks for their agents. ### Consumer Disruption Debate - Speakers debate whether agents will take over personal decisions like ordering food (DoorDash) or content selection (Netflix) - While direct consumer disruption might be clickbaity, the CTO of DoorDash confirmed agent commerce will be transformative to the industry. ### Ghost GDP and Macro View - Ghost GDP suggests productivity gains lead to wealth concentration among fewer workers who then spend less, causing economic shrinkage - Productivity gains are historically good, but short-term dislocation occurs if workers cannot adjust to new jobs quickly enough.