Semafor1h 1m
Former Twitter CEO Parag Agarwal on X, Elon, & Building A New Internet For AI Agents | Mixed Signals
Parag Agarwal, Elon, Ben Smith, Max Tani
Interview
Core Thesis: The Shift to an Agent-Centric Internet
- Parag Agarwal posits that AI agents will consume the web at a scale 1,000x greater than humans due to the removal of human time and attention constraints.
- Current web infrastructure optimized for human browsing (e.g., "10 billion" queries/day) cannot function at the required volume for agents, necessitating new technology generations.
- Unlike humans, agents are not limited by patience; they require search engines capable of multi-step, high-frequency queries within sub-second latency windows.
- Agents currently lack real-time data (e.g., recent sports scores or school records), breaking the "mirage" of omniscience unless granted direct, optimized web access.
Technical and Operational Changes Required
- Search ranking logic must shift from "above the fold" human UX to delivering specific, authoritative data points (e.g., extracting a revenue figure from page 77 of a 200-page PDF).
- Agent input queries are expected to be more precise and context-rich compared to the "underspecified" two-to-three keyword searches typical of human users.
- Parallel, the new company, optimizes for agents by pre-processing content to reduce token waste and computational costs compared to agents "fumbling" through the web.
- Agarwal emphasizes that agents are machines, not humans; building products requires optimizing for compute outputs rather than anthropomorphizing the systems.
Economic Implications for Media and Content Creators
- Traditional business models (ads and subscriptions) face existential risk as agents do not click ads and do not possess "wallets" for subscriptions.
- Without new models, content owners will be incentivized to block agents, which Agarwal labels a "value-destructive" move that shrinks the open web.
- The "advertising engine" is described as the most efficient mechanism for differential pricing, allowing platforms to subsidize 90% of users with free access while monetizing high-value segments.
- Current subscription models ($20/month) are inefficient for extracting maximum value from high-willingness-to-pay users (e.g., hedge funds paying $1,000/month for specific data).
- Agarwal proposes a new B2B-like economy where content pricing scales based on the value of the information to the agent's task, not the user interface.
- He suggests a future where a hedge fund agent pays significantly more for a news article than a casual reader agent because the former's reliance on that data is higher.
Strategic Vision and Market Dynamics
- Parallel is positioning itself as a "Rebel Alliance" against content-ownership models that sell data exclusively to large labs (OpenAI, Google), fearing a future where the open ecosystem is priced out of renewal cycles.
- The goal is to align incentives so that high-quality human-generated content is rewarded proportionally to the compute (GPU) usage required to process it.
- Agarwal argues that while the "pie" of web value will grow 1,000x, the distribution mechanism must evolve to prevent the open web from becoming a "free gift" for big tech.
- He views the current trajectory of deals between media and labs as unsustainable, predicting that value capture for media partners could drop 5–10x annually as lab costs rise.
Reflections on X (Twitter) and Social Platforms
- Agarwal asserts that social platforms will survive the AI era because human-to-human connection and the "experiential" nature of scrolling remain durable needs.
- He identifies "Community Notes" (formerly Birdwatch) as the only successful product iteration from his tenure that he continues to support.
- He rejects the notion that X/Twitter has been the primary driver of US political shifts to the left (his tenure) or the right (Musk era), citing its relatively small share of total time-on-app compared to short-video platforms.
- Agarwal admits he would have focused more on product iteration, transparency, and deeper integration of AI into ranking algorithms had he remained CEO for the past four years.
- He declined to state definitively whether the current version of X is "good for the world," noting the difficulty in measuring its outsized influence on other influencers versus its total audience size.
Host Commentary and Outlook
- Hosts acknowledge the "optimism vs. reality" gap in Agarwal's vision, noting the historical difficulty of bridging the divide between new technological potential and viable business models.
- The discussion highlights a potential return to a "free market" for information pricing, contrasting it with the current concentrated power of a few tech labs to buy data.
- Max Tani expresses caution regarding the "gulf" between current revenue and the projected agent economy, comparing it to the dot-com bubble's unfulfilled promises.
- Ben Smith notes the resilience of the X platform despite significant leadership changes, suggesting the core utility of short-burst information sharing remains unmatched by competitors.