20VC: Five Predictions for a World of Agents | The Ads Business Model Will Die | Biggest Lessons from Working with Elon Musk at Twitter with Parag Agrawal, Parallel

What you'll learn

• 1 (03:32) **Parallel Defined as "Google for Agents"** - Parag explains that agents need web search just like humans, but the technology and business models must be rebuilt from scratch.
• 2 (06:42) **The Signal-to-Noise Optimization Dance** - Parag breaks down how web search allocates compute to save the model from wasted context, and why the optimal balance shifts with model cost.
• 3 (08:14) **Dealing with Agent Ambiguity: Turbo vs. Advanced Modes** - How Parallel lets agents specify their constraints (accuracy, latency, cost) via API parameters, with distinct product modes.
• 4 (10:04) **Current Customer Base: Knowledge Work, Not Just Coding** - Parag reveals that web search is only in ~5% of coding prompts, while law, insurance underwriting, and science are far more web-search-heavy.
• 5 (12:09) **Why Smarter Models Don't Hurt Parallel's Business** - Parag argues that smarter models won't reduce web search demand because parametric memory is lossy and models are being distilled.
• 6 (13:45) **Two Trends: Frontier Models Grow, Smaller Models Get Cheaper** - Parag predicts both the largest frontier models and the smallest efficient models will coexist, with no end to scaling laws in sight.
• 7 (15:21) **Debunking the 90/10 Token/Dollar Split** - Parag disagrees that 90% of tokens will go through open models while 90% of dollars go through frontier models, citing path dependency and the need for competition.

+ Full timestamped outline available in the app