The Future of AI Agents | Jesse Zhang Interview

Decagon founder Jesse Zhang argues customer service is the highest-conviction enterprise AI use case because ROI is instantly quantifiable and human escalation paths eliminate deployment risk.

• Decagon’s discovery process: ask customers exact dollar willingness-to-pay, approval chain, and ROI framing — highest sessions hit low-to-mid six figures, order of magnitude above competing ideas.
• Customer service wins over other AI use cases because resolution rate improvement (15–20% → 50–80%) maps directly to headcount cost, making ROI trivially justifiable to CFOs.
• Built-in human escalation path makes enterprise deployment safe: enterprises can launch to 5% of users with zero downside if AI fails.
• AI cost margins don’t matter now — Zhang’s only rule is no negative gross margins; argues costs will fall exponentially and market share matters more.
• Forward-deployed engineering only makes economic sense at ~$1M+ deal sizes; conflating Palantir’s model with standard startup hands-on support is a mistake spreading through AI startups.
• Fine-tuned small open-source models increasingly replace frontier LLMs for specific agent routing decisions, improving latency and cost without sacrificing quality where intelligence isn’t needed.
• Zhang’s five private AI bets: Cognition, Cursor, Etched (hardware), Pika (video models), Chai (healthcare foundation models).

2025-10-06 · Watch on YouTube