1. Opening and Host Introduction

• Packer Radio opens as a Sequoia partner and introduces today's hosts.
• "We're just the appetizer today — the real main course comes later."
• "Yesterday a founder told me, 'I'll probably arrive around 9:35,' and that turned out to be exactly when Jensen walks on stage."
• Today's goal: Share what we've learned over the past year and offer our perspective on the AI market.

2. The AI Market Today and Tomorrow: A Framework for Thinking

2-1. Don Valentine's Question Framework

• What is happening?
• Why does it matter?
• Why now?
• What should we do about it?

2-2. Market Size and Growth

• "Cloud generated $400 billion in revenue and surpassed the software market. The AI services market starts from an even larger baseline."
• AI threatens the profit pools of both the services and software sectors.
• "Many companies are evolving from simple tools to copilots to autopilots — moving from software budgets into labor budgets."

2-3. Technology Waves and the Arrival of AI

• "This layer cake shows the waves of technology that have been stacking up for decades."
• AI is not only inevitable — it is now imminent.
• "Compute, networks, data, talent — all the conditions are in place."
• "The speed of technology diffusion keeps accelerating."

Three Elements of Technology Diffusion

• People must know about your product,

• They must want it, and

• They must be able to buy it.

• "The moment ChatGPT launched, the entire world took notice."

• "Now 5.6 billion people are connected to the internet. The moment the starting gun fired, the barriers to entry disappeared."

• The physics of technology deployment have fundamentally changed.

3. How to Win in AI: Where and How to Play

3-1. The Value of the Application Layer

• "We care not about unicorns but about revenue and cash flow."
• "Past winners were mostly in the application layer. AI will be no different."
• "But now foundation models are pushing deep into the application layer."

Strategy for Startups

• "Start from the customer. Focus on vertical specialization and functional specialization."
• "Tackle complex problems, problems where humans need to be involved. That is where the competition is."

3-2. The 5% Differentiation AI Companies Need

• "When we look at AI companies, 95% is the same as any other company. The remaining 5% is what's AI-specific."

1) Real Revenue vs. Vibe Revenue

• "Don't be fooled by vibe revenue. Make sure real behavioral change is happening — always check the metrics: Adoption, Engagement, Retention."
• "Customer trust matters more than the product. Earn the trust and the product can improve."

2) Margins

• "Current COGS will keep coming down. Move from tools to outcomes, climb the value chain — prices rise with it, and you eventually secure healthy margins."

3) Data Flywheel

• "If you can't answer which business metrics your data flywheel actually moves, it isn't a real flywheel."
• "This can become the most powerful moat."

4) Nature Abhors a Vacuum

• "Right now the market is making a tremendous sucking sound. The wave of technology adoption overwhelms every macro headwind."
• "If you don't seize this opportunity, someone else will. Now is the time to go full speed."

4. AI Today: A Year in Review from the Customer and Technology Perspective

4-1. User Engagement in AI Apps Surging

• "Last year the daily-to-monthly user ratio for AI apps was low — now ChatGPT has reached Reddit-level engagement."
• "More people are getting real value from AI and weaving it into their daily lives."

4-2. Practical AI Use Cases

• Advertising: "Generating ad copy that is surprisingly accurate and beautiful"

• Education: "Easily visualizing new concepts"

• Healthcare: "Better diagnoses through apps like Open Evidence"

• "We're barely scratching the surface. As AI models grow more powerful, the depth of what they can do will grow too."

4-3. The Leap in Voice AI

• "2024 was the 'Her' moment. Voice generation has fully crossed the uncanny valley."
• "Sesame's voice demo was genuinely astonishing. The gap between science fiction and reality is closing fast."

4-4. Innovation in Coding

• "With the launch of Anthropic's Claude 3.5 Sonnet, AI coding firmly established itself in the market."
• "AI coding is now completely reshaping the accessibility, speed, and economics of software creation."

4-5. Technical Limits and Breakthroughs

• "Pre-training scaling is slowing, but new breakthroughs are emerging: reasoning, synthetic data, tool use, agent scaffolding."
• "AI can now perform increasingly complex tasks."

5. Value Creation in the AI Stack and the Future of Agents

5-1. Value Concentrating in the Application Layer

• "Ultimately, value will concentrate in the application layer."
• "Companies like Harvey and Open Evidence are creating customer-centric value."

5-2. Agent-First Startups

• "The first generation of AI killer apps has arrived. Going forward, many new startups will evolve as agent-first."
• "Agents will progress from prototypes into increasingly robust systems."

Two Evolutionary Paths for Agents

• 1) Orchestration + rigorous testing and evaluation
• 2) End-to-end task-specific tuning

5-3. The Opportunity in Vertical Specialized Agents

• "For founders with deep expertise in a specific domain, vertical agents represent a massive opportunity."
• "In security, AI is already outperforming human penetration testers; in networking, AI troubleshooters are surpassing the best human experts."
• "These data points show that vertical agents can exceed human performance on specific problems."

5-4. The Age of Abundance and AI's Impact

• "The coding domain was first to enter the age of abundance. When labor becomes cheap and plentiful, 'AI slop' could flood the market."
• "Taste will become a scarce and valuable asset."

6. The Mid- to Long-Term AI Outlook: Agent Economies and How We Will Change

6-1. From Agent Swarms to Agent Economies

• "Last year agents were just beginning to evolve into businesses. Now agent swarms are becoming central to the AI stack."
• "Going forward, agents will exchange not just information but also resources, transactions, and trust — forming an agent economy."
• "This economy does not exclude humans. It is a structure where agents and humans work together."

6-2. Three Technical Challenges

1) Persistent Identity

• "If an agent changes every day, you can't trust it. Agents must maintain their own personality and understanding over time."
• "Real memory, self-learning, and consistent behavior are required."

2) Seamless Communication Protocols

• "Imagine the PC era without TCP/IP. We are only now building that protocol layer."
• "Protocols like MCP and others will enable the transmission of information, value, and trust."

3) Security

• "In an era of agents we can never meet in person, trust and security become even more critical."
• "An entirely new industry will emerge in this space."

6-3. How We Will Change: New Mindsets and Ways of Working

1) Probabilistic (Not Deterministic) Mindset

• "Computers no longer always give the same answer. AI may or may not remember that the number was 73."
• "This is a completely different way of thinking from the past."

2) Manager Mindset

• "Understanding what agents can and cannot do becomes essential."
• "Now management matters more — complex oversight, feedback, and process gatekeeping."

3) Greater Leverage, Less Certainty

• "We can accomplish more, but we must manage uncertainty and risk."
• "The people in this room are the ones best positioned to adapt to this world."

6-4. The Future AI Will Create

• "Individual functions will be replaced by agents, and gradually entire processes will be completed by AI."
• "There are no solo unicorns yet, but companies are emerging that grow faster with far fewer people."
• "Ultimately, individual work, corporate structure, and the economy as a whole will be recreated."

7. Closing

• "Thank you for joining us at AI Ascent today. I'm truly excited for the journey ahead."

🔑 Key Takeaways

• Explosive growth of the AI market
• Value concentrating in the application layer
• Agents and agent economies
• Data flywheels, trust, and margins
• Accelerating technology diffusion
• Vertical specialization and solving complex problems
• Probabilistic thinking and the evolving manager role
• The importance of security and trust
• The age of abundance and the value of taste

"Now is the time to go full speed."

"Customer trust matters more than the product."

"AI is now imminent. All the conditions are in place."

"The speed of technology diffusion keeps accelerating."

"Vertical agents can exceed human performance on specific problems."

"This is an era of greater leverage and less certainty."

"AI will recreate our work, our companies, and the economy as a whole."

That concludes a complete, chronological summary of all the major highlights from the Sequoia AI Ascent 2025 keynote, along with the most memorable lines from the stage. 🚀