He saved OpenAI, invented the “Like” button, and built Google Maps: Bret Taylor (Sierra)

Key Concepts

• Agents: AI-powered systems designed to autonomously perform specific tasks or jobs.
• Outcomes-Based Pricing: A pricing model where customers pay based on the actual value or results achieved by the software or service.
• Frontier/Foundation Models: Large, powerful AI models requiring significant capital expenditure (CapEx) to develop.
• Applied AI: The application of AI technologies to solve specific business problems or create new products and services.
• Model Context Protocol (MCP): A method for providing additional context to AI models to improve their decision-making and performance.
• Product-Led Growth (PLG): A go-to-market strategy where the product itself drives customer acquisition, activation, and retention.
• Direct Sales: A traditional sales approach involving direct interaction between sales representatives and potential customers.
• Systems Thinking: A holistic approach to problem-solving that considers the interconnectedness of different components within a system.

Fail Corner: The Google Local Story

• Initial Product: Bret Taylor's first prominent mistake was as a product manager for Google Local, a "me too" version of Yahoo Yellow Pages. It was essentially grafting Yellow Pages search on top of Google Search.
• Lack of Differentiation: The product was not differentiated and didn't offer a compelling reason for users to switch from existing solutions like Yahoo Yellow Pages or the traditional Yellow Pages.
• Tough Product Review: The product received a tough review from Marissa Mayer and Larry Page.
• Second Chance & Google Maps: Taylor was given another shot to create a V2, which led to the development of Google Maps. The team inverted the hierarchy and made the map the canvas.
• Key Integration: Google Maps integrated mapping, local search, and driving directions, which were previously separate product categories.
• Launch Success: Google Maps had 10 million users on the first day.
• Satellite Imagery Integration: Integrating satellite imagery from Keyhole (later Google Earth) led to 90 million users on the same day.
• Product Lessons:
  • Creating an entirely new experience is more effective than just making a better copy.
  • New technologies should create native experiences rather than digitizing what came before.
  • "Sizzle to the steak": Satellite imagery wasn't the most important part of Google Maps, but it created a viral moment.
  • Distinguish between why people decide to use a product and its enduring value.

Mindsets and Habits for Success

• Flexible Identity: Having a flexible view of one's own identity and being able to transform into what the company needs.
• Builder Mentality: Identifying as a builder of products and companies.
• Confluence of Technology and Capitalism: Believing in the power of technology and capitalism to produce incredible outcomes for customers.
• Sheryl Sandberg's Influence: Sheryl Sandberg's advice to focus on the most impactful thing to do each day significantly changed Taylor's approach to new jobs.
• Impact-Driven: Deriving joy from having an impact and being receptive to doing things that may not be one's favorite.
• Intellectual Honesty: Being intellectually honest about why a product is not succeeding and avoiding incorrect storytelling.
• Self-Awareness: Being self-aware of one's strengths and weaknesses and avoiding the tendency to apply one's superpower to every problem.
• Good Co-founder/Leadership Team: Having a good co-founder and leadership team to ensure that the correct problems are being addressed.
• Seeking External Advice: Soliciting advice from the right people and being open to hearing what one doesn't want to hear.
• Understanding the "Why": When seeking advice, asking "why" to understand the framework behind the advice.
• Good Judgment: Developing good judgment through self-reflection and holding oneself accountable for bad decisions.

FriendFeed Failure

• Engineering Focus: FriendFeed, a social network co-founded by Taylor, had a team primarily composed of engineers who focused on product development.
• Lack of Distribution Focus: The company failed to focus on distribution and getting celebrities and public figures onto the platform, unlike Twitter.
• Celebrity Adoption: Twitter gained traction when Obama, Ashton Kutcher, and Oprah Winfrey joined the platform.
• Product vs. Distribution: FriendFeed had a better product but lost to Twitter due to a lack of focus on distribution.
• Google Influence: Taylor attributes this failure to the Google environment, where distribution was less scrutinized due to AdWords.

Whose Advice to Listen To

• Good Judgment: It comes down to good judgment and being a judge of people's character.
• Confidence vs. Quality: There's not a strong correlation between the confidence with which someone expresses an opinion and the quality of that opinion.
• Asking for Referrals: Asking people who to talk to get good advice.
• Understanding the Framework: Asking "why" to understand the framework someone is using to give advice.
• Statistical Significance: Recognizing that most advice is not statistically significant.

The Future of Coding

• Computer Science Fundamentals: Studying computer science is valuable for understanding systems thinking, algorithms, and complexity theory.
• Code-Generating Machines: The act of creating software will transform from typing into a terminal to operating a code-generating machine.
• Systems Thinking: Operating a code-generating machine requires systems thinking.
• AI Assistance: AI will facilitate creating software, but the operator's job is to make a product or solve a problem.
• Loose Attachment: Having a loose attachment to the way we do our jobs.
• New Programming System: There should be a new programming system designed for LLMs rather than humans.
• Human Programmer Productivity: Many abstractions are based on human programmer productivity.
• Code Verification: The focus should be on verifying that the generated code does what it's intended to do and being able to change it easily.
• Rust Example: Rust's compile-time memory safety is an example of a desirable feature.
• Leverage for Humans: Enabling humans to have as much leverage as possible by using computers to do the work.
• AI Supervision: AI supervising AI and doing code reviews.
• Loosening Constraints: Loosening constraints like coding being free.
• Robustness and Agility: The focus should be on building increasingly complex systems and changing them with agility.

Teaching Kids in the Age of AI

• AI Integration: Encouraging kids to make AI a part of their lives.
• Calculus Exam Analogy: Drawing an analogy to the introduction of calculators in calculus exams and the need to rethink problems.
• Broken Evaluation Mechanisms: Recognizing that many evaluation mechanisms are broken by the existence of ChatGPT.
• Teaching Thinking and Learning: Focusing on teaching kids how to think and learn.
• Personalized Tutor: Viewing AI models as personalized tutors that can teach in different styles.
• Amplifier of Agency: Recognizing that AI is an amplifier of agency for kids who have aspirations to learn.
• Empathy for Teachers: Having empathy for public school teachers who are facing challenges due to the rapid advancement of AI.
• Democratizing Force: Viewing AI as a democratizing force that can provide access to tutoring and advanced learning opportunities.

AI Market Dynamics

• Three Segments:
  • Frontier Model Market: Dominated by hyperscalers and big labs due to high CapEx requirements. Startups are unlikely to succeed in this market.
  • Tooling Market: Selling "pickaxes in the Gold Rush" (data labeling, data platforms, eval tools). This market is at risk of being obviated by infrastructure providers.
  • Applied AI Market: Building agents to solve specific business problems. This market is similar to the SaaS market and will likely have higher margins.
• Agents as the New App: Believing that agents will be the product form factor in the AI market.
• Technical Moat: The technical moat of orchestrating an agentic process will likely become easier over time.
• Focus on Business Outcomes: Agent companies will increasingly focus on workflows and business outcomes rather than the underlying technology.

Agent-Pilled Perspective

• Productivity Gains: Agents will drive significant productivity gains in the economy, similar to the early days of computing.
• Autonomous Job Accomplishment: Software is going from helping individuals be slightly more productive to actually accomplishing a job autonomously.
• Measurable Value: Agents provide measurable value, making it easier to justify their cost.
• Outcomes-Based Pricing: The market will move towards outcomes-based pricing because it's the correct way to build and sell software.

Outcomes-Based Pricing Explained

• Definition: A pricing model where customers pay based on the actual value or results achieved by the software or service.
• Sierra Example: Sierra charges customers based on the number of customer service calls deflected by its AI agents.
• Alignment with Customer's Business Model: Aligning the business model with the customer's business model.
• Consumption-Based vs. Outcomes-Based: Outcomes-based pricing is different from usage-based pricing (e.g., tokens) because usage doesn't necessarily correlate with value.
• Partner vs. Vendor: Aspiring to be a partner rather than a vendor.
• Customer-Centric Orientation: Creating an extremely customer-centric orientation.

Productivity Gains from AI

• Immature Tools: The tools and products are somewhat immature.
• Code Review Challenges: Reviewing code produced by AI can be more difficult than editing code written by oneself.
• AI Supervision: Having AI supervise AI is effective.
• Root Cause Analysis: Focusing on root cause analysis to improve the code generated by AI.
• Context Engineering: Providing the right context to AI models to improve their decision-making.
• Virtuous Cycle of Improvement: Creating a system that lets customers create a virtuous cycle of improvement.
• Sierra's Gains: Customers see anywhere between 50 and 90% of their customer service interactions completely automated.

Go-to-Market Strategies for AI

• Developer-Led: Appealing to individual engineers within the department of the CTO. Works for platform products and startups.
• Product-Led Growth (PLG): Users can sign up from the website and buy with a credit card. Works when the user and buyer are the same person (e.g., small business software).
• Direct Sales: Selling into large lines of business in a traditional sales motion. Requires engaging with the buyer of the software.
• First Principles Thinking: Thinking through the process of purchasing and evaluating the value of the software.
• Leveraging Direct Sales: Considering leveraging direct sales more often.

Lightning Round

• Recommended Books:
  • Competing Against Luck by Clayton Christensen (Jobs to be Done framework)
  • Endurance (story of Shackleton's trip to the South Pole)
• Favorite Recent Movie: Inception
• Favorite Product: Cursor
• Life Motto: "The best way to predict the future is to invent it" (attributed to Alan Kay)
• Like Button Story: The original framing was "one-click comment." The first version had a heart, but it was replaced with "like" because it was more neutral.

Conclusion

Bret Taylor's insights highlight the transformative potential of AI, particularly through the development of autonomous agents. He emphasizes the importance of focusing on business outcomes, adopting outcomes-based pricing, and understanding the underlying systems and context required for AI to be truly effective. His experiences, both successful and unsuccessful, offer valuable lessons for entrepreneurs navigating the evolving AI landscape.