He saved OpenAI, invented the “Like” button, and built Google Maps: Bret Taylor on the future of careers, coding, agents, and more
Brett Taylor, co-founder & CEO of Sierra and Chairman of OpenAI, discusses his career journey, including co-creating Google Maps and inventing the like button. He shares insights on AI's future, the rise of agents, and effective strategies for founders.
Deep Dive Analysis
13 Topic Outline
Bret Taylor's First Major Product Mistake: Google Local
The Genesis and Success of Google Maps
Mindsets for Success Across Diverse Roles
Lessons from FriendFeed's Failure and Marketing
How to Discern Good Advice and Build Judgment
The Future of Coding in an AI-Driven World
Preparing Children for an AI-Transformed Educational Landscape
Three Key Segments of the AI Market
The Shift to Outcome-Based Pricing in AI
Achieving Real Productivity Gains with AI
Effective Go-to-Market Strategies for AI Products
Lightning Round: Books, Movies, Products, and Motto
The Origin Story of the Like Button
6 Key Concepts
Systems Thinking (in Software)
This involves designing a system that can produce a desired, delightful output even when given uncontrolled or messy input, rather than just designing a perfect static mock-up. It emphasizes understanding how different components interact to achieve a goal at scale, especially when dealing with unpredictable real-world data.
Programming System for LLMs
A conceptual new approach to software development where the focus shifts from human-centric programming languages (like Python, designed for ergonomics) to systems optimized for AI code generation. This system would prioritize verifiability, robustness, and leverage AI to supervise AI (e.g., for code reviews), allowing humans to operate a 'code generating machine' with high leverage.
AI Market Segments
The AI market can be broadly divided into three areas: frontier models (requiring massive CapEx, consolidating to hyperscalers), tooling (selling 'pickaxes' like data labeling or eval tools, but risky due to proximity to infrastructure providers), and applied AI/agents (building products that achieve specific business outcomes autonomously, seen as the future 'app' form factor).
Agent-Pilled
A term describing the strong belief that AI agents will fundamentally transform the software industry by autonomously accomplishing jobs, rather than merely making individuals slightly more productive. This shift is expected to drive significant, measurable productivity gains and lead to a widespread adoption of outcome-based pricing models.
Outcome-Based Pricing
A business model where a software vendor charges customers based on the measurable business results or value achieved by the product, rather than traditional metrics like features, seats, or usage. This model is particularly effective for autonomous AI agents where impact can be directly attributed and measured, aligning the vendor's and customer's business goals.
Context Engineering (for AI Models)
A technique used to improve the accuracy and performance of AI models, particularly coding agents. It involves performing root cause analysis on incorrect outputs to identify missing context and then providing that necessary context to the model, aiming to prevent similar errors in future generations and create a virtuous cycle of improvement.
11 Questions Answered
After a tough product review for Google Local, which was a 'me too' version of Yahoo Yellow Pages, Bret was given a second chance. He and his team inverted the product's hierarchy, making the map the central canvas, which led to the creation of Google Maps, integrating mapping, local search, and driving directions into a new, compelling experience.
A key mindset is having a flexible view of his own identity, seeing himself as a 'builder' rather than strictly an engineer or product person. He also adopted Sheryl Sandberg's advice to ask daily, 'What is the most impactful thing I can do today?' which shifted his focus from personal preferences to maximizing company success.
Founders must cultivate intellectual honesty to avoid incorrect storytelling about why their product succeeds or fails. It's crucial to question one's own biases (e.g., an engineer defaulting to engineering solutions) and engage in very real conversations with co-founders and leadership to accurately identify the root cause of problems.
It requires good judgment, as confidence in advice doesn't correlate with quality. Seek out common answers when asking 'who should I talk to?' and always ask 'why' to understand the framework behind the advice, recognizing that most advice is anecdotal rather than statistically significant.
Yes, studying computer science remains valuable for developing systems thinking, which is crucial for operating future 'code generating machines.' While the act of typing code may change, understanding what's hard, easy, possible, and impossible in systems will be essential for leveraging AI effectively.
Parents should encourage their children to integrate AI, like ChatGPT, into their lives as a utility for learning and problem-solving. AI can act as a personalized tutor, adapting to different learning styles and democratizing access to advanced educational resources, though the education system needs to adapt its evaluation methods.
The market has three main segments: frontier models (dominated by hyperscalers due to CapEx), tooling (supporting AI development, but with risk of competition from large providers), and applied AI/agents (building products that deliver specific business outcomes autonomously, offering significant value and becoming the new 'app' form factor).
Outcome-based pricing charges customers based on the measurable business results achieved by the product. Sierra, for example, charges for customer service interactions that their AI agents successfully resolve or 'contain,' aligning their business model directly with the cost savings and value delivered to their clients.
While AI tools like Cursor show promise, current productivity gains are often counterintuitive because AI-generated code can be incorrect, requiring significant cognitive load to fix. To achieve real gains, companies need to implement systems for 'context engineering' and root cause analysis to continuously improve the AI's output and provide necessary context.
Founders should choose a go-to-market model aligned with their product category: developer-led (for platform products appealing to engineers with latitude), product-led growth (when user and buyer are the same, common for small businesses), or direct sales (when the buyer and user are different, increasingly relevant for many AI enterprise solutions).
The 'like' button was invented at FriendFeed as a 'one-click comment' to reduce single-word acknowledgments (like 'cool' or 'wow') in comment threads. The initial design used a heart icon, but it was deemed too strong for all contexts, leading to the more neutral 'like' sentiment which could apply to a broader range of posts, including tragedies.
30 Actionable Insights
1. Prioritize High-Impact Tasks Daily
Every morning, ask yourself what is the most impactful thing you can do today to maximize the likelihood of achieving your goals, even if it’s something you don’t initially enjoy.
2. Maintain Flexible Founder Identity
As a founder, cultivate a flexible view of your identity, allowing yourself to transform into whatever role the company needs you to be at any given time, from selling to engineering.
3. Practice Intellectual Honesty in Problem Diagnosis
When diagnosing problems (e.g., losing a deal), maintain intellectual honesty to uncover the true root cause, rather than accepting surface-level explanations that might be misleading.
4. Question Solutions Aligned with Your Strengths
Be self-aware that you might subconsciously favor solutions that align with your personal strengths; by default, question if your ‘superpower’ is truly the best solution or if it’s chosen out of comfort.
5. Understand the “Why” Behind Advice
When receiving advice, repeatedly ask ‘why’ to understand the underlying framework and experiences informing the counsel, allowing you to apply it with nuance rather than as a rigid rule.
6. Reflect to Improve Judgment
After making a bad decision, dedicate time to reflect on it, understand the ‘why,’ and continuously work to improve your judgment, as this is crucial for entrepreneurial and product management success.
7. Solicit Diverse External Advice
Actively seek advice from a diverse range of people outside your immediate circle to gain broader perspectives and identify potential blind spots in your strategy.
8. Discern Advice Quality from Confidence
Recognize that the confidence with which someone delivers advice does not correlate with its quality; exercise good judgment to evaluate the substance of the advice.
9. Create Differentiated Product Experiences
Avoid simply creating a better copy of an existing product; instead, aim to build an entirely new, compelling, and differentiated experience to increase your chances of success.
10. Reimagine, Don’t Just Digitize
When developing new products with new technologies, disassemble existing concepts and reassemble them into an entirely new experience, rather than just digitizing what came before, to create meaningful breakthroughs.
11. Design Systems for Uncontrolled Input
When designing products, especially those with user-generated content, focus on creating systems that can produce a delightful experience even with messy, uncontrolled input.
12. Add “Sizzle” to Your Product
Incorporate elements that, while not the core value, create excitement and viral moments, acting as ‘sizzle to the steak’ to attract widespread attention.
13. Distinguish Initial Use from Enduring Value
When designing products, understand the difference between why users initially decide to use it and what provides its enduring value, as these are related but distinct.
14. Learn Systems Thinking via Computer Science
Study computer science to develop strong systems thinking, which is crucial for operating code-generating AI machines and solving complex problems at scale, even as the act of coding evolves.
15. Maintain Loose Attachment to Job Methods
Cultivate a very loose attachment to the specific methods and tools you use to perform your job, as AI will significantly transform how work is done.
16. Leverage AI as a Personalized Tutor
Encourage children and students to use AI tools like ChatGPT as personalized tutors that can adapt to their learning style (visual, audio, reading) and provide tailored explanations and quizzes.
17. Encourage Constructive AI Use in Learning
Actively encourage children to use AI tools like ChatGPT constructively in their learning process, treating it as a powerful utility for problem-solving and skill acquisition.
18. Avoid Frontier Model Startups
Startups should generally avoid building frontier or foundation AI models due to the immense CapEx requirements and the rapid deterioration of model value, making it an unviable business model for most.
19. Assess Risks in AI Tooling Market
When building AI tooling, be aware of the risk that large foundation model providers may offer competing products; differentiate strongly to ensure continued customer choice.
20. Build Applied AI Agents
For startups, focus on building applied AI agents that deliver specific business outcomes, as this market segment offers higher margins and will likely evolve into a product-centric SaaS model.
21. Adopt Outcomes-Based Pricing
Shift towards outcomes-based pricing for your software, especially AI agents, as it aligns your business model with customer value, drives true productivity, and is measurable.
22. Align Pricing with Customer Business Outcomes
Structure your pricing model to directly align with the measurable business outcomes achieved by your customers, such as paying per resolved customer service interaction, to foster partnership and shared success.
23. Implement AI Self-Reflection for Robustness
Improve the robustness of AI systems by having AI supervise AI, such as using one agent to find errors in another, to significantly increase accuracy and reliability.
24. Root Cause AI Errors for System Improvement
When AI tools produce incorrect outputs, perform root cause analysis to understand why (e.g., lack of context) and adjust the system or context engineering to prevent future errors, rather than just fixing individual instances.
25. Create Virtuous Cycles with AI for Improvement
Design AI systems to enable a virtuous cycle of continuous improvement, where AI helps identify opportunities, diagnose frustrations, and suggest new capabilities to enhance performance.
26. Select Go-to-Market Based on Purchase Process
Carefully choose your go-to-market strategy (developer-led, product-led growth, direct sales) by thoroughly considering the specific purchasing and evaluation processes of your target customers.
27. Reconsider Direct Sales for AI
For AI products where the buyer and user are often different, consider leveraging direct sales as a go-to-market strategy, as it has become more relevant in the AI market.
28. Study “Jobs to Be Done” Framework
Read ‘Competing Against Luck’ to understand the ‘Jobs to Be Done’ framework, which provides a valuable perspective on delivering product value, and consider using AI to summarize it.
29. Read “Endurance” for Grit
Read ‘Endurance,’ the true story of Shackleton’s expedition, for an unparalleled example of grit and resilience, which can be inspiring during challenging entrepreneurial times.
30. Invent the Future
Embrace the motto ‘The best way to predict the future is to invent it’ to drive an entrepreneurial mindset focused on building and creating new things.
6 Key Quotes
I think the whole market is going to go towards agents. I think the whole market is going towards outcomes-based pricing. It's just so obviously the correct way to build and sell software.
Bret Taylor
What is the most impactful thing I can do today?
Sheryl Sandberg (as taught to Bret Taylor)
I think that a lot of education doesn't presume you have a super intelligence in your pocket.
Bret Taylor
If you think the thing that you've been doing your whole career is the way to fix your problem, it's at least 30% likely that you've chosen that because of comfort and familiarity, not truth.
Bret Taylor
The best way to predict the future is to invent it.
Alan Kay (as attributed by Bret Taylor)
I don't think mobile phones are great in school or great for kids. And I, I, I personally advocate for waiting a long time, but I think that ChatGPT is more like Google search.
Bret Taylor