Sequoia Essay: "Generative AI’s Act o1"
Sequoia Capital’s essays, such as "RIP Good Times" (2008) and "Black Swan" (2020), are essential reading for founders navigating market turbulence, emphasizing disciplined operations, profitability, and cash conservation. Sequoia’s long-term vision is reflected in works like “Elements of Enduring Companies” (2013), which stresses solving real problems, building strong cultures, and hiring top talent. Insights from "Measuring What Matters" (2019) and "The Sequoia Guide to Market Expansion" focus on tracking key metrics and thoughtful scaling.
In "Generative AI: A Creative New World" (2022), Sequoia highlighted the transformative power of AI, urging businesses to harness AI-driven innovation to redefine creativity and automation across industries.
The first phase of generative AI, or "Act 1," was driven primarily by advancements in technology. One year ago Sequoia’s essay "Generative AI’s Act Two" was published. In this essay, Sequoia concluded that the market is transitioning into "Act 2," where the focus shifts to a customer-centric approach. In this phase, generative AI will aim to solve real human problems comprehensively, addressing needs from start to finish.
On October 9th of this year, Sequoia’s essay "Generative AI’s Act o1" appeared. It is subtitled "The Agentic Reasoning Era Begins".
Sequoia's essays are highly recommended reading.
The following summary of "Act o1" was created with the help of ChatGPT 4.0:
Shift in focus from System 1 to System 2 thinking: AI is evolving from "thinking fast" (rapid, pattern-based responses) to "thinking slow" (deliberate reasoning at inference time), enabling agentic applications.
Consolidation of Generative AI's foundation layer: Market stabilization with key players like Microsoft/OpenAI, AWS/Anthropic, Meta, and Google/DeepMind. These scaled players control vast capital and computational resources.
Emergence of the reasoning layer: The next frontier is developing AI systems with reasoning abilities at inference time, inspired by models like AlphaGo. These new cognitive architectures will shape interactions with AI.
Strawberry (o1) model: OpenAI's most important 2024 update, with enhanced general reasoning using inference-time compute, marking a major advancement in AI capabilities for logic-based tasks.
AlphaGo and LLM analogy: Like AlphaGo, LLMs are evolving to stop, think, and reason, using a search process to explore potential future outcomes at inference time, leading to better responses over time.
Challenges of reasoning in open-ended tasks: Constructing value functions in tasks like essay writing is complex, as reasoning needs a scoring mechanism. Strawberry excels in logical domains but struggles in open-ended, unstructured areas.
System 2 thinking in AI: While System 1 (pre-trained responses) is effective for tasks like fact recall, System 2 is crucial for complex problem-solving, requiring models to reason through novel situations.
New scaling law for AI models: AI's reasoning abilities scale with inference-time compute. The more time given for reasoning, the better the model performs, opening the door to solving complex problems like the Riemann Hypothesis.
Application layer opportunities: While foundation models are consolidating, the messy real-world application layer remains a key area for value creation, with domain-specific cognitive architectures driving breakthroughs.
Rise of agentic applications: New AI apps like Harvey (AI lawyer), Glean (AI assistant), Factory (AI software engineer), and Sierra (AI customer support) are emerging across industries, lowering service delivery costs and expanding markets.
Service-as-a-software paradigm: AI is transforming the services market, with applications performing labor-intensive tasks, moving beyond traditional software models toward selling work outcomes instead of software licenses.
Investor focus: The most promising areas for investors are application-layer companies and developer tools, as they bring AI’s reasoning capabilities into the real world.
Future of AI reasoning and multi-agent systems: Research continues to advance reasoning and inference-time compute. The rise of multi-agent systems, like Factory’s "droids," will further expand AI's capabilities in real-world tasks.
AGI possibilities: The next phase of AI development might simulate perception, reasoning, and action in ways that feel like independent thought, potentially leading toward artificial general intelligence (AGI).
Conclusion
Sequoia's essays offer invaluable guidance for startups at every stage, delivering not only survival strategies for tough times but also actionable frameworks for building durable, innovative companies. Their approach consistently sets a high bar for founders and leaders, enabling them to adapt, scale, and leverage transformative technologies—like generative AI—that drive significant market impact.