From Copilots to Autopilots, Agentic Apps, Service as a software: What lies ahead

Notes from Sequoia Capital's latest essay on GenAI future ͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­͏   ­ Forwarded this email? [Subscribe here]() for more [From Copilots to Autopilots, Agentic Apps, Service as a software: What lies ahead]( Notes from Sequoia Capital's latest essay on GenAI future [Ashish Sinha](cnha) Oct 11   [READ IN APP](   Sequoia Capital has published its new essay on Gen AI opportunities and is an excellent read. [Upgrade to paid]( Sharing 7 big ideas from the [essay]( (tl;dr: Application layers are a WIN / Service-as-a-software is for real) --------------------------------------------------------------- The foundation layer of the Generative AI market is stabilizing in an equilibrium with a key set of scaled players and alliances, including Microsoft/OpenAI, AWS/Anthropic, Meta and Google/DeepMind. Only scaled players with economic engines and access to vast sums of capital remain in play. --------------------------------------------------------------- - The Rise of Inference-Time ComputeOpenAI's o1 model (Strawberry) introduces a new paradigm of "thinking slow" - using inference-time compute for complex reasoning. This approach, inspired by AlphaGo, allows AI to deliberate and problem-solve beyond simple pattern matching, potentially leading to breakthroughs in fields like mathematics and biology. [Share]( [A large room with a sign that says, what are we doing today?]( - Emergence of a New Scaling LawThe [o1 paper]( reveals a new scaling law: increased inference-time compute correlates with better reasoning capabilities. This shift may lead to the development of "inference clouds" that can dynamically scale compute based on task complexity, moving away from massive pre-training clusters. - The Evolution of AI Applications AI apps have evolved from simple "wrappers" on foundation models to sophisticated systems with custom cognitive architectures. These architectures typically include multiple foundation models, routing mechanisms, vector/graph databases for RAG, compliance guardrails, and application logic mimicking human reasoning workflows. - Service-as-a-Software Business ModelAI is enabling a shift from software-as-a-service to service-as-software. Companies like Sierra are selling outcomes (e.g., $ per customer issue resolved) rather than traditional software licenses. This approach targets the multi-trillion dollar services market, vastly expanding the addressable market for AI companies. - Emergence of Agentic Applications A new cohort of AI applications is emerging that can perform complex knowledge work. Examples include Sierra (conversational AI platform for businesses), Harvey (AI lawyer), Factory (AI software engineer), and XBOW (AI pentester). These apps are expanding existing markets and creating new ones by dramatically reducing the marginal cost of delivering high-skill services. - AI-Native Product DevelopmentAI-native startups are demonstrating the power of rethinking products from the ground up. Day.ai, for instance, has created an auto-generated CRM that configures itself based on email and calendar data, challenging incumbent CRM systems that require extensive manual configuration. - The Copilot to Autopilot TransitionMany AI companies are deploying their solutions first as copilots (human-in-the-loop) before transitioning to autopilots (fully autonomous). This approach allows companies to build trust and improve their systems through real-world usage before attempting full automation of complex tasks. You're currently a free subscriber to [Masters of AI by NextBigWhat](. For the full experience, [upgrade your subscription.]( [Upgrade to paid](   [Like]( [Comment]( [Restack](   © 2024 Zakti Techmedia Private Limited 677, 27th main, HSR Layout, Bangalore-560102 [Unsubscribe]() [Get the app]( writing]()