Nature doesn't like vacuum: Sequoia's memo on AI (and trillion dollar opportunities)
+ rise of vibe revenue
Folks at Sequoia shared some very interesting take on rise of AI and importantly, the opportunities that lies ahead at AI Ascent 2025 keynote.
Sharing big ideas from their talk:
The Immense Scale of the AI Opportunity
The market AI addresses, starting with services and software, is at least an order of magnitude larger than the software market at the beginning of the cloud transition.
The endpoint for AI services and software profit pools has the chance to be absolutely massive in 10-20 years. AI is actively attacking both these profit pools simultaneously.
Companies can progress from selling a tool to selling software, then becoming a co-pilot, then an autopilot, ultimately transitioning from selling into a software budget to selling an outcome or selling work into a labor budget. Both these Total Addressable Markets (TAMs) are available.
Why Now? The Unprecedented Speed of Adoption
AI is imminent, not just inevitable. The necessary precedent conditions—compute, networks, data distribution, and talent—are all in place.
Technology waves are additive and accelerating. Things are happening faster than ever before.
The physics of distribution have fundamentally changed.
Rapid Discovery & Spread: Platforms like Reddit and X, with 1.2 to 1.8 billion combined monthly active users, provide potent channels to discover new technology, which barely existed during prior transitions.
Ubiquitous Connectivity: Internet connectivity is now effectively global, reaching 5.6 billion people across nearly every household and business. During the cloud transition, only 200 million people were connected.
These factors mean the "rails are in place". When the "starting gun went off," there were no barriers to adoption. This new reality of technology distribution facilitates unprecedented speed.
Where Value Accrues: The Application Layer Battleground
Value, similar to prior technology transitions, is primarily created and will ultimately accrue at the application layer.
This layer is the intensifying battleground. Foundation models are increasingly competing within the application layer, powered by developments like the second scaling law, test time compute, reasoning with tool use, and inter-agent communication.
A first cohort of "killer apps" in AI has emerged, including examples like ChatGPT, Harvey, Glean, Sierra, Cursor, and A Bridge, spanning a rich and diverse set of end markets.
Winning in the Application Layer: Strategic Approaches for Builders
Building an AI company is 95% standard company building (solving important problems, attracting great people) and 5% AI-specific.
Think Customer Back, Not Tech Out: Build moats by starting from the customer problem and perspective. Provide end-to-end solutions that solve the problem directly, rather than just delivering a tool.
Embrace Vertical and Function Specificity: Focus on specific, complex problems within a domain or function that may require sophisticated solutions or even a human in the loop. Be "of the industry, for the industry".
Build Data Flywheels (That Matter): Use proprietary usage data from your product to create a data moat. Critically, this data flywheel must tie directly to a business metric; otherwise, it's ineffective ("bullshit"). This is one of the best moats you can build.
What Matters: The Metrics of Success
Real Revenue vs. "Vibe Revenue": Be vigilant about the quality of revenue. Distinguish "tire kicking" from true, durable behavior change. Inspect adoption, engagement, and retention metrics to understand what customers are actually doing. Avoid self-delusion about revenue quality.
Customer Trust is Paramount: At this stage in the market cycle, earning customer trust is more important than the product itself. If customers trust you, they will believe you can improve the product. Good vibes with customers are essential.
Margins Have a Path: While current gross margins may be low, look for a clear path to healthy margins over time. The cost component (COGS, e.g., cost per token) is decreasing. The price component should increase as you successfully move from selling tools to selling outcomes and capturing more value up the chain.
The Agent Wave: From Prototypes to an Economy
Agents are Today's Focus: AI Ascent a year ago focused on agents beginning to form into businesses; they are now a critical part of the AI stack, with agents working together.
Agent-First Companies Emerging: Many new AI companies will be agent-first, with agents evolving from prototypes to robust systems.
Building Robust Agents: Companies are pursuing two primary paths: orchestration with rigorous testing/evaluation or tuning agents on end-to-end tasks.
Vertical Agents Offer Opportunity: This is a significant opportunity for founders with deep domain knowledge. Agents trained end-to-end for specific workflows (often using techniques like reinforcement learning on synthetic and user data) are showing early evidence of outperforming expert humans in specific tasks (security, DevOps, networking).
Predicting an Abundance Era: The coding market, having tipped, offers a preview of this era. With AI, labor becomes cheap and plentiful; taste becomes the scarce asset. This trend is a harbinger for how other industries will be changed.
The Next Major Wave: The Agent Economy: In the mid to long term, agent swarms will mature into an agent economy where agents transfer resources, make transactions, track each other, and understand trust/reliability. Humans and agents will work together in this new economy.
Building the Agent Economy: Technical Imperatives
Achieving the agent economy requires addressing several technical challenges.
Persistent Identity: Agents need a persistent personality and understanding of themselves. Equally important, they need a persistent understanding of the human user. True memory and self-learning remain major challenges.
Seamless Communication Protocols: Protocols are needed for agents to transfer information, value, and trust, similar to how TCP/IP enabled the internet. Initiatives like MCP are steps in this direction.
Elevated Security and Trust: When interaction isn't face-to-face, the importance of security and trust is dramatically increased. A significant industry will emerge specifically around enabling trust and security in agent interactions.
Adapting Mindsets for the AI Era
The AI era requires a shift in human mindsets.
The Stochastic Mindset: Departure from deterministic computing; requires comfort with and ability to manage uncertainty.
The Management Mindset: Shift from individual contributor thinking to managing complex systems and decisions, including understanding what agents can and cannot do.
More Leverage, Less Certainty: The new reality is the ability to achieve unprecedented leverage but requires managing increased uncertainty and risk effectively.
This transition is expected to reinvent individual work, rewire companies, and recreate the economy. Companies are already scaling faster with fewer people.
Go at Maximum Velocity
There is a tremendous "sucking sound" in the market for AI right now. The rising tide of technology adoption is powerful enough to swamp macroeconomic volatility.
Nature hates a vacuum. If you don't get in front of the opportunity, someone else will.
Notwithstanding focusing on moats and metrics, this is a "run like heck" business. Now is the time to operate at maximum velocity.