Sequoia partners Packer, Sonia, and Constantine led the opening session at AI Ascent, providing reflections on the current state and future of AI, with a focus on market opportunities, technology layers, and practical business considerations.
Key topics included the growing role of AI applications, the emergence of agent economies, technical challenges such as persistent identity and security, and the necessity for startups to deliver genuine value and trust.
The presentations highlighted recent progress in AI engagement, the rapid evolution of agent technology, and expected industry shifts toward greater leverage and uncertainty in the workplace.
The session set the stage for further talks from industry leaders and innovators featured throughout the day's event.
Action Items
(none explicitly mentioned in transcript)
State of AI Market and Opportunities
Sequoia's simple framework for evaluating markets: What is it? So what? Why does it matter? Why now? What now?
AI impacts both the services and software markets, attacking traditional profit pools with both co-pilot (assistive) and autopilot (automated) evolution.
The scale and opportunity in AI dwarf prior technology transitions, supported by much wider and faster adoption rails (e.g., internet reach, social platform virality).
Current value creation and competition is concentrated in the application layer, with foundation models increasingly encroaching into applications.
Startups are advised to focus on customer needs ("customer back" approach), vertical or function-specific complexity, and leveraging unique data from usage to build moats.
Building Successful AI Companies
95% of building an AI company is classic company-building; only the last 5% is AI-specific.
Key AI-specific metrics:
Revenue: Distinguish “vibe”/non-durable revenue from true product-market fit through engagement, adoption, and retention metrics.
Margins: Focus on a clear path toward robust gross margins, even if not strong initially, as COGS should decrease and value capture should rise.
Data Flywheel: Proprietary usage data feeding business metric improvements is a strong moat, but must be tied to real outcomes.
Trust between company and customer is critical, especially as product quality will continue to improve.
Recent Advances and Engagement in AI
AI native applications have significantly improved engagement ratios, moving beyond initial hype to demonstrable daily use (e.g., ChatGPT matching Reddit engagement levels).
Voice-based AI applications have crossed the "uncanny valley," making naturalistic voice interaction broadly compelling.
Coding has emerged as a breakout application for AI, democratizing software creation and fundamentally altering economics and accessibility.
Technology breakthroughs include advanced reasoning, synthetic data, tool use, and agentic scaffolding, with research and product boundaries increasingly blurred (e.g., Deep Research, Notebook LM).
Predictions for the Mid-to-Long Term AI Landscape
The next major wave is agent economies: networks of persistent, resource-exchanging, trust-aware AI agents working collaboratively and with humans.
Three main technical challenges ahead:
Persistent identity for agents and users to maintain continuity and trust.
Seamless communication protocols to facilitate secure, scalable agent interaction (e.g., early development of standards like MCP).
Security and trust as foundational, with an emerging industry focused on these areas.
The rise of a "stochastic mindset": moving from deterministic to probabilistic expectations in computing, requiring new management and risk strategies.
Increasing leverage with less certainty, with companies scaling faster and leaner, possibly reaching new standards of productivity and organization.
Decisions
(No formal decisions recorded in this transcript.)
Open Questions / Follow-Ups
How will companies best address the technical challenges of persistent agent identity and true memory?
What standards or protocols will ultimately emerge as the backbone for agent communication and security?
How will management frameworks adapt to a stochastic and highly leveraged workforce model, especially with AI agents integrated into core processes?