a16zAI in 2026: 3 Predictions For What’s To Come (a16z Big Ideas)
CHAPTERS
- 0:00 – 0:28
2026 Big Ideas overview: autonomous science, consumer connection, AI-reinforced business models
Erik Torenberg previews three 2026 predictions from a16z partners. The episode will cover autonomous labs for scientific discovery, consumer AI shifting from productivity to connectivity, and AI applications that create compounding business advantages.
- •Three themes for 2026: autonomous labs, consumer connectivity, and reinforced business models
- •Autonomous labs as a new paradigm for accelerating research
- •Consumer AI moving beyond work assistance into relationships and engagement
- •AI apps creating defensibility and widening gaps between leaders and laggards
- 0:28 – 1:29
Autonomous labs: pairing AI reasoning with robotic lab automation
Oliver Hsu explains what’s newly possible: combining advanced AI reasoning and experiment planning with physical lab automation. Near-term progress looks like tight collaboration between scientists, AI software, and robots across multiple scientific domains.
- •Lab automation isn’t new; the new piece is AI reasoning + experiment planning integrated with robotics
- •Human-in-the-loop collaboration is the near-term reality
- •Applies across life sciences, chemicals, and materials research
- •Goal is faster iteration and higher research throughput
- 1:29 – 2:00
Interpretability and traceability as requirements for AI-assisted research
Hsu argues that interpretability matters more in scientific workflows than in many other AI applications. Researchers must understand why an AI proposes a sequence of experiments and have reliable records of each step taken.
- •AI systems behave like non-deterministic computers—harder to reason about in labs
- •Science requires understanding “why” behind proposed experimental steps
- •Purpose-built scientific AI will emphasize logging, provenance, and step-by-step recording
- •Trust and adoption depend on transparent experimentation workflows
- 2:00 – 2:30
From assisted science to the destination: fully closed-loop self-driving labs
Hsu distinguishes today’s foundation-building from the end state of autonomous science. The long-term destination is a closed loop where AI designs, executes, and iterates experiments with minimal or no human intervention.
- •Fully autonomous “self-driving science” is further out
- •Today’s progress is incremental: better reasoning + better automation
- •Closed-loop autonomy requires robust planning, execution, and learning cycles
- •Autonomous science is framed as a long-term destination with near-term milestones
- 2:30 – 3:31
What must mature first: math/physics reasoning, simulation, world models, robot learning
Hsu outlines the capability stack needed to close the loop in scientific discovery. Progress is uneven across these areas, and autonomous science advances as these underlying technologies reach readiness.
- •Science spans theory, computation, and experimentation
- •Key enabling areas: mathematical reasoning, physical reasoning, simulation, world models, robot learning
- •Uneven capability maturity slows end-to-end autonomy
- •Breakthroughs will come as components become reliable enough to integrate
- 3:31 – 5:03
Where adoption happens first: demand-side markets for research outputs
Hsu predicts early adoption will be shaped by market pull rather than just technical feasibility. Industries with established buyers for research results will value speed, capability, and cost advantages most.
- •Early adoption depends on mature markets with clear ROI
- •Examples: life sciences/pharma, chemicals, parts of materials science
- •Faster and cheaper R&D matters when buyers already pay for outputs
- •Autonomous labs likely appear first where procurement and incentives already exist
- 5:03 – 6:21
Startup ecosystem and public-private partnerships accelerating AI-driven science
Hsu points to startups attempting different layers of autonomous science and to growing government/industry collaborations. These partnerships aim to speed scientific discovery by combining data, infrastructure, and advanced AI capabilities.
- •Examples of startups: Periodic Labs, Medra, ChemFi, Yoneda Labs
- •Some companies focus on automation; others aim at an “AI scientist”
- •Public-private initiatives like DOE’s Genesis Mission
- •Partnerships (e.g., DeepMind + UK government) signal broader institutional momentum
- 6:21 – 7:22
Consumer AI in 2026: shifting from productivity tools to connectivity products
Bryan Kim predicts consumer AI will evolve from helping users work to helping them connect. The focus moves toward self-understanding and relationship-building rather than purely efficiency gains.
- •2026 as an inflection point: productivity → connectivity
- •AI helping users ‘see themselves clearly’ and support relationships
- •Consumer mindshare shifts away from traditional apps toward AI-native interactions
- •Multiple use cases: digital companionship and facilitating real-world relationships
- 7:22 – 7:52
Startups vs incumbents: new interaction models create openings to win
Kim addresses competition with large platforms that already have networks and distribution. He argues AI introduces new interaction primitives and creative ‘atomic units’ that may not fit neatly into incumbent products, enabling startups to break through.
- •Incumbents have platform advantages and networks
- •AI can introduce novel interaction models that are hard to clone quickly
- •New creative formats and primitives may be native to new products, not old platforms
- •Belief: startups can win by owning new user behaviors
- 7:52 – 8:22
AI as a relationship facilitator: agents that mediate and initiate connection
Kim imagines users becoming comfortable sharing more of their inner life with AI. He suggests AI-to-AI communication could prompt check-ins, surface hard conversations, and create new openings for deeper connection.
- •Users are increasingly willing to share personal context with AI
- •AI could broker conversations: ‘my AI talks to your AI’
- •Products can prompt check-ins and initiate relationship-supporting actions
- •Goal: help people feel ‘seen’ by others through facilitated communication
- 8:22 – 9:31
Personalization via digital footprint: understanding the user without constant narration
Kim frames consumer product design around addressing core emotions—especially feeling seen. He argues the key technical/product challenge is fast, accurate personalization, potentially via photos, online/offline artifacts, and broader digital footprint ingestion.
- •Consumer mantra: address the core emotion (connection/being seen)
- •AI must understand the user deeply to facilitate relationships
- •Possible inputs: digital footprint, online/offline conversations, photo roll
- •Next wave: products that translate personalization into emotional resonance
- 9:31 – 10:01
AI that reinforces business models: revenue pull beats cost-cutting narratives
David Haber argues the best AI applications don’t just reduce costs; they strengthen how customers make money. When AI drives revenue or outcomes, adoption is less constrained and market pull becomes much stronger.
- •Contrast: automation/cost reduction vs revenue reinforcement
- •If AI improves earnings, customers adopt more aggressively
- •Stronger product-market pull when incentives align with revenue creation
- •Look for AI apps embedded in core workflows that affect outcomes
- 10:01 – 12:55
Case studies (plaintiff law and lending): outcome-driven AI creates compounding advantages
Haber illustrates the thesis with two examples: Eve in plaintiff law and Salient in loan servicing. He then explains defensibility through end-to-end workflow ownership and proprietary outcomes data that improves decision-making over time.
- •Eve (plaintiff law): contingency fees mean AI enables more cases and higher earnings, not fewer billable hours
- •Salient (loan servicing): voice agents reduce costs and improve collection rates (better outcomes)
- •Defensibility via deep workflow embedding from intake to outcome
- •Proprietary outcomes data compounds: smarter triage, better demand letters, improved win rates
