a16zAI in 2026: 3 Predictions For What’s To Come (a16z Big Ideas)
CHAPTERS
Big Ideas for 2026: Autonomous science, connected consumer AI, and AI-reinforced business models
The episode tees up three 2026 predictions from a16z partners. It frames AI’s next phase as (1) accelerating discovery via autonomous labs, (2) shifting consumer AI from productivity to human connection, and (3) creating durable winners where AI strengthens revenue-driving business models.
- •Three themes: autonomous labs, consumer connectivity, compounding advantages in AI apps
- •AI impact expands beyond efficiency into new workflows and relationships
- •Focus on where adoption pull is strongest: science markets, consumer behavior, and revenue reinforcement
From lab automation to “autonomous labs”: combining AI reasoning with robotics
Oliver Hsu explains what’s new about today’s “autonomous labs” versus traditional lab automation. The breakthrough is pairing AI reasoning/experiment planning with robotic execution to create a tighter human–AI–robot collaboration loop in real lab environments.
- •Lab automation isn’t new; the novelty is AI reasoning + experiment planning + robot learning
- •Near-term model: collaborative scientist + AI application + lab robot
- •Applicability across life sciences, chemicals, and materials science
- •Goal: faster iteration cycles and accelerated breakthroughs
Interpretability and audit trails as prerequisites for AI-driven research
Hsu argues that scientific settings demand more transparency than many general AI applications. Because AI systems can be non-deterministic, researchers need clear records of why the system chose specific experiment steps and how results informed subsequent iterations.
- •Interpretability is critical in scientific discovery workflows
- •Need to capture/record each step of planning and execution (traceability)
- •Purpose-built scientific AI systems will likely emphasize explainability
- •Trust and reproducibility are key adoption gates in labs
The destination: closed-loop, self-driving science (and why it’s still farther out)
The long-term vision is fully autonomous, closed-loop science: AI generates hypotheses, runs experiments, ingests results, and iterates without human intervention. Hsu positions this as a destination that depends on uneven progress across multiple capability areas.
- •“Self-driving science” = closed-loop iteration with minimal/no human intervention
- •Science loop spans theory, computation, and experimentation
- •Enablers: mathematical reasoning, physical reasoning, simulation/world models, robot learning
- •Progress is uneven; near-term advances will be incremental toward the loop
Where adoption happens first: market maturity and clear buyers of research outputs
Hsu predicts early adoption will correlate with domains that have established demand for research outputs and can pay for speed/cost advantages. Mature markets like pharma, chemicals, and parts of materials science have clearer ROI pathways for autonomous lab systems.
- •Adoption driven by demand-side market readiness, not just technical feasibility
- •Life sciences/pharma as prime early adopter category
- •Chemicals and materials science also have mature buyers for research output
- •Speed, capability gains, and potential cost advantages drive willingness to pay
Startup landscape and examples: building blocks of autonomous science
Hsu highlights early companies pursuing pieces of the autonomous science stack, from life sciences to chemistry-focused efforts. These startups represent different wedges into autonomous discovery: automation, AI “scientist” software, and verticalized lab workflows.
- •Periodic Labs cited as an example “taking a swing” at autonomous science
- •Medra (life sciences/pharma) highlighted as an early-stage example
- •ChemFi and Yoneda Labs noted in the chemistry industry context
- •Startups span lab automation and AI scientist approaches
Public–private partnerships accelerating AI-driven scientific discovery
Beyond startups, Hsu points to government, academia, and industry collaborations as key accelerants. These partnerships aggregate resources, data, and infrastructure—potentially speeding translation of AI advances into real scientific progress.
- •DOE’s Genesis Mission brings together academia, government, national labs, and AI companies
- •DeepMind–UK government partnership mentioned as another signal
- •Ecosystem approach: public sector + private sector + academia
- •Collaboration helps scale compute, labs, and applied research programs
Consumer AI in 2026: the shift from productivity tools to connectivity tools
Bryan Kim predicts consumer AI will move from helping users “do work” to helping them “stay connected.” The emphasis becomes emotional utility—helping people feel understood, maintain relationships, and spend time in new AI-native interaction models.
- •2026 inflection: consumer AI products shift from productivity to connectivity
- •AI becomes a mechanism to reflect self-understanding and support relationships
- •AI may take mindshare/time from traditional consumer products
- •Two modes: augment digital connection and facilitate in-person relationships
Startups vs incumbents: winning via new interaction models and creative primitives
Kim argues that incumbents’ network effects aren’t insurmountable if AI enables fundamentally new interaction patterns. If the “atomic units” of sharing and creativity change, startups can win by building AI-native behaviors that don’t fit existing platforms well.
- •Incumbents have platforms and networks, but AI introduces new interaction models
- •Hard-to-replicate behaviors can emerge outside incumbent product constraints
- •New “creative outlets” and “atomic units” of content/interaction create openings
- •Startups can win by owning AI-native engagement loops
AI as a relationship facilitator: “my AI talks to your AI”
Kim describes a future where people share more of their inner life with AI and allow agents to coordinate socially. AI-to-AI communication could prompt check-ins, open difficult conversations, and create relationship moments that wouldn’t happen otherwise.
- •People are increasingly comfortable sharing deeply with AI
- •Concept: personal AIs coordinating to prompt supportive actions and conversations
- •AI can help people feel “seen” by others
- •Consumer product design focus: address the core emotion (connection)
Personalization through digital footprint ingestion (and the tradeoff)
To facilitate connection, Kim suggests AI must understand users quickly without requiring exhaustive manual setup. He points to ingesting digital footprints—messages, online activity, photos—as a path to rapid personalization, implying new expectations around permissioning and trust.
- •Personalization requires fast, accurate user understanding
- •Potential inputs: digital footprint, online/offline conversations, photo roll
- •Goal: avoid users needing to narrate their life story
- •Trust, consent, and data handling become central to product adoption
AI that reinforces business models: adoption is strongest when revenue grows, not just costs fall
David Haber argues the most powerful AI application companies will be those that strengthen customers’ business models—especially by increasing revenue or outcomes—rather than only automating labor. When AI directly improves earnings, customer demand can be uncapped.
- •Cost reduction alone creates a weaker adoption narrative than revenue reinforcement
- •When AI drives better outcomes, customers want more of it
- •“Market pull” is strongest when ROI maps to revenue, not just efficiency
- •Framework for identifying durable AI app winners
Case studies: plaintiff law (Eve) and loan servicing voice agents (Salient)
Haber details two portfolio examples where AI improves customer outcomes. In plaintiff law, AI helps contingency-based firms take more cases and win more; in lending, compliant multilingual voice agents reduce costs and improve collection rates.
- •Eve: plaintiff attorneys on contingency; AI enables more throughput and higher earnings
- •AI supports drafting/reasoning but aligns with revenue (not billable hours)
- •Salient: voice agents for lending workflows (welcome calls, reminders) across many languages
- •Outcome improvement: better collection rates, not only call-center efficiency
Compounding advantage and defensibility: end-to-end workflow + proprietary outcomes data
Haber explains how AI apps build moats by embedding deeply in daily workflows and collecting proprietary data that improves decisions over time. For Eve, outcomes data from intake-to-resolution is private and enables smarter triage, valuation, and strategy—making the platform stronger with every case.
- •Defensibility from owning the end-to-end workflow (intake → outcome)
- •Deep embedding drives switching costs and sustained usage
- •Proprietary outcomes data isn’t available on the public internet for model training
- •Feedback loop: more cases processed → smarter intake/triage/demand letters → better outcomes
