a16zThe AI Opportunity that goes beyond Models
CHAPTERS
AI as the next major platform shift: from PCs to cloud/mobile to AI apps
Alex frames AI as the next product cycle layered on top of prior infrastructure waves (cloud + mobile), which accelerates adoption and value creation. He argues most net-new software revenue is now coming from AI across both infrastructure and applications.
- •Historical product cycles: PC → internet → cloud → mobile → AI
- •AI’s acceleration depends on prior waves (smartphones + cloud distribution)
- •AI adoption is unusually fast compared to previous platform shifts
- •Value creation is happening at both infra and application layers
From novelty to habit: consumer usage inflection and enterprise pull
The discussion moves from early ChatGPT-era capabilities to today’s richer modalities and real utility. They cite rapid growth in usage and examples of AI becoming embedded in everyday workflows, including enterprise spend signals.
- •AI has moved beyond text/image into real-time, native audio interaction
- •Weekly ChatGPT usage is widespread; minutes per user are growing quickly
- •Anecdotes illustrate “countably infinite” daily use cases
- •Enterprise adoption is accelerating despite skepticism (e.g., claims deployments aren’t working)
Why AI apps are in a “golden age”: unprecedented company growth dynamics
Alex argues AI apps are enabling historically rare growth rates (e.g., near-zero to massive revenue quickly) because they deliver immediate, measurable economic value. He ties this to a simple behavioral premise: people want to be “richer and lazier.”
- •AI drives unusually fast go-to-market and revenue scaling
- •Customers buy because ROI is tangible (time and cost savings)
- •“Richer and lazier” as the core demand driver for GenAI
- •AI apps increasingly feel like essential utilities rather than experiments
Three investable AI application archetypes: AI-native software, software eating labor, and the walled garden
Alex lays out three core categories a16z is focusing on, emphasizing defensibility against labs and incumbents. The goal is to invest where durable moats can exist even as foundation models commoditize capabilities.
- •1) Traditional software categories going AI-native
- •2) New categories where software replaces/augments labor
- •3) “Walled garden”: proprietary data + AI delivering finished products
- •Key diligence question: what won’t OpenAI/Microsoft/labs do (or can’t easily do)?
Traditional software goes AI-native: greenfield vs. brownfield and systems of record
They explain why displacing incumbents is hard in “brownfield” (rip-and-replace) but more feasible in “greenfield” (new company creation or inflection points). The most defensible AI-native plays become systems of record with high switching costs.
- •Greenfield opportunities: new companies or inflection points (e.g., QuickBooks → ERP)
- •Brownfield displacement (e.g., Mailchimp/NetSuite replacement) is difficult
- •Incumbents will add AI and get stronger; startups must pick entry points carefully
- •Systems of record create stickiness; usage-based/outcome pricing pressures seat-based models
Software eating labor: the biggest market and the new pricing frontier
Alex argues the labor market dwarfs software, creating a massive opportunity for AI to do job-like work rather than sell tooling. The key challenge becomes pricing and defensibility: charging between “software spend” and “labor spend” while preventing easy substitution.
- •Labor spend is far larger than software spend
- •AI can do a large fraction of a job (24/7, multilingual, consistent)
- •Pricing is unclear: too high vs labor, too low vs traditional software budgets
- •To defend margins, products must become sticky end solutions (often systems of record)
Case study — Eve: plaintiff-side legal AI that owns the end-to-end workflow
David explains why plaintiff law is especially aligned with AI because contingency fees reward productivity rather than billable hours. Eve is positioned as a workflow owner from intake to litigation artifacts, and it compounds advantage through private outcome data.
- •Plaintiff attorneys take few cases; AI improves intake and case selection ROI
- •Voice agents gather evidence and triage cases by predicted value
- •Eve drafts chronologies, demand letters, complaints—covering the full workflow
- •Moat: 100% of cases flowing through the product + proprietary outcome data feedback loop
Defensibility vs. differentiation: why “AI features” aren’t enough
In Q&A, they distinguish flashy AI capabilities from durable competitive advantage. The key is workflow ownership plus proprietary data loops that improve outcomes and make the product mission-critical.
- •Differentiation (e.g., multilingual voice) can be copied; defensibility is harder
- •System-of-record positioning makes switching painful
- •Private data from real operations becomes a compounding advantage
- •Mission-critical adoption is validated by full workflow penetration and outcomes
Case study — Salient: AI for auto loan servicing that increases collections (not just cost-cutting)
Alex uses Salient to illustrate “software eating labor” in a high-friction call-center domain. The wedge is not merely reducing headcount, but increasing collections materially while ensuring compliance and improving reliability.
- •Problem space: servicing, collections, insurance follow-ups—high volume + high friction
- •Value prop: ~50% higher collections plus lower operating cost
- •Compliance moat: state-by-state statute awareness and consistent scripting
- •Data + execution moat: millions of calls create better scripts and outcomes over time
Vertical software can still be huge—especially when paired with labor automation and embedded finance
They discuss how niche/vertical software can scale into massive businesses (e.g., Toast) when it becomes a full operating system and adds monetization layers. The lesson: AI labor automation must be embedded in a broader, sticky platform to avoid price-based churn.
- •Vertical operating systems are hard to displace (examples: Toast, ServiceTitan)
- •Skepticism often underestimates TAM when platforms expand into financial services
- •AI labor features should be part of a broader workflow platform
- •Stickiness and multi-product bundling reduce vulnerability to undercutting
The walled garden strategy: proprietary data becomes more valuable when AI turns it into a finished product
Alex explains “walled gardens” as owning unique data access (even if raw data is publicly obtainable) and using AI to transform it into higher-value outputs. This shifts businesses from selling raw data subscriptions to delivering decisions, memos, and outcomes.
- •Examples of data moats: FlightAware (ADS-B), Bloomberg, LexisNexis, CoStar, DomainTools
- •AI increases willingness to pay by packaging raw data into finished deliverables
- •Proprietary often means hard-to-assemble history/time-series data, not necessarily secret data
- •Opportunity: digitize fragmented public records and monetize via AI-native workflows
Examples of walled gardens in practice: OpenEvidence, vLex, AskLéo
They highlight companies leveraging exclusive or aggregated datasets to outperform general-purpose chatbots. The common thread is defensible access to high-value corpora and the ability to sell directly to end users with substantially higher pricing.
- •OpenEvidence: exclusive licensing to medical journals; used broadly by clinicians
- •vLex: digitized legal corpora (e.g., Spain) + AI drove major revenue expansion
- •AskLéo: procurement intelligence via proprietary contract datasets
- •Strategic shift: don’t just sell data to intermediaries—sell the finished product to customers
Incumbents vs. startups: why AI can benefit both, and where disruption still happens
Alex contrasts AI with prior waves (cloud/mobile) where incumbents dismissed the shift. Here, everyone agrees AI is valuable, so incumbents will respond; startups must win via greenfield wedges, new categories, proprietary data, and workflow dominance.
- •Unlike cloud/mobile, incumbents aren’t dismissing AI; they will monetize aggressively
- •Startups should avoid pure brownfield “AI + replacement” battles without a wedge
- •“Why now?” often equals: AI enables economically viable finished products
- •Best startup opportunities: greenfield systems of record, labor automation, walled gardens
AI roll-ups: when buying distribution makes sense vs. traditional PE consolidation
Alex discusses AI-enabled services roll-ups, separating low-leverage geographic roll-ups (e.g., clinics) from digitally scalable roll-ups where acquiring a customer base accelerates adoption. The promising variant is buying a declining but well-distributed business to inject AI and expand rapidly.
- •Traditional roll-ups require many acquisitions; VC isn’t optimized for that playbook
- •More compelling: acquire one business with strong distribution/clients, then transform with AI
- •Example rationale: buy a debt collector/MSP to get customers, then improve outcomes dramatically
- •Key question: is the market local and fragmented or national and digitally scalable?
Consumer AI: AI-native categories, new category creation, and proprietary-data moats (plus model aggregation)
Anish maps Alex’s three archetypes to consumer, with examples in creative tools and therapy. He also argues that consumer winners may be model aggregators (like Kayak for airlines) because labs can’t easily provide multi-model “single pane of glass” experiences.
- •AI-native replacement: Krea as an AI-native ‘Photoshop’ for new designers
- •Category creation: ElevenLabs makes voice/audio a major new market
- •Proprietary data: Slingshot uses therapist scribe notes to train models powering ‘Ash’
- •Model aggregation thesis: users want multiple specialized models in one interface; labs default to first-party models only
Investment process, team, and Q&A: retention, enterprise sales motion, and execution support
They explain a16z’s approach to sourcing and winning competitive deals through expertise, publishing, and a conviction-driven process. In Q&A, they note retention has been strong when products are embedded in workflows, and enterprise adoption often requires forward-deployed engineering as much as sales.
- •“Find, pick, win” by being market experts (content + benchmarks) and moving fast on top deals
- •Conviction-oriented IC: ensure diligence quality; avoid political vote-trading
- •Retention: strongest where startups build full ecosystems, not just a single AI feature
- •Enterprise sales: heavy inbound in hot categories; increasing need for forward-deployed engineering and applied-AI guidance
