a16zThe AI Opportunity that goes beyond Models
CHAPTERS
- 0:01 – 2:17
Product cycles and why AI is the next platform shift
Alex frames AI as the newest major product cycle following PCs, the internet, cloud, and mobile—each with infrastructure and application winners. He argues AI is compounding on prior cycles (cloud + smartphones + global connectivity), enabling unusually fast adoption and new software revenue growth.
- •Historical platform shifts: PC, internet, cloud, mobile—each creates infra + app layers
- •AI builds on prior infrastructure rather than replacing it
- •Adoption is accelerating because distribution (smartphones/Wi‑Fi) is already global
- •Net-new software growth is increasingly AI-driven across infra and apps
- 2:17 – 6:42
From demos to real value: consumer ubiquity and enterprise inflection
The discussion moves from early “magic trick” demos to AI becoming embedded in daily routines and enterprise workflows. Alex cites strong usage growth (e.g., ChatGPT weekly users) and enterprise spend signals, arguing deployments are working despite skepticism.
- •Rapid capability expansion in just ~2 years (text → multimodal/real-time)
- •Consumer behavior shift: AI becoming a habitual utility (minutes per user rising)
- •Enterprise adoption inflecting—tools now saving time and money
- •Counterpoint to ‘AI deployments aren’t working’ narratives
- 6:42 – 9:13
The golden age of apps: why AI companies hit $100M revenue faster
Alex explains why AI application companies can scale revenue at unprecedented speeds versus traditional SaaS. The core driver is immediate ROI: customers pay because AI makes them both more productive and more profitable.
- •AI apps reaching extreme growth rates (0→$100M in 1–2 years)
- •Purchasing is value-driven, not hype-driven
- •“Richer and lazier” as a behavioral north star for AI adoption
- •Sets up three investment themes: AI-native software, software eating labor, and walled gardens
- 9:13 – 12:45
Theme 1 — Traditional software goes AI-native (and the greenfield wedge)
Alex describes how existing categories will be rebuilt as AI-native systems of record, similar to the cloud-native shift. He contrasts hard-to-win brownfield displacement with greenfield entry points where customers are forced to choose a new system.
- •AI-native versions of core categories (ERP, support, RPA, payroll, etc.)
- •Greenfield vs. brownfield: new businesses/inflection points are the openings
- •Systems of record create stickiness and pricing power
- •Incumbents will add AI—startups win by rebuilding the system for new buyers
- 12:45 – 14:16
Bingo board strategy, systems of record, and outcome-based pricing shifts
The “bingo board” illustrates how nearly every major software category will be upgraded with AI. Alex notes business model transitions (e.g., per-seat support becomes obsolete as automation rises) and reiterates that enduring winners become workflow hubs.
- •Incumbents will monetize AI add-ons due to switching costs
- •Systems of record are difficult to rip out—moats come from embedded workflows
- •Support and other tools shifting from per-seat to outcome-based pricing
- •Investment focus: build/own the new system of record, not a thin feature
- 14:16 – 17:02
Theme 2 — Software eating labor: turning jobs into products
Alex argues the labor market is far larger than software, creating a massive opportunity for AI to take on job responsibilities. The challenge is pricing and defensibility: products must become sticky and outcome-driven, not easily undercut widgets.
- •AI can perform a large share of many roles (24/7, multilingual, consistent)
- •New market: selling “labor as software,” not just replacing existing SaaS
- •Pricing sits between labor cost and SaaS norms—requires thoughtful packaging
- •Moats matter more as building software becomes easier (vibe coding)
- 17:02 – 21:10
Case study — Eve: end-to-end workflow + proprietary outcomes data as a moat
David presents Eve in plaintiff-side legal workflows, where contingency economics align incentives with productivity gains. Eve aims to own intake-to-outcome, using voice agents and document automation while compounding advantage through non-public outcomes data.
- •Plaintiff attorneys take few cases—AI improves screening and expands the market
- •Voice agent gathers evidence; system drafts key artifacts (chronologies, demand letters, complaints)
- •Defensibility comes from end-to-end workflow ownership (system of record)
- •Outcomes data isn’t public—creates compounding advantage over labs and general tools
- 21:10 – 24:29
Defensibility vs. differentiation: mission-critical AI apps
Responding to questions, the group distinguishes flashy AI features from durable defensibility. The key pattern for retention and “mission critical” status is workflow capture plus proprietary data loops—not simply access to a model capability.
- •Differentiation (e.g., multilingual voice) is not defensibility by itself
- •Mission critical comes from being the operating workflow and context layer
- •Proprietary feedback loops (outcomes, usage data) compound over time
- •System-of-record posture reduces switchability even with consumption pricing
- 24:29 – 31:50
Case study — Salient: AI in collections that drives revenue, not just savings
Alex explains Salient’s growth by focusing on measurable value creation—collecting more money and reducing compliance risk—rather than merely cutting call-center costs. The company’s moat is operational: scripts, statutes, and learning from millions of calls.
- •Auto loan servicing/collections: AI handles unpleasant, high-churn work
- •Value prop: ~50% higher collections and better compliance, not only lower cost
- •Moat: large-scale call data + rapid ingestion of evolving state/federal statutes
- •Stickiness increases when paired with a system-of-record workflow
- 31:50 – 35:22
Theme 3 — The walled garden: proprietary data + AI enables ‘finished products’
Alex introduces the walled-garden strategy: assemble unique or time-accumulated datasets and use AI to deliver complete outcomes, not raw data. This shifts pricing power from low-cost subscriptions to high-value decisions and workflows.
- •Examples of proprietary data moats: FlightAware, PitchBook, LexisNexis, CoStar, Bloomberg
- •Many data sources are ‘free’ but costly to aggregate, digitize, and maintain over time
- •AI increases value by turning data into decision-ready products (not just access)
- •The ‘farm adding restaurants’ analogy: platforms may compete—walled gardens counterbalance
- 35:22 – 40:22
Walled garden examples: OpenEvidence, vLex (Ask Leo), and how incumbents respond
Alex details how exclusive or hard-to-recreate content (medical journals, legal corpora, contract libraries) creates defensibility against general-purpose models. He also explains the ‘why now’—AI enables finished-product delivery that makes old data businesses dramatically more valuable.
- •OpenEvidence: exclusive medical journal access + ChatGPT-like UI for clinicians
- •vLex: scaled legal corpus; AI layer drove major revenue expansion
- •Ask Leo: contract/procurement intelligence built on proprietary contract data
- •‘Why now’: AI converts low-value raw material into high-value outcomes
- 40:22 – 49:29
Incumbents vs. startups: where disruption works (and where it doesn’t)
Jen pushes on the competitive dynamic, and Alex argues AI is unusually favorable to both incumbents and startups. Incumbents can monetize hostages and embed AI quickly, while startups win in greenfield moments, new labor markets, and new data moats.
- •AI differs from cloud/mobile: incumbents largely agree it’s valuable and will adopt it
- •Startups should avoid pure brownfield displacement without a forcing function
- •Greenfield triggers: new company formation, inflection points, new categories
- •New proprietary datasets can create businesses that weren’t viable pre-AI
- 49:29 – 53:43
AI roll-ups: when buying distribution/customers makes sense (and when it doesn’t)
Alex outlines when AI-enabled roll-ups are attractive: buying a platform with customers to replace slow enterprise sales, especially in national/digital service markets. He’s skeptical of locality-bound roll-ups (clinics, many professional services) that require repeated acquisitions to grow.
- •PE-style roll-ups vs. AI-native transformation: different playbooks
- •Good fit: buy one business with blue-chip customers, then scale via AI efficiency
- •Bad fit: local, fragmented businesses requiring hundreds of acquisitions (e.g., clinics)
- •Example logic: debt collection/MSPs can be national and digitally delivered
- 53:43 – 55:44
Consumer AI playbook: AI-native categories, new categories, and proprietary therapy data
Anish maps Alex’s three themes directly onto consumer AI. He highlights AI-native creative tools, category creation in voice, and proprietary-data strategies like Slingshot’s therapist-scribe pipeline feeding a consumer therapy product.
- •AI-native replacement: Krea as an ‘AI Photoshop’ for new designers
- •Category creation: ElevenLabs expanded voice/audio into a major market
- •Proprietary data: Slingshot collects therapist session notes → trains model → consumer app (Ash)
- •Consumer defensibility mirrors enterprise: workflow + data + product depth
- 55:44 – 57:05
Model aggregation as a strategy: why ‘Kayak for models’ can beat single-lab apps
Anish argues that in many consumer categories, the best product aggregates multiple models because models have distinct strengths. Labs and big tech are constrained to first-party models, giving aggregators an advantage as the “single pane of glass.”
- •Analogy: Kayak aggregates airline inventory; users prefer cross-provider choice
- •Models are not perfect substitutes—specialization matters
- •Aggregators can route tasks to the best model for the job
- •Labs/big tech limited by first-party incentives and product focus
- 57:05 – 1:06:42
How a16z evaluates and wins deals: content-led sourcing, conviction, and process
Alex explains the fund’s operating model: publish category research to become the obvious partner, then move fast on high-quality opportunities. Investment decisions emphasize conviction with process safeguards, avoiding committee politics while ensuring thorough competitive diligence.
- •‘Process + interrupt’: deep research plus rapid response to exceptional deals
- •Content as a competitive edge for sourcing and credibility with founders
- •Conviction-led IC with a “two-key” process (expert lead + process validation)
- •Team structure balances deal-finding expertise with operator/board leverage
- 1:06:42 – 1:09:44
Q&A — retention signals and the evolving enterprise go-to-market motion
In closing Q&A, Anish and David address retention and enterprise sales dynamics. They report limited switching so far when startups build full ecosystems around primitives, and note strong inbound demand plus rising importance of forward-deployed engineering for large customers.
- •Retention strongest when products wrap rich workflows around AI primitives
- •Customers look to startups as ongoing AI solutions partners amid fast-changing capabilities
- •More inbound than typical enterprise software—some companies avoid outbound early
- •Forward-deployed engineers increasingly critical to drive adoption in large orgs
