CHAPTERS
- 0:04 – 1:58
Top 100 AI Apps report: what’s changed since 2023 (and why we’re still early)
Anish and Olivia set the context for the sixth edition of a16z’s Top 100 AI Apps report, highlighting both rapid product growth and surprisingly low global penetration. Olivia explains why the last six months have been especially dynamic for consumer AI, with new form factors and platform battles intensifying.
- •Six editions over three years: massive growth, but still early-stage adoption
- •ChatGPT remains the dominant global AI product; weekly active usage is still a small share of the world
- •Competition is heating up for consumer/prosumer mindshare across major model platforms
- •Non-AI-native products (e.g., Canva, Notion) now qualify as “majority AI-enabled”
- •AI is expanding beyond the prompt box into browsers and desktop apps
- 1:58 – 4:18
How ChatGPT, Gemini, and Claude are differentiating (and the real market-share gaps)
Olivia breaks down usage data showing ChatGPT’s clear lead on both web and mobile, despite online discourse suggesting tighter competition. She then explains how each platform is carving out distinct user segments and ecosystems, with surprisingly low overlap between their emerging “app stores.”
- •ChatGPT leads decisively: ~2.5–2.7x Gemini; ~30x Claude on web and ~80x on mobile
- •User behavior is expanding: people increasingly use multiple AI tools for different jobs
- •Claude is leaning into prosumer/knowledge-work integrations (e.g., Excel/PowerPoint)
- •ChatGPT’s and Claude’s app ecosystems have 200+ apps each, but only ~11% overlap
- •Gemini’s traction is closely tied to creative-model releases and Google product embeds
- 4:18 – 5:56
ChatGPT’s “AI for everyone” strategy vs. subscription-first monetization
The discussion turns to platform strategy: ChatGPT aims to win broad consumer distribution first, then monetize via multiple paths. Olivia contrasts this with Claude’s clearer subscription-led approach and explains why ChatGPT’s directory/app-store strategy could resemble Google’s long-term playbook.
- •ChatGPT’s goal: capture the broadest possible consumer base (“AI for everyone”)
- •Claude’s monetization is more straightforward: subscriptions and high-ACV work tools
- •ChatGPT may monetize beyond subscriptions via ads and/or transaction take rates
- •Consumer-facing plugins/marketplaces (travel, nutrition, finance) expand monetization surface
- •The bull case may not fully show up in current data but compounds over time
- 5:56 – 9:12
Compounding advantages: context, network effects, and “log in with ChatGPT”
Olivia explains why “context” and “memory” could become long-term moats for horizontal AI platforms, especially as products become stickier and more social. The conversation explores network effects, developer incentives, and a potential identity/inference layer where users bring memory and compute to third-party apps.
- •Lock-in may rise as memory/context becomes harder to export between products
- •Social usage (e.g., group chats) creates classic network effects and switching friction
- •Developers may prioritize the biggest platform for early launches and deeper integrations
- •A hinted “log in with ChatGPT” layer could let users port memory/tokens to other apps
- •Open question: separating work vs. personal identity/memory to avoid “crossing streams”
- 9:12 – 11:33
Google’s Gemini comeback: DeepMind-driven creativity and the challenge of legacy surfaces
Anish and Olivia assess Google’s shift from the Bard era to a more confident, multimodal creative push. They argue Google’s strongest innovation is happening in greenfield products (like NotebookLM) rather than entrenched enterprise surfaces, where inertia and organizational complexity slow change.
- •Google has improved execution and product “vibes,” leaning into multimodality and creativity
- •DeepMind-led model-first products drive much of Gemini’s momentum
- •NotebookLM is cited as a standout example of a truly new consumer experience
- •Legacy products (Docs/Sheets) face high inertia, risk of user disruption, and internal overhead
- •Google may still “hold the line” due to enterprise lock-in and distribution strength
- 11:33 – 14:23
Global AI ecosystems: why China and Russia look structurally different
Olivia shares new geographic analysis from the report, showing how regulation, censorship, and sanctions shape distinct AI stacks in China and Russia. Both countries emerge as major outliers with parallel product ecosystems, creating large markets where local or alternative models dominate.
- •China has the lowest combined ChatGPT+Gemini usage (~15%) due to restrictions
- •Chinese users rely more on domestic tools/models (e.g., Doubao, DeepSeek, Kimi)
- •Russia similarly shows a parallel ecosystem (e.g., GigaChat, Yandex) driven by constraints
- •Russia is DeepSeek’s #2 market after China
- •These are large markets that can surface local winners into global rankings
- 14:23 – 17:55
Per-capita adoption heat map: which countries embrace AI fastest (and why)
The conversation expands from ecosystem structure to per-capita usage patterns across major LLM products. Olivia explains why places like Singapore and the UAE rank highly, while the U.S. sits lower than expected—driven by workforce mix and differences in cultural trust and optimism toward AI.
- •Per-capita leaders include Singapore (#1), Hong Kong, UAE, and South Korea
- •The U.S. ranks around #20 despite producing many leading AI products
- •Workforce composition matters: tech-first, white-collar economies adopt faster
- •Trust and cultural sentiment vary widely; U.S. trust in AI cited around 32% in a survey
- •Adoption likely diverges further as creative/cultural use cases become more prominent
- 17:55 – 20:52
Creative tools evolution: from image-generator dominance to workflow and modality battles
Olivia traces how creative apps shaped early consumer AI—starting with Midjourney—and how that mix is changing as foundation models commoditize basic image generation. She argues enduring winners now either offer distinctive aesthetics/workflows or operate in modalities where platforms haven’t fully consolidated.
- •Early consumer AI was dominated by creative tools; hallucinations were an asset for art
- •Standalone image generators face pressure as ChatGPT/Gemini handle commodity images well
- •Surviving image players tend to be aesthetically opinionated or workflow-rich (e.g., Ideogram, Midjourney)
- •Music/voice/video remain more open, enabling breakouts like Suno and ElevenLabs
- •Video likely won’t be “one model to rule them all,” favoring multi-model platforms (e.g., Krea)
- 20:52 – 24:16
Sora’s launch as a social experiment: explosive adoption, exportable content, unclear native feed
Anish and Olivia unpack Sora’s breakout launch and what it revealed about AI-native social. While Sora proved strong as a creation tool (rapid user growth and meaningful DAUs), they argue its social feed struggled because the best Sora content competes more effectively when exported to mainstream platforms.
- •Sora hit #1 in the U.S. App Store for 20 consecutive days and reached 1M users fast
- •Sora still has meaningful daily usage (millions of DAUs), though downloads cooled
- •“Cameos” enabled likeness-based meme creation and early virality (including celebrity participation)
- •Exportability pushed the best content to TikTok/Reels/YouTube, weakening a Sora-only feed
- •AI-only social still lacks a durable “status game,” though licensed content could create niches
- 24:16 – 26:32
Agents surge: OpenClaw’s developer breakout and what it signals
The episode pivots to agents, focusing on OpenClaw’s rapid rise in developer mindshare after the report’s data window. Olivia explains why OpenClaw is massively influential in technical communities, what its growth limits suggest about mainstream adoption, and why “OpenClaw for X” is becoming a common startup pattern.
- •OpenClaw would have debuted around #30 on the web list if February data were included
- •It became the #1 GitHub repo by stars all time, surpassing major projects
- •User growth appears strong among technical users but flatter for new mainstream sign-ups
- •OpenClaw’s architecture is inspiring many “vertical OpenClaw” startups
- •Tension to watch: multi-model openness vs. alignment with a single lab/provider
- 26:32 – 29:01
Manus as a consumer-grade agent: reliability, distribution, and the “horizontal app” risk
Olivia contrasts Manus with OpenClaw, describing Manus as an early breakthrough in consumer-accessible, autonomous agent behavior across tools. She also highlights a strategic reality: as agent capabilities become table stakes at the model layer, broad horizontal agent apps may need massive distribution to win long term.
- •Manus stands out for consumer-grade autonomy (email/web/slides/spreadsheets) and reliability
- •Earlier agent attempts (Operator/Mariner-era) felt less dependable to users
- •Horizontal agent products may be vulnerable once big platforms replicate core capabilities
- •Distribution and existing enterprise relationships can decide winners for broad agents
- •Vertical agents may remain better startup territory than all-purpose consumer agents
- 29:01 – 38:32
Beyond the prompt box: desktop AI, AI browsers, teen usage patterns, and memory as the next moat
In the closing stretch, Olivia explains why desktop apps and AI-native browsers are emerging as major new surfaces—even if measurement lags behind web/mobile. She then shares data on how teenagers use AI today, argues agents will become invisible infrastructure across software, and closes on memory as a core product advantage that will make future onboarding feel obsolete.
- •Desktop AI is rising (e.g., meeting/voice tools), but desktop usage is harder to track than web/mobile
- •AI browsers (Comet/Atlas) show promise, but browser switching costs demand killer features
- •Teen usage data: homework is dominant; creative use is rising; casual conversation and emotional support are emerging
- •Agents will become ubiquitous—users won’t think of them as “agents,” just capabilities
- •Memory/personalization will be decisive; products that don’t “know you” will feel broken, but persona boundaries (work vs personal) must be handled well
