a16zThe Current State of Consumer AI
a16z
Consumer AI spending and the new report 0:00
About half of Americans say they use AI, but only around 4 percent of US consumers actually pay for an AI subscription, and the heaviest spenders put roughly 93 dollars a month on their personal cards. This conversation centers on the seventh edition of a top 100 consumer AI apps report, which for the first time adds real consumer card spend data instead of just website traffic. The report finds only 11 new products across web and mobile combined, the fewest of any edition so far, even though 29 of the 50 products ranked by spend never showed up on the traffic lists at all, showing how much of consumer AI right now is a power user game.
Personal agents become the new frontier 3:31
The biggest shift since the last report is the rise of personal agents, the kind you might actually hand to a family member. OpenClaw, once projected to rank very high in traffic, has now vanished from the rankings after its team was absorbed into OpenAI, while rivals like Grok, Comet, and a new product called Dots have taken its place. Two standouts are Muse and Instinct. Instinct reportedly reached 100,000 users growing 10 percent a day, with 40 percent of users linking a credit card within three weeks and spending over 1,000 dollars in their first month. Muse drew about 500,000 downloads and 250,000 active users in its first 12 days, but compared with Threads, which hit 16 million US and Canada downloads in 22 days versus Muse's 5 million over the same stretch, it still has far less mainstream pull.
Trust, cost, and who uses agents 9:30
Nobody has yet cracked person to person network effects for these assistants, partly because they know users so intimately that people hesitate to blend that access with other people's data, the same reason many keep separate work and personal chatbot accounts. Trust and privacy are described as the real limit on growth, since an agent with access to your email or credit card could share things or act in ways you don't expect. On cost, David Paffenholz, who runs a site tracking over 170 agents and a community of more than 1,500 early adopters, found their top use case is still coding and technical automation, which explains why some assistants cost hundreds or thousands of dollars a month to run, while mainstream-focused products only cost in the tens of dollars.
Spending on AI is Tiny and Concentrated 14:01
About half of Americans report using AI, yet only around four and a half percent of US consumers actually pay a subscription for an AI product, according to Ramp data, though that share has roughly doubled over the past year. Even within that small paying group, spending is heavily skewed: the top ten percent of payers generate more than half of all revenue, the top one percent alone account for twenty percent, while the bottom half of payers contribute only about sixteen percent. The single top-spending user pays ninety three dollars a month personally, while the median paying user spends twenty five dollars, and this money flows mostly toward developer, productivity, and creative tools like n8n, Granola, Higgsfield, and Manus.
Makers, Not Just Time Savers 19:02
The speakers see this small group of high spenders as a sign of a new generation of makers, people using personal credit cards to code, make art, or produce videos they never could before. One portfolio founder, Eugenia of Wabi, is quoted as saying most people are not looking to save time, they are looking for ways to spend their time, which is why entertainment products like Netflix, TikTok, and YouTube dominate daily use. The hope expressed is that AI moves beyond making tasks faster toward giving ordinary people the tools to actually create and bring ideas to life, much like the early frenzy around coding agents that kept people glued to their laptops.
Subscriptions, Costs, and the Rise of Ads 20:30
Roughly eighty five percent of consumer AI products monetize through subscriptions and another sixty two percent through credits or usage payments, with only about thirteen percent relying on ads, an inversion of how most big consumer internet companies historically made money. High inference costs are blamed for this early reliance on direct payment, and the hope is that as costs fall, advertising and transaction fees will return as viable models. OpenAI is cited as already reaching a billion dollar annual run rate from ads, helped by 1.2 billion weekly active users and deep personal knowledge of each user, with features like login with ChatGPT suggesting more targeted advertising ahead. Early ad experiences are described as subtle and well labeled, fitting naturally into commercial queries like trip planning or product shopping rather than interrupting personal conversations, with one example involving a search for an American-made keychain that led to a previously unknown factory in Detroit.
Niche AI Products and Advertising 28:33
OpenEvidence, an AI product built for doctors, has already reached 50 to 60 percent density among US physicians, showing that a narrow but valuable audience can support strong advertising. The expectation is that more of these verticalized consumer AI products, built for specific professional audiences, will follow this same path of early ad success.
Chatbot Rankings and Spending Patterns 29:30
ChatGPT remains the dominant consumer AI product by both usage and revenue, staying about six times ahead of Claude and twice ahead of Gemini on the web. The bigger surprise is that Claude has passed Gemini in number of paid subscribers, despite Gemini's much larger install base and Google's natural distribution advantages, a shift credited to strong launches like Claude Code and heavy press attention since around February. ChatGPT still leads in total paid subscribers, roughly three times more than Gemini or Claude, but Anthropic has leaned hard into subscriptions instead of ads, with about seven and a half percent of its subscribers on the hundred-dollar-plus Max plan compared to only about one percent for ChatGPT and Gemini.
Creative Tools Split by Modality 32:30
Audio tools like ElevenLabs and Suno have stayed near the top of traffic and spending rankings because the big labs have mostly avoided building competing voice and music models, partly to dodge the IP headaches Suno has faced. In images, OpenAI's Images 2.0 and Google's Nano Banana and VO have pulled casual traffic away from standalone generators like Midjourney, though power users still pay for Midjourney's distinct aesthetic sense. Video remains a harder frontier, where Chinese companies with fewer data restrictions have an edge given how complex it is to generate synced visuals and sound, and tools that work well with AI agents are expected to gain strength.
Consumer Tools Turning Enterprise 37:02
Products like Granola, Whisper Flow, Superhuman, Replit, and Gamma are following a pattern where consumer tools grow fast through product-led growth and then get pulled into enterprise use because they are work-adjacent, eventually adding privacy, security, and team plans. Big incumbents like Google have been slow to reinvent core tools like Docs, Gmail, or Calendar, and even OpenAI and Anthropic struggle to innovate much outside their own chat interfaces, leaving room for startups such as Plaud, an AI notetaker device, to find real openings.
Software Layer Becomes the Value 41:30
As models keep improving, the opportunity for startups is to build products that get better alongside the models without being replaced by them, since big labs move slower due to their size and internal complexity. The belief is that value is shifting back to the software and experience layer, where community, context, and interface design become the real asset, giving startups a chance to reinvent categories faster than larger companies can, even if some big players do manage to adapt.
Building Context That Can't Be Copied 43:30
One speaker describes using a product called Town to write emails and help with reports, noting that it has built detailed playbooks of who he is and how he sounds in a way no other product has matched. He explains that the gap between an email that sounds 99.9 percent like him versus 85 percent like him is the difference between a 10 second fix and a 10 plus minute one. Both speakers agree that dismissing such tools as a mere harness or wrapper misses the point, since real products are being built that collect context and act on it, with the underlying model becoming just one piece of a much richer experience.
Where The White Space Remains 45:31
Olivia points out that most consumers still use AI mainly as a search replacement, for things like homework help or writing emails, and that current startup success has clustered around getting things done, including productivity, photo and video editing, and search. She identifies wide open categories, especially network categories that need a multiplayer product, such as dating and recruiting, where no one has yet made the list. Social AI has also not taken off, remaining mostly people posting AI generated content on existing platforms.
Shopping, Entertainment, And Storytelling 47:31
The discussion turns to shopping, home buying, and retail, with speculation about whether these will live inside agents like Instinct or Muse, as standalone products, or some mix of both. Gaming and entertainment are flagged as areas where models still aren't good enough to produce content people want to watch, aside from the exception of AI micro dramas. One speaker compares this moment to TikTok's rise, suggesting AI could help people become better storytellers and create new networks of shared content once models improve further.
Saving Time Versus Spending Time 49:00
The conversation closes by returning to the distinction between products that save people time and those that let people spend time meaningfully, noting that almost everything built in consumer AI so far falls into the save time category even though spend time products often become the biggest companies. Both speakers agree many builders are already working on this, and the episode ends with a reminder to check out the seventh edition of the top 100 consumer AI apps list, with a promise to return for edition eight.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
