a16z just published the 7th edition of its Top 100 Gen AI Consumer Apps and discussed it on its podcast. The headline trend: personal agents that act for you. Here is what the data says about agents that shop and pay, every number with its source.
1. Paying for AI is still niche, and extremely concentrated
- 4.5% of US consumers had an active paid AI subscription in August 2026.
- Inside that group, the top 1% of payers account for 19.5% of observed consumer AI spend; the bottom 50% account for 16.6%.
- Median payer: $25/month.
Source: a16z, Top 100 Consumer AI Apps – 7th edition, data from YipitData's US consumer-card panel.
2. The top is pulling away
| Segment | Jan 2025 | Aug 2026 | Change |
|---|---|---|---|
| Top 1% | $504/mo | $903/mo | +79% |
| Top 10% | $119/mo | $156/mo | +31% |
| Top 40% | $27/mo | $30/mo | +10% |
Source: a16z chart shown in the episode, YipitData panel; $903 also in the report.
These are people who build, make and sell: developer, productivity and creative tools dominate their spend. Spend and traffic are different maps of the market — 29 of the top 50 products by spend don't appear on a16z's traffic lists, and only 11 new apps made the lists this time, the fewest debuts ever.
3. Consumer AI barely monetizes transactions
Share of top AI web products offering each model: subscriptions 84.1%, usage/credits 63.6%, advertising 13.6%, transaction fees 2.3% (a16z, 7th ed.).
The largest consumer internet companies earn from ads and transactions. AI hasn't unlocked transactions yet, and ads are only arriving now: ChatGPT's ad business passed a $1B annualized run rate on Aug 31, 2026, per OpenAI's David Dugan (PPC Land).
4. After coding, commerce is the next job for agents
In a 1,502-member power-user group for AI personal assistants (data by AssistantBenchmark, charted by a16z), the most-discussed tasks after coding (103) were travel (66), shopping & returns (58) and finance & investing (56).
Early agents already touch money: Instinct passed 100K users growing ~10% a day (The Information via RuntimeWire); on the podcast, the company-stated figures were ~40% of users connecting a card within three weeks and over $1,000 average spend in month one.
5. The merchant is the wall, and the rails are fragmenting
On the podcast: Amazon doesn't let Meta's Muse browse and buy on its site, while Muse partners to buy via Shopify. Every store now decides which agents get in.
In about a year the industry shipped:
- Google AP2, 60+ partners — Google Cloud
- OpenAI + Stripe ACP / Instant Checkout — Stripe
- Visa Trusted Agent Protocol (AI traffic to US retail sites +4,700%) — Visa
- MCP moved to the Linux Foundation, 97M+ monthly SDK downloads, 10K servers — MCP blog
- Shopify + Google UCP — Shopify
- Tempo mainnet + MPP, 100+ services at launch — The Block
- x402: 75.41M transactions, $24.24M in 30 days (Aug 2026) — x402.org
What this means
There won't be one agent. ChatGPT leads paid subscribers by ~3x over Claude or Gemini, yet Claude passed Gemini this year (a16z). Many agents × many protocols × millions of stores is an N×M integration problem nobody wants to solve store by store.
Agentic commerce needs: one merchant integration any agent can reach; a machine-readable catalog with fixed-price quotes; agent-native payment straight to the seller; and verifiable receipts and limits that build trust.
Where APIbase fits
That is the layer we're building. APIbase Integrator gives a store one link and an MCP storefront: any MCP-capable agent can discover it, get a quote and pay in USDC via x402 (Base) or MPP (Tempo), straight to the merchant's wallet — APIbase never holds the money. The same endpoint already serves 1,300+ API tools from 400+ providers to agents, pay-per-call.
Top comments (0)