Which stores opened checkout to AI agents this week, and why did fewer payers send more money?
Shoppers in the US can now buy from Newegg without leaving Google's AI Mode or the Gemini app. TikTok said it is building
one-click buying and a shopping agent, and x402 payments on Solana can now be batched into a single settlement. Our own
readout: the sellers we track took in 17.7% more than the week before, from fewer payments and far fewer payers. This
issue covers 2026-10-05 to 2026-10-11.
What agents did this week
Every figure below is recorded in this issue's claims ledger and can be recomputed with the published score-v1.0.0
method. Our nightly sync stopped for a week after a scheduling fault (see the note at the end of this section); the
missed days were synced before this week was computed, and every day in it is fully covered.
Scope. These numbers cover USDC received on
Base by the
2,0886 seller addresses in our catalog, which is
built from the x402 Bazaar. They are not "the x402 market." "Verified" means a payment of at least $0.01 that is not a
self-transfer and not from a sender flagged as an exchange or a fan-out. It is not the seller-level "verified tier."
Every number in this section is verified on-chain — you can recompute it.
More dollars, fewer payments, fewer payers. The sellers we track received
$249,962.351 , up from
$212,326.692 , across
172,0671 verified payments. The average payment rose from
$1.202 to
$1.451 ; the median stayed at
$0.021 .
6851 sellers were paid, against
7542 , and
2,5141 distinct addresses paid them, against
3,6972 .
One seller lost two-thirds of its payers. AX1 Console, the two-cent token-verdict tool from last week's issue,
received 81,8043 payments
($1,636.083 ) from
4523 senders, against
96,5103 payments from
1,3583 senders the week before. It still received nearly half
of all payments. (A sender can pay several sellers, so sender counts cannot be subtracted the way payment counts can.) Every other seller together received
90,2633 payments, up from
80,6703 .
Dust, shown beside verified. Of 291,3924
transfers to tracked sellers, 105,5434 were
dust (36.2%), moving
$399.384 together. The week before:
139,8444 dust transfers (42.5%),
$431.964 .
Where the dollars went. The same three categories led again: search-retrieval
($183,961.125 ), payments-infrastructure
($183,344.225 ) and storage-state
($170,503.405 ). Inference fell from
$48,558.145 to
$27,283.425 . Sellers whose listings have left the catalog
received $19,456.215 between them. A seller listed in
several categories counts in each, so category totals add up to more than the week's total.
The catalog shrank for the first time. After roughly doubling in two weeks, the x402 Bazaar peaked at
35,1627 items in our 2026-10-07 snapshot and fell to
32,0447 by 2026-10-11. It is also less concentrated: in our
2026-10-13 snapshot the ten largest payment addresses hold
48.1%7 of listings, against 59.8% a week earlier.
How people used agents this week
Instinct users got shopping lists they had not asked for
report
Three early users of Instinct, a personal AI agent, told Business Insider that it sent them product
lists built from things they had mentioned in earlier conversations. One user said:
"I didn't ask for anything." Business Insider wrote that the lists
"were not labeled as ads", and that it could not tell whether Instinct earns
money from the purchases. Instinct did not comment.
who · date Thibault Spirlet, Business Insider · 2026-10-05
caveat Three users' accounts, as reported. We have not verified them.
changes our ledgers? No.
Shoppers described what happened when they let agents buy for them
test
Business Insider, 2026-10-10. One user asked Instinct for chicken wings and drumsticks
without giving an amount. "Later, 20 pounds of chicken wings showed up on his doorstep."
Another had Meta's Muse buy ballet shoes at Nordstrom after approving the payment; the agent left out her
loyalty number. A third uses single-use cards: "I wouldn't give it access to my best credit card."
The New York Times, 2026-10-06. On The Daily, reporter Eli Tan described connecting Muse to his
cards, bank accounts, email, calendar and DoorDash. On air, Muse offered to order the host's lunch on a card it
had on file, after he approved the order. The transcript does not say whether the order went through.
caveat Personal accounts, as reported. Neither outlet ran a controlled test.
changes our ledgers? No.
What numbers were published this week?
report
Two surveys were published with dates in this week. Neither counts purchases completed by agents.
American Express: what merchants and shoppers expect. In a survey of 502 US
business leaders, "62% of Merchants expect AI agents shopping on behalf of customers to have a moderate or
significant impact on their business". In a companion survey of 2,005 US online shoppers, 52% were comfortable
letting an agent buy low-cost everyday items, and "just 15% say they would feel comfortable" with
an agent buying high-value items.
who · date American Express, surveys run by Teneo (August 2026) · 2026-10-06
what it counts Stated expectations and comfort, not behavior. Amex says the figures are rounded.
Shopify: plans for the holiday season. Shopify says
"65% of shoppers say they'll use AI for at least one shopping task this season", and
"Only one-third of shoppers say they trusted an AI agent to buy on their behalf."
who · date Shopify, survey of 18,000 shoppers in nine countries by Sapio Research (July 2026) · 2026-10-06
what it counts Survey answers about intentions and past trust, not purchases.
The big moves
Newegg's checkout went live inside Google's AI Mode and Gemini
move
Shoppers in the US can now buy eligible Newegg products without leaving AI Mode in Google Search or the Gemini app,
paying with Google Pay. The checkout runs on the Universal Commerce Protocol
(UCP). The release says "Newegg is the merchant of record for every order"
and that it is "live now for shoppers in the United States".
who Newegg; Google (Ashish Gupta, VP/GM of Merchant Shopping, is quoted)
counts vs dollars No order counts or sales were published.
changes our ledgers? No. It is consistent with the "live" status already
recorded for Google's UCP checkout.
TikTok said it is building one-click buying and a shopping agent
move
TikTok announced Buy Direct, for buying from a brand "with just one click" from the For You feed,
and Shopping Assistant, an agent that answers questions about a product and can complete the purchase in the chat. Its
wording: "TikTok is building new Agentic Commerce shopping experiences". It named Salesforce,
Shopify, Shoplazza and Stripe as partners, and gave no launch date or markets.
caveat TikTok says "building"; TechCrunch's headline says it is rolling out. We print TikTok's word.
changes our ledgers? Pending verification. Candidate /economy entry, as announced.
Constructor added Stripe checkout inside the shopping agents it runs for retailers
move
Constructor makes search and shopping agents that retailers put on their own sites. It added Agentic Checkout, so a
shopper can pay with Stripe's Link or another method without leaving the agent's chat. Constructor calls itself
"the first to offer Stripe-powered checkout directly within an onsite shopping agent".
counts vs dollars Constructor says the number of companies using its agents
"increased 900%" over the past year, with no base number. No retailers or order counts were named.
"First" is the company's own claim.
changes our ledgers? No.
x402 payments on Solana can now be batched: many payments, one settlement
move
Solana's account announced: "Batch payments are live for @x402 on Solana via @PayAINetwork and @BlockRunAI".
An agent opens a USDC payment channel, pays as many times as it needs, and settles once on-chain. The use named is
paying for many AI model calls with a single transaction.
who · date Solana; PayAI Network; BlockRunAI · 2026-10-07
sourceSolana's post on X (date from the post's ID; about 117,000 views, read 2026-10-13)
why it matters to our numbers Our readout covers x402
payments in USDC on Base, not Solana. Batched payments also settle less often than they are made, so on-chain counts would
undercount the payments behind them.
changes our ledgers? No.
Meta, Sierra and partners announced a standard for how personal agents deal with businesses
move
The Personal Agent Protocol "defines how personal agents interact with businesses": an agent starts
as a guest, then signs in for the user with read-only or write access, and works through the business's website, its APIs or
its own agent. Named partners include Shopify, Stripe and Walmart. The authors
"plan to publish the v0.1 specification later this month". Payments are described as a possible later
extension.
who · date Sierra (Bret Taylor and Clay Bavor) with Meta · 2026-10-06
caveat Announced, with no published specification yet.
changes our ledgers? No.
Briefly
Each item below is one sentence, dated and sourced. Our ledger flag is "No" for all three.
2026-10-05: Instinct began letting early users add a shared agent to group chats, to plan trips, buy
tickets and split the cost. Source: Noah Shinn's post on X.
2026-10-08: Crossmint released a developer toolkit that stores a card, uses Visa and Mastercard agent
credentials where the card is eligible, and checks out through Shop Pay, UCP or browser automation; it says
"It is live and self-serve today." Source:
Crossmint's release.
2026-10-09: HSBC and Ant Digital Technologies ran a test in which an agent paid for a digital service with a
tokenized bank deposit on a test network; the release says it "does not represent a commercial launch or live customer
proposition." Source: the joint release.
Trading agents this week
eToro connected investment accounts to Claude, ChatGPT and Grok, with approval on every action
move
eToro released connectors that let an AI assistant read a user's eToro account, or, where offered, act on it. In eToro's
words, "Any action is proposed for review and approval", and deposits and withdrawals stay inside
eToro. The connectors are "not currently available to clients in the US, Australia or Singapore".
counts vs dollars eToro also cites its own survey: 56% of retail investors use or
are open to using AI tools to pick or change investments. That is stated openness, not use. No usage figures for the
connectors were published.
changes our ledgers? No.
A new "AI bot made a fortune" post offers nothing to check
claim
A post on 2026-10-07 by @SimslearnAi on X, seen about 42,700 times by 2026-10-13, describes a Polymarket bot built with
Claude and lists: "$200. 27 days ago. Current balance: $14,300. 271 trades. 74% win rate."
It names no account and no wallet. It ends: "Comment “Claude” if you want the framework."
The "$78,083 in 5 days" claim from last week's issue was posted again this week by other accounts, nearly word for word.
Status, if we run the check: No public record offered, unless an account turns up. Polymarket accounts are
public, so a named account would make it checkable.
sourcethe 2026-10-07 post
(date and view count from X's public data, read 2026-10-13). Accounts that only reposted these claims are not named.
What our data shows: the token-commentary tool still dominates the trading category
our data
Last week we reported that most payments in our trading-execution category went to AX1 Console, a two-cent tool that our
category rule files under trading because its description says "not trading advice." It received
81,8043 payments this week, worth
$1,636.083 . The whole category fell out of the top eight
by dollars: the eighth-largest category took in
$4,627.475 , so trading-execution took in less. These are payments for tools, not trading results.
method Categories come from
capability-v1, a published text rule. The rule has not changed.
The account was 58 days old when the claim was first posted, not 14 months.
The claim's "40,266 trades" was the account's count of markets. It had made about 3.3 million trades by then.
The $794,000 matches the profit figure the account's profile showed that day. Polymarket's newer profit figures for the
same day are far lower.
Nothing public ties the account to a Brazilian student or to any AI model.
The claim first appeared on 2026-05-22, posted by the same account as the "fired Citadel quant" claim we checked
earlier, with the same referral codes. All six
checks are on the fact-check page.
Where are the current numbers?
This issue is a dated document. Its figures are frozen to 2026-10-13 and recorded in its claims ledger. The weekly
answers to eight standing questions are at the weekly statistics. The
market dashboard, rankings and directory show
generation 10, published 2026-10-13. For which of this week's moves changed what agents can buy, see
/economy.
This week's verified figures: the Monday-to-Sunday UTC window 2026-10-05 → 2026-10-11, every day
coverage-published at 1.0 (all 1,292 tier-A sellers synced past each day's close), Base blocks 52,184,527 → 52,486,926,
buyer flags version flags-2026-10-13, computed 2026-10-13T08:26:57Z with the chain observed through block 52,544,868
(2026-10-13T08:11:23Z). Method score-v1.0.0, payment-level predicate. Recompute from
/api/v1/stats/weekly and
the methodology.
Prior week: 2026-09-28 → 2026-10-04, same method and computation. Pulse #5
printed it as $212,267.97 across 175,953 payments, 723 sellers paid, 3,685 senders and 86 first-time earners. It now reads
$212,326.69, 177,180, 754, 3,697 and 103, because sellers synced after Pulse #5 was computed added payments they had
received that week. Our weekly records are append-only, so the revision is visible.
AX1 Console is the seller address 0x7284d41b5b852f2bd4c99bdf95043d84452d299c (two x402 Bazaar listings at $0.02,
both still listed). Its counts use the same blocks and predicate as fn-1 and fn-2, filtered to that address. The figures for
every other seller are the week's totals minus AX1 Console's.
Dust is any transfer under $0.01 (the dust floor). Gross volume
including dust and excluded senders was $251,040.86 this week and $214,063.10 the week before. That is a context figure,
never labeled verified.
Categories come from capability-v1. A seller listed in
several categories counts in each, so this week's category totals sum to $649,789.70 against the $249,962.35 deduplicated
total. "Listings have left the catalog" is the group the statistic labels "(no active listing)": 29 sellers paid this week
with no active Bazaar listing.
Seller addresses in our catalog when the week was computed (2026-10-13).
Catalog sizes are the item counts of our daily snapshots
bazaar-2026-10-07 and
bazaar-2026-10-11. The top-ten share comes from snapshot
bazaar-2026-10-13 (48.1%) and bazaar-2026-10-06 (59.8%), as published in /api/v1/meta
on those days.