Technology

The Structure Behind AI Agents Starting to 'Shop' — From a Experiment to a Trillion Future

⚡ FAST READ AI agents are fundamentally reshaping commerce, and companies that control payments, automation, and foundation models may establish a new era of platform dominance. Pattern: Platform Power × Winner Takes All Base scenario: Companies that dominate the infrastructur...

Nowpattern

10 5月 2026 — 10 min read

⚡ FAST READ

AI agents are fundamentally reshaping commerce, and companies that control payments, automation, and foundation models may establish a new era of platform dominance.

Pattern: Platform Power × Winner Takes All

Base scenario: Companies that dominate the infrastructure of the AI agent economy will levy platform taxes at the payment, automation, and inference layers, rewriting the rules of commerce.

Watch: How Visa and Mastercard compete to set the standard protocol for AI agent payments.

Why it matters: In the same week that a Reddit experiment giving an AI agent $50 to buy a PC went viral, Visa predicted that "millions will shop through AI agents by the 2026 holiday season." Meanwhile, Gartner announced that "by 2028, AI agents will mediate 90% of B2B purchasing, reaching $15 trillion in scale." This is not merely a tech experiment. It marks a turning point where the fundamental structure of commerce — who buys, who sells, and who processes payments — is being rewritten.

Genre:#Technology #Economy & Trade

Event:#Tech Breakthrough #Structural Shift

Dynamics(Nowpattern):#Platform Power #Winner Takes All

📝 Summary: ⚡ FAST READ AI agents are fundamentally reshaping commerce, and companies that control payments, automation, and foundation models may establish a new era of platform dominance. Pattern: Platform Power × Winner Takes All Base scenario: Companies that dominate the infrastructure of the AI agent...

What happened

February 2026 — A Reddit user published an experiment giving an AI agent $50 to autonomously purchase a PC. The agent automatically performed price comparison, spec evaluation, and purchase decisions
February 2026 — A fully automated AI voice agent reservation workflow for restaurants was published on the n8n platform, integrating VAPI + PostgreSQL + Twilio for 24/7 unmanned operation
November 2025 — Gartner predicted "by 2028, AI agents will mediate 90% of B2B purchasing, reaching $15 trillion in scale" (Gartner IT Symposium/Xpo 2025)
December 2025 — Visa predicted "millions will shop through AI agents by the 2026 holiday season," announcing its Intelligent Commerce program
October 2025 — Visa and Mastercard simultaneously announced agentic payment tools. Mastercard introduced "Agent Pay" while Visa launched "Trusted Agent Protocol"
October 2025 — n8n raised $180M (led by Accel, with NVidia NVentures participation) at a $2.5B valuation. ARR reached $40M
January 2026 — The World Economic Forum announced "the AI agent market will reach $236B by 2034"
2025 — AI agent-driven retail site traffic on Black Friday surged 805% year-over-year. AI-led sales exceeded $22B globally
2025 — Coinbase announced Agentic Wallets, enabling AI agents to autonomously hold, send, and trade cryptocurrency
February 2026 — Cloudflare announced its agentic commerce security framework, providing AI agent authentication infrastructure compatible with Visa and Mastercard

The Big Picture

Historical Context

For the past 30 years, e-commerce has been built on the premise that "humans shop using browsers." Amazon (1994), Rakuten (1997), Shopify (2006) — every platform was designed with the fundamental assumption that "humans click."

That assumption began to crumble in late 2025. Three catalysts converged.

First, the dramatic improvement in LLMs' (Large Language Models) tool-use capabilities. GPT-4o, Claude 3.5, and Gemini 2.0 acquired the ability to call APIs, navigate the web, and input payment information.

Second, payment infrastructure adapted. When Visa and Mastercard simultaneously announced agentic payment tools in October 2025, the institutional foundation for "AI making payments" was established.

Third, no-code automation platforms matured. n8n achieved $180M in funding and a $2.5B valuation in 2025, providing an environment where non-engineers can build and deploy AI agents.

All three elements aligned simultaneously in late 2025 through early 2026. Historically, this parallels the structural transformation of the mid-1990s when the internet entered commerce.

When PizzaHut launched online ordering in 1994, many laughed: "Ordering food from a PC? That's niche." Two years later, Amazon began selling books, and within five years, e-commerce had fundamentally transformed commerce. The same pattern is now beginning — this time as a shift from "humans buy" to "AI buys."

Stakeholder Map

Actor	Public Stance	Real Motive	What They Gain	What They Lose
Visa / Mastercard	Improving payment security	Monopolize the payment layer of the agent economy	Standard-setter status for AI agent payment protocols	Traditional fee model compressed by low-margin machine-to-machine transactions
n8n / No-code Platforms	Democratizing automation	Capture platform tax as the infrastructure layer of the AI agent economy	Enterprise AI automation dependency (ARR $40M, $2.5B valuation)	Open-source commercialization limits, risk of AWS/Google forks
OpenAI / Anthropic / Google	Safe AI development	Dominate the OS (foundation model) of the agent economy	Explosive growth in API calls and usage-based revenue	Regulatory backlash from agent runaway incidents
Amazon / Retail Platforms	Enhancing consumer experience	Become the default purchase destination for AI agents	Lock-in of agent-driven sales	Constant exposure to price-comparison AI erodes brand loyalty
Consumers	Convenient, cheaper shopping	Save time and effort	Price optimization, time savings	Black-boxed purchasing decisions, loss of privacy
SMBs (Restaurants, etc.)	Operational efficiency	Cut labor costs, achieve 24/7 availability	Automated reservations via n8n voice agents at $50/month, 24/7	Elimination risk: AI-incompatible businesses become invisible

Data Points

$15 trillion — Gartner forecast: Total B2B purchases mediated by AI agents by 2028
90% — Gartner forecast: Share of B2B purchases mediated by AI agents by 2028
$236B — WEF forecast: AI agent market size by 2034
805% — YoY increase in AI agent-driven retail traffic on Black Friday 2025
$22B+ — Global AI-driven sales on Black Friday 2025
$2.5B — n8n valuation as of October 2025 ($180M raise, Accel-led)
$40M ARR — n8n annual recurring revenue (as of July 2025)
36% — Share of consumers willing to let AI agents shop on their behalf
$50/month — Server cost for n8n self-hosted, 150K executions/month

The delta: On the surface, this looks like a tech experiment in "AI shopping." But the essence is a civilizational shift in "the role of humans in commerce." The fact that Visa and Mastercard simultaneously entered agentic payments reveals their conviction that "the next customer is not a human but an algorithm."

Between the Lines — What the Headlines Don't Say

Visa and Mastercard publicly tout improved security for AI agent payments, but their real goal is to monopolize the payment layer of the agent economy and maintain fee revenue. However, machine-to-machine transactions compress fees, forcing them to find new revenue sources. Foundation model companies like OpenAI also aim for explosive API call growth and usage-based revenue under the guise of "safe AI development," while fearing regulatory crackdowns from agent runaway risks.

NOW PATTERN

Platform Power × Winner Takes All

Whoever controls the infrastructure layer of the AI agent economy — payments, automation, foundation models — writes the rules of commerce for the next 30 years.

Platform Power: The War Over "Who Controls the Agent's Wallet"

In October 2025, Visa and Mastercard announced agentic payment tools in the same week. This was no coincidence. Both companies built their empires over 30 years on the act of "a human presenting a card." They moved simultaneously because they could see a future where that premise vanishes.

Visa's Intelligent Commerce program gives developers APIs that plug identity checks, spending controls, and tokenized card credentials into AI agents.
— Visa Investor Relations, December 2025

Mastercard's Agent Pay uses 'agentic tokens' and requires that AI agents be registered and verified before they can initiate payments.
— Mastercard Press Release, October 2025

Visa's "Trusted Agent Protocol" and Mastercard's "Agent Pay" are technically mechanisms to connect AI agents to payment networks, but strategically they represent a hegemonic war over "the standard for the next 30 years of payment infrastructure."

The structural breakdown:
- Layer 1 (Foundation Models): OpenAI / Anthropic / Google — The AI's "brain"
- Layer 2 (Automation Infrastructure): n8n / Make / Zapier — The AI's "hands"
- Layer 3 (Payment Infrastructure): Visa / Mastercard — The AI's "wallet"
- Layer 4 (Crypto Payments): Coinbase Agentic Wallets — The AI's "other wallet"

The winner at each layer collects "platform tax" from the agent economy. This mirrors the competition between Amazon, PayPal, and Visa during the e-commerce dawn of the 1990s — but with a fundamental difference: "the customer is not a human but an algorithm."

Unlike humans, AI agents have no brand loyalty. They don't buy on emotion. They compare all options in 0.1 seconds. This means "complete efficiency in price discovery," potentially triggering a structural collapse of brand premiums and retail margins.

Winner Takes All: What a $50 Experiment Reveals About the "AI-Native Economy"

The Reddit experiment of "giving an AI agent $50 to buy a PC" is still rough around the edges technically. But what this experiment demonstrates is not technological maturity — it's a cultural shift: "humans have begun crossing the psychological threshold of delegating purchasing decisions to AI."

36% of customers are open to AI agents making purchases on their behalf.
— World Economic Forum, January 2026

By 2028, 90% of all B2B purchases will be handled by AI agents, channeling more than $15 trillion in spending through automated exchanges.
— Gartner IT Symposium/Xpo 2025

The 36% of consumers willing to let AI shop for them closely resembles early e-commerce adoption numbers. In 1998, about 30% of consumers said they were comfortable entering their credit card number online. Five years later, that figure exceeded 80%, and e-commerce became irreversibly established.

Gartner's prediction essentially forecasts a similar "trust threshold breakthrough" occurring between 2025 and 2028. In B2B specifically, 90% of MRO (Maintenance, Repair, and Operations) repeat orders are automatable, making AI agent integration technically and organizationally low-barrier.

Meanwhile, the restaurant AI voice agent case demonstrates "winner takes all" dynamics. An automated reservation system built with n8n + VAPI + PostgreSQL operates 24/7/365 at $50/month in server costs. Zero labor cost 24-hour availability gives adopting businesses an overwhelming competitive advantage. Conversely, non-adopters get eliminated as "the restaurant that never answers the phone."

This echoes the digital divide for SMBs, but this time the divide is crueler. The 2000s question of "whether or not to have a website" led to gradual elimination. The AI agent economy creates a more direct exclusion mechanism: "whether or not you appear in what AI agents recommend."

Where the Dynamics Intersect

The two dynamics — Platform Power and Winner Takes All — are inseparably linked in the AI agent economy.

When platform layers (Visa/Mastercard/n8n/OpenAI) dominate payment, automation, and inference infrastructure, only businesses that ride that infrastructure enter the AI agent's "consideration set." Businesses that can't get on the infrastructure literally become "invisible to AI."

This resembles SEO (Search Engine Optimization) from 20 years ago, but with a fundamental difference. In SEO, humans were the final decision-makers. In the AI agent economy, recommendation, judgment, and purchase are all performed by AI. Humans only hold the intent to "want to buy something," delegating the entire remaining process to AI.

If Gartner's prediction of "90% B2B purchasing automation by 2028" materializes, the structure of global commerce will irreversibly shift from a "human-choice economy" to an "algorithm-choice economy." In this transition, those who control the infrastructure layer capture 30 years' worth of platform tax.

Pattern History

1994: The Birth of E-Commerce — From PizzaHut Online Ordering to Amazon

In 1994, PizzaHut launched internet-based pizza ordering. That same year, Jeff Bezos founded Amazon. By 1998, PayPal began providing payment infrastructure.

At the time, consumer trust in "shopping on the internet" hovered around 30%. Yet within five years it exceeded 80%, and e-commerce became irreversibly established. Crucially, companies that controlled the three layers of "payment infrastructure," "marketplace," and "logistics" during this transition became the winners for the next 30 years: Visa/PayPal (payments), Amazon (marketplace), FedEx/UPS (logistics).

Structural parallels to today: Emergence of a new purchasing channel (browser → AI agent). Consumer trust at 30-36%. A battle for payment infrastructure.

2007: iPhone and the Structural Shift to Mobile Commerce

The iPhone's 2007 launch shifted the e-commerce battleground from desktop to mobile. Apple Pay (2014) established the mobile payment standard, and the App Store imposed a 30% platform tax.

"Shopping on a phone" was initially niche, but by 2020 over 70% of all e-commerce was mobile. Once again, infrastructure layers (Apple/Google) captured platform tax, and non-adapting retailers were eliminated.

Structural parallels to today: Generational shift in the purchasing interface. Infrastructure layer (Apple → Visa/n8n) captures platform tax. Non-adapting businesses eliminated.

What History Shows

The pattern history reveals is clear. Every time the "interface" of commerce changes, three things happen:

① Early consumer trust starts around 30%, but exceeds 80% within five years
② Companies that control the infrastructure layer collect platform tax for 30 years
③ Businesses that fail to adapt to the new interface are eliminated within 5-10 years

The 2026 AI agent economy has entered the third cycle of this historical pattern. "Buy via browser" → "Buy via smartphone" → "AI buys for you." With each cycle, human involvement in decisions decreases, and the scope of automation expands.

What's Next

Base Case — Gradual Penetration — B2B Leads, Consumer Market Takes Off in 2028 (Probability: 55-65%)

B2B MRO (Maintenance, Repair, Operations) procurement automation leads the way. By 2028, 50-70% of B2B purchases will go through AI agents. Consumer-facing AI shopping begins with Visa/Mastercard pilots in late 2026, reaching mainstream adoption by 2028. AI voice agent adoption among SMB restaurants exceeds 20% by end of 2027. No-code platforms like n8n become the standard for "non-engineer AI agent building."

Investment/action implications: Watch automation platform stocks (n8n is still private). Visa/Mastercard benefit as first movers in agentic payments over the medium term. Restaurant/retail automation SaaS (AI voice reservations, auto-ordering) — 2026-2027 is the investment window.

Bull Case — Explosive Growth of the Agent Economy — "AI-Native Commerce" Goes Mainstream by 2027 (Probability: 15-25%)

OpenAI/Anthropic models achieve full autonomous web browsing and payment execution. Visa/Mastercard payment tokens become standardized, and AI agent-driven sales exceed 20% of total e-commerce by the 2027 holiday season. Gartner's $15T forecast begins materializing ahead of schedule.

Investment/action implications: Increase exposure to foundation model companies (Microsoft/OpenAI ecosystem, Google, Amazon). Agent payment infrastructure companies (Cloudflare, Stripe) see rapid growth. Traditional advertising/marketing firms face structural headwinds from "AI doesn't see ads" problem.

Bear Case — Regulatory Walls and Security Incidents — 2-3 Year Delay for the Agent Economy (Probability: 15-25%)

A major AI agent misorder or fraud incident triggers consumer protection regulation. Strict enforcement of the EU AI Act mandates additional authentication/approval processes for agent payments. Consumer distrust of AI intensifies, keeping agent-mediated purchases below 5% of total commerce through 2028.

Investment/action implications: Cybersecurity firms (CrowdStrike, Palo Alto Networks) benefit. Watch AI insurance and risk management startups. In this extended-timeline scenario, existing large-cap retail stocks become safe havens.

Triggers to Watch

Visa Intelligent Commerce goes live: Q2-Q3 2026 transition from pilot to commercial deployment. Success accelerates the agent economy
First major agent payment incident: An AI agent misorder or fraud case would trigger regulation and confidence erosion toward the bear scenario
n8n IPO / major acquisition: IPO of a no-code AI automation platform would accelerate market awareness of agent economy infrastructure
OpenAI "Computer Use" full rollout: If foundation models achieve full autonomous web browsing and payment, B2C AI agent adoption accelerates dramatically
Gartner B2B automation interim report: Early 2027 Gartner report confirming actual automation rates would validate the market direction

OPEN LOOP

Next trigger: Discovery of security vulnerabilities in AI agent payments and resulting regulatory debates (2027 Q1-Q2).

Continue tracking: Data privacy and ethical challenges in the AI agent economy: transparency and accountability in algorithmic purchasing decisions.

Sources: