Agentic Commerce and the AI Tech Stack | Dorman Review

The Dawn of the Machine Economy: Anatomy of the Agent's Stack

'Soon people will have AI agents browse, select, purchase and manage on their behalf.'

• Jack Forestell, Chief Product and Strategy Officer, Visa

❓ What You'll Learn

Why AI agents can find anything online but still cannot pay for it.
How the new payment rails work, and why the big protocols are fighting over them.
Why online stores are being rebuilt for machines instead of eyeballs.
How fraud systems break down when the buyer is software.
Where the money is going as this stack gets built.

💎 Why It Matters

Agentic commerce is expected to move somewhere between $1.7 trillion and $5 trillion by 2030.

Around 85% of current fraud systems cannot handle agent traffic, so the entire security model has to be rebuilt.

Money is already following the shift. Startups building the payment rails for agents, including Basis Theory, Nekuda and Skyfire, have raised roughly $50 million between them.

🔍 Problem

Agents research instantly but fail at checkout, unable to swipe cards or process billing autonomously.

Millions of independent stores remain invisible to AI buyers because merchants must manually integrate new protocols.

Multi-agent systems are fragile, frequently stalling due to high latency, unresolved logic loops, and soaring token costs.

💡 Solution

Universal APIs replace storefronts, allowing machines to read catalogs and execute checkouts uniformly.

Agents use single-use tokens and signed permissions instead of card forms, proving consent without exposing sensitive data.

Crypto rails handle micro-transactions, with on-chain wallets enforcing strict spending limits while complex AI logic runs off-chain.

🏁 Players

The AI Platforms & Agent Surfaces

Perplexity • Operating the 'Buy with Pro' service, embedding native checkout directly into conversational search experiences.
Meta • Earmarking up to $135 billion for AI infrastructure and developing native agentic shopping tools to surface products from its massive advertiser catalog.
Anthropic • Created the Model Context Protocol (MCP), providing the foundational open-source infrastructure for AI agents to coordinate tasks and access data securely.

The Protocol Builders

OpenAI and Stripe • Creators of the Agentic Commerce Protocol, the standard behind Instant Checkout in ChatGPT. It uses Shared Payment Tokens so an agent can pay without ever holding the buyer's card details.
Google • Built the Agent Payments Protocol with more than 60 partners including Mastercard, PayPal and American Express. Its focus is proof: every purchase carries a signed record of what the user actually approved.
Coinbase • Revived a forgotten piece of the web with x402, which turns any API into something a machine can pay for instantly using stablecoins. Ideal for payments worth fractions of a cent.
Shopify • The quiet kingmaker. By switching on agent checkout across its merchant base, it made over a million stores reachable by AI overnight.

The Merchant Aggregators

Firmly • Backed by Ark Invest and FJ Labs, building a unified Buy Now platform that lets merchants sell across AI chatbots, social platforms, and connected TVs through a single integration.
Commercetools • Launched Agent Gateway, giving autonomous systems full context of massive catalogs (10M+ items) via an API-first, headless architecture.

The Payment Networks

Visa • Opened its network to agents through Intelligent Commerce and added a Trusted Agent Protocol to separate genuine AI buyers from malicious bots.
Mastercard • Answered with Agent Pay, letting agents recommend a product and then choose the best way to pay for it.

The Startups Building the Rails

Basis Theory • Raised $33 million to build payment vaults made for AI commerce, and now leads the Agentic Commerce Consortium.
Skyfire • Runs a 'Know Your Agent' system, giving every agent a verifiable identity before it is allowed to spend anything.
Nekuda • Raised $5 million for a secure agent wallet and now works directly with Visa on permission rules for autonomous spending.

The Trust and Orchestration Layer

Forter • Rebuilding fraud detection for a world where the buyer is software, so good agents get through and bad ones do not.
Cloudflare • Sits at the edge of the internet and decides which bots are welcome, which makes it the unofficial gatekeeper of agent traffic.
LangGraph and CrewAI • The frameworks that let several agents work together in defined roles, which is what real autonomy actually requires.

🔮 Predictions by Aybars D.

The rise of the agentic product manager • Brands will stop hiring people to perfect the look of a cart. The new job is making sure an AI can read your catalog properly, because that is where the sale now happens.
Whoever simplifies the mess wins • There are already four competing standards. The companies that hide all of them behind one simple integration will capture the millions of merchants nobody else can reach.
Business buying goes first • Agents will handle server capacity, supplies, and recurring invoices long before anyone lets them buy a watch. Clean data and repeat orders make B2B the easy starting point.

☁️ Opportunities

The checkout gap. Enormous money has gone into payments and discovery, yet almost nobody is solving the final step of actually completing the order.
Trust at the edge. Tools that tell a real buying agent apart from a hostile bot, before either reaches the store, will become essential infrastructure.
Agent wallets with hard limits. Smart contracts that cap spending no matter what the AI decides, while leaving it free to analyse the market, solve the biggest fear holding this back.

🏔️ Risks

The coverage gap • As long as every merchant has to integrate by hand, agents will only be able to shop across a small slice of the open web.
Corporate paralysis • Brands that still treat their product data as a marketing asset, rather than as working infrastructure, will not appear in the machine economy.
The trust deficit • Handing real spending power to software makes consumers and banks nervous for good reason. Without airtight proof of consent, adoption stalls.

🔑 Key Lessons

Your catalog is the new storefront. Design used to win customers. Now clean, machine-readable data does, because the agent never sees your homepage.
Identity is the real currency. Banks will only approve autonomous purchases if the agent can be tied back to a verified human beyond any doubt.
One model cannot do it alone. Finding, checking, and paying are separate jobs, and they need separate agents working in sequence.

🔥 Hot Takes

Shopping websites are becoming legacy technology. Once software can search the whole market, verify the specs, and buy in seconds, asking a human to scroll through tabs looks like a waste of a life.
These 'open' standards are land grabs. Google and OpenAI are not being generous. They are racing to own the toll booth on a transaction network worth trillions.

😠 Haters

'People will never let a machine spend their money unsupervised.'

They already do. Auto-pay, algorithmic trading and rolling subscriptions all spend without asking. Agents just add judgment to the process, inside limits the user sets.

'The safe move is blocking all automated traffic.'

Blocking every bot means blocking your future customers. The winners will let verified buying agents through and stop the rest.

🔗 Links

Agentic Commerce Protocol • The open standard from OpenAI and Stripe for agent checkout.
Stripe on Shared Payment Tokens • How agents pay without ever touching card details.
Google Agent Payments Protocol • The authorization layer built with 60 payment partners.
x402 • The open standard for instant machine payments over stablecoins.
Visa Intelligent Commerce • A global card network opening its rails to AI agents

See you in the next one. To be continued..

Written by Aybars Dorman | July 23, 2026

#AI #Technology #Innovation #Ecommerce #ArtificialIntelligence #Payments #DeepTech #Fintech #AgenticCommerce #MachineEconomy #AIAgents #Web3 #Stablecoins #ProtocolWars #AutonomousPayments #RetailTech #DormanReview