Source: a16z crypto; Compiled by: Golden Finance

Agentic Commerce has arrived.

ACP (Agentic Commerce Protocol) and UCP (Universal Commerce Protocol) promise to enable direct ordering in ChatGPT and Gemini. Soon, hundreds of millions of consumers around the world will be able to discover better products, merchants will see improved conversion rates, and platforms will take a 5-10% cut.

But ChatGPT’s checkout feature is just an incremental improvement. It won’t reshape society the way the internet did in the early 2000s. What can really change society is "Open Agentic Commerce".

To understand why, we have to go back to the 1990s.

There were two competing versions of "the Internet":

1. AOL (America Online) Edition: One price, includes email, weather, additional approved content, and ultimately the entire Time Warner library.

2. Open protocol version: HTTP, DNS, HTML, and a browser called Mosaic.

The Mosaic at the time looked very crude in comparison. Mosaic There are so few sites on the Internet that no search is necessary; an alphabetical index will suffice. Eight years later, AOL completed a $350 billion merger of equals with Time Warner. The market seems to have made its decision: Curated content bundles are the way of the future.

But soon after, Mosaic and open protocols finally won out, and human civilization entered the digital age.

Why? Imagine if the Walled Garden had won: Mark Zuckerberg wanted to start Facebook in 2004, he needed to sign a distribution agreement with AOL first; two Stanford kids (the founders of Google) wanted to index the web, they had to ask for permission from CompuServe first; a man wanted to sell books in his garage (Bezos), he had to make a demo pitch to MSN's content team.

"Go back to school, boys," the big guys would say. Then none of this will happen. The entire digital economy we take for granted will simply not exist.

Open protocols mean no gatekeepers. Anyone with a server and a domain name can reach the entire Internet. The fringes are innovating, the center cannot keep up, and the result is one of the largest wealth creation events in human history. It’s a fundamental tenet of capitalism: innovation comes from the margins.

Back in 1997: Tim Berners-Lee (the father of the World Wide Web), Marc Andreessen, and others were developing protocols and browsers. At the time, running a server cost hundreds of thousands of dollars. Why a content server should respond to an unknown consumer was not obvious at the time. Doing so is expensive and there is no clear financial incentive.

So they created a status message called "402" that the server could send to the consumer: "This content is available for a small fee." But there was no reasonable way to pay digitally. PayPal didn’t exist yet, and credit card flat fees were tens of cents, too expensive for 1-cent micropayments.

But the World Wide Web took off.

Google conceived a strange business model for the Internet: advertising. In traditional media, the main economic relationship exists between content producers and consumers. Borrowing from the economics of broadcasting, Google brings in a third party—advertisers—to pay for the relationship between producers and consumers.

A stroke of genius. Now producers can monetize readers’ attention without having to establish a prior relationship with consumers. Google sits in the middle of the money stream, between advertisers and content producers, taking whatever cut they want.

Thus, the need for micropayments is circumvented. Open source software emerged, the cloud revolution occurred, and hosting costs plummeted 100x. Google has become the biggest supporter of a free and open Internet. The more consumers search, the more Google makes. So they invested hundreds of billions of dollars to make the Internet fast, cheap, and ubiquitous.

Then the 2010s came and nothing happened. Interest rates are low, technology changes slowly, and the "walled gardens" are cultivating their health and gathering strength.

In 2022, ChatGPT is released, and the world will change again. Large language models (LLMs) can do more than deliver search results. They can generate and compile multiple results into a convenient summary, often without touching the content itself.

In the GPT-4 era, it became clear: Agents were the next step. LLMs are able to use computers like humans, but cheaper and more efficiently.

At that moment, the economics of the Internet changed.

From 1997 to 2024, the business model of the Internet has been "distraction." Humans will be distracted by advertisements when reading web pages, thus monetizing their fragmented attention. But the LLM/agent will not be distracted.

There is a wonderful irony here: advertising created a free and open Internet, which turned into a data set of 10 trillion Tokens, feeding LLM, and LLM eventually led to the demise of advertising.

Since the advent of GPT-4, Stack Overflow traffic has dropped by 75%, technology news traffic has dropped by 60%, and more. Technology consumers are early adopters, but this trend will sweep across all information on the web.

Checkout is not important in ChatGPT. The Internet is the square of civilization, and the original economic contract is now obsolete.

There are still small corners of the Internet that are resistant to Google’s reach, those walled gardens with truly differentiated content: Facebook, TikTok, LinkedIn. Relying on thousands of highly paid engineers working around the clock, they are resistant to automated crawlers.

But the Walled Garden's defenses have just collapsed. "Computer-use agents" can perfectly simulate the traffic of real human users. Snake oil salesmen will be prescribing all kinds of remedies over the next 10 years, and VCs will follow suit. But there is no antidote. The walls of the fortress have lost their meaning in front of the fighter planes.

What’s next? ——Open intelligent business

Checkout in ChatGPT is just the "AOL" of agency business. It's a carefully filtered directory, a walled garden for a better user experience. To sell through it, merchants need months of business development (BD), rigorous legal documentation, a concrete five-year plan, substantial revenue, a strong user base, and a story that satisfies shareholders.

And "open agent commerce" is today's HTTP. It’s a simple set of protocols for agents to pay for everything they need: data, cloud hosting, communications, and many more things we haven’t imagined yet.

The two leaders are x402 from Coinbase and mpp from Tempo and Stripe. 28 years after the “402” status code was invented, we finally have a workable implementation. Stablecoins on modern blockchains have fixed transaction costs of less than 1 cent, precisely solving the fixed fee problem that killed micropayments in 1997.

An agent that can only buy things from pre-approved merchants is just an employee with a restricted corporate card. An agent with an open protocol is an entrepreneur with a bank account.

There are no business negotiations, no whitelists, just simple permissionless standards.

These protocols only focus on two things:

• Agent: "How do I send money?"

• Merchant: "How can I be sure that this agent has paid?"

LLMs are very good at calling tools they have never seen before. Starting with Claude 4.5+ and Codex 5.2+ models, an agent can discover an API, read its schema, and use it correctly without pre-training.

The current discussion focuses on "Skills". You can think of them as natural language programs that can be combined like building blocks. A non-technical founder can just write a Slack message and have it executed as software:

"Go get pizza from the place closest to me with good reviews and track status every 10 minutes. When the driver is 5 minutes away, turn on the porch light. If delivery takes less than 30 minutes, tip the driver $5."

No coding required, no computer science degree required. The agent reads the intent, writes a native computer program on the fly, executes it, and then destroys it. Programming as a subject will become optional, and literacy in the native human language will suffice.

"Skills" are valid, but they are the product of a transitional period. They need someone to write, publish, security review, and update, and agents need to preload them. This is too much trouble.

The talk of “skills” masks a deeper transformation: agents being able to combine capabilities in unprecedented ways.

Pizza is just a small example. Let’s look at a real one: An agent managing a small business’s supply chain discovered that prices at a certain packaging supplier had increased by 15% due to tariffs. It automatically discovered three local alternatives, requested samples from each, negotiated volume pricing, and completed the switch. All this was completed before the boss’s morning practice ended.

No API partnerships, no procurement teams, no bidding processes. There is only one broker with a balance and an open protocol.

Discovery

Agents can pay, and agents can combine, but they are not yet very good at "finding" what they need.

The remaining problem is discovery. For the agent: "How do I find what I want to buy?" For the merchant: "How do I tell the agent about my service?"

AgentCash appears. It's a single balance leading to every API on the internet. When an agent hits a snag, it can call thousands of APIs, spend a few pennies, and keep working.

The key is that AgentCash bundles payments and merchant discovery. Merchants can register their servers at x402scan.com or mppscan.com and be instantly exposed to all 2000+ AgentCash agents.

In 1997, the World Wide Web had no business model and no one knew why servers would bother with strangers. Open protocols and a clever patch called "advertising" solved the problem, and civilization digitized. In 2026, this patch is becoming obsolete. Open protocols and a 28-year-old status code are about to replace it.

Welcome to the era of open agent business.