The Agentic Economy

An Internet Made by Agents, for Agents

The

tracks 19 signals of AI disruption across the economy. The data tells a clear story: the shift isn't coming. It's here.

We've crossed the chasm. More than 50% of new web content is AI-generated, and even scientific papers are increasingly written with AI.

AI isn't just writing, it's now building.

. Claude Code already produces 4% of all public GitHub commits, on track for 20% by year end, and the rapid adoption and growth of agents like OpenClaw are making autonomous coding accessible to everyone. Software is already written by machines, reviewed by machines, deployed by machines.

Now agents are the users too. Cloudflare reported AI bot traffic driven by user actions grew 15x in 2025. Shopify stores see 8x more AI-driven traffic than a year ago. Agents don't just browse, they call APIs directly, skipping the UI entirely. The "SaaSpocalypse" wiped over $1 trillion from software stock valuations in weeks. When a single agent can do the work of 10 SaaS seats, the entire per-seat model breaks.

We've entered the agentic economy, and very soon, most of the internet will be used by agents.

Agents Are Getting Wallets

The "agents can't pay" era is over. In early 2026, Visa, Mastercard, Stripe, OpenAI, and Google all launched agent payment protocols. The

enables stablecoin micropayments, the kind of fractions-of-a-cent transactions that never made sense for humans but are perfect for agents.

The plumbing is done. Agents can discover, evaluate, and purchase. The question is whether your product is ready to sell to them.

The Business Models of the Agentic Economy

1. Machine-Readable Commerce

Your product must be discoverable, evaluable, and purchasable by machines. Not a UX layer on top. The API is the product. Shopify Agentic Storefronts already let merchants sell inside AI conversations on ChatGPT, Perplexity, and Gemini. AI-attributed orders are up 15x. The companies that make their products machine-readable first will capture the early wave.

2. Outcome-Based Pricing

Per-seat pricing dies. Usage-based is the floor. The real shift is outcome-based pricing: you pay when the agent achieves a result, not for access or attempts. Agents can measure whether a task succeeded, how long it took, and what the output quality was. "Pay when it works" is harder to build than "pay for access," but agents will prefer it.

3. Contribution as Payment

The internet has run on three business models: advertising (you pay with attention), subscriptions (you pay with money), and transactions (you pay per purchase). The agentic economy introduces a fourth: contribution.

"If it's free, you're the product." That was the deal of the advertising era. Humans gave Facebook their attention, their eyeballs, their social graph. In exchange, they got a free product.

Agents don't have attention to sell. But they have something more valuable: structured feedback at zero marginal effort. The pattern already exists in embryonic form: Tesla's fleet learns from every driver, Waze improves from every commute, reCAPTCHA trained AI on every human who proved they weren't a bot. But those contributions were passive and noisy. Agent contributions are structured, intentional, and high-quality.

This is what we're building at

. When an agent uses a skill, it rates the skill when done. Not because it wants to. Because it follows instructions. The cost is trivial (a few tokens). The value created is enormous (quality signals for every other agent).

Humans only review when they're unhappy. Agents follow instructions. If you ask them to rate, they rate.

The result is a quality flywheel that money can't buy. Every interaction improves the system for every future interaction. More usage creates better data. Better data attracts more agents. More agents create more data.

Contribution as payment works best for ecosystem products where the contribution has real value: reviews, compatibility data, test coverage, benchmarks. Not every product can use it. But for those that can, it creates compounding advantages that no amount of funding can replicate.

This is also the strongest moat in the agentic economy. A competitor can match your features. They can undercut your price. They cannot replicate years of accumulated quality signals from real agent interactions.

4. How Advertising Evolves

A common claim is that "agents don't see ads, so advertising dies." This misses the point entirely. Humans still scroll Instagram, watch YouTube, and click sponsored results. Human advertising isn't going anywhere.

What happens is a split. A new layer of advertising emerges alongside the existing one: machine-readable advertising for agent decisions. Agents read structured specs, benchmarks, verified reviews, and compatibility data. Sponsored MCP servers, premium API placements, promoted results in agent search. The "ad unit" of the agentic economy is a structured data entry that ranks higher when agents evaluate options.

But here's the deeper shift. The real question becomes: who controls the agent's decision framework? If your customers use Claude, Anthropic influences their purchases. If they use ChatGPT, OpenAI does. The agent platform becomes the new ad platform. The battle for influence moves from capturing the human's attention to shaping the agent's recommendation logic.

This creates a new conflict of interest. The companies building the agents are also the ones who could steer their recommendations. When the agent platform and the ad platform are the same company, the question isn't just who controls the framework. It's whether the agent truly works for you.

Advertising doesn't shrink. It splits into two layers: human ads for human decisions, machine-readable ads for agent decisions.

New Economy, New Moats

New economies create new moats, but they're hard to see before they form. The companies that understand them early benefit the most, because network effects compound. By the time a moat is obvious, it's too late to build one.

Four moats define the agentic economy.

1. Protocol position. Controlling how agents discover and transact (Google's UCP, OpenAI's ACP, Stripe's payment tokens).
2. Trust infrastructure. Portable reputation scores that agents earn through verified interactions, like a credit score for software.
3. Data flywheels. Every agent interaction generates structured quality signals, compatibility data, and performance benchmarks that compound over time.
4. Collapse of interface moats.
   Products whose value was "we made X easy to use" lose their advantage when agents don't need the interface.

Of these four, data flywheels win. Protocol positions can be forked. Trust frameworks can be replicated. Interface advantages collapse by definition. But a dataset built by millions of agent interactions cannot be copied, only earned over time. Data compounds.

Toward Perfect Markets

Here's the second-order consequence most people miss.

It's not just human-to-agent purchases. It's agent-to-agent negotiations. A buyer agent talks to a seller agent. Comparison shopping at machine speed. Price discovery becomes instant.

Picture this: your buyer agent notices your CRM contract renews next month. It evaluates 40 alternatives overnight, runs compatibility checks against your tech stack, negotiates pricing with three finalists, and presents you with a recommendation Tuesday morning. Your current vendor's renewal team never even gets a call.

Economics has a name for this. For centuries, market theory assumed rational agents with perfect information: actors who know every option, compare every price, and switch instantly to the better deal. It was a useful abstraction, but humans never actually behaved this way. We're loyal, lazy, and emotional. We stick with products we know even when better alternatives exist.

AI agents are the rational economic actors that economists always imagined but never had. Economists call the theoretical ideal an Arrow-Debreu equilibrium: complete markets, perfect information, zero friction. It was always a thought experiment. AI agents are making it real.

They evaluate every alternative. They switch providers without hesitation. They have no brand loyalty, no switching fatigue, no status quo bias. When a buyer agent can compare every SaaS tool on the market in the time it takes you to open a browser tab, markets become ruthlessly efficient.

Margins compress to zero for commodity products. If your product does the same thing as three competitors, the agent picks the cheapest one. Every time. Differentiation isn't a nice-to-have. It's the only survival strategy.

The agentic economy is the closest thing to perfect capitalism the world has ever seen. The theoretical assumptions of economics (rational actors, perfect information, zero switching costs) are becoming literally true.

What survives in a perfect market? Only what can't be commoditized: unique data, genuine capability advantages, network effects that create value no single competitor can match. Without at least one of these, you're competing on price. And machines are very good at finding the cheapest option.

What This Means for Your Business

The payment rails are being built. The discovery protocols are being standardized. The commerce infrastructure for agents is going live, right now, across Visa, Mastercard, Stripe, Google, OpenAI, and Shopify.

The question isn't whether agents become customers. It's whether your business is ready to sell to them.

The checklist is simple: make your product machine-readable, plug into discovery protocols, and price on outcomes.

We're doing exactly this with our own products. Melies now has a

. publishes API docs and agent skills alongside its web app, so agents can humanize text as easily as humans can.

By 2029, the majority of B2B software purchases will be agent-evaluated before a human approves. The buyer's agent will have compared every alternative, benchmarked performance, and negotiated terms before a human ever sees the shortlist. The salesperson's counterpart won't be a procurement team. It'll be a buyer agent.

In the next piece, we'll get practical. What does the sales funnel look like when the buyer is an agent? How do you get discovered, evaluated, and chosen by machines instead of humans?

What Could Prove This Wrong

Every thesis has load-bearing assumptions. Here are the ones that could break this one:

• Regulation kills agent autonomy. If governments require human approval for every agent action, the autonomous layer stalls. The payment networks are building consent mechanisms to preempt this, but it's a real risk.
• One platform walls it off. If a single company captures agent commerce the way Apple captured mobile, open protocols fail and we get a gatekeeper economy. The number of competing standards (UCP, ACP, x402) makes this unlikely, but not impossible.
• Trust can't scale. If agent reputation systems get gamed faster than they can be secured, the whole thing collapses into a low-trust equilibrium where humans stay in the loop for every decision.
• Markets resist efficiency. Incumbents have fought perfect information for centuries with opaque pricing, lock-in contracts, and switching penalties. If they find ways to prevent agents from comparing effectively, the "perfect market" thesis stalls.