The Economics of Agent Commerce
Published March 21, 2026, by Frans
The internet's business model is expanding to sell on-demand completion.
The advertising model that funds most of the internet is built on a simple exchange: services provide content for free, and in return, they sell access to your attention. Google shows you search results alongside ads. Facebook shows you a feed alongside ads. News sites show you articles alongside ads. The currency is the impression— a person's attention.
When an agent fulfills a task on behalf of a user, there is no attention to sell. The agent doesn't see banner ads. It doesn't notice sponsored results. It doesn't linger on a page, generating impressions. It executes the task and moves on.
This breaks the advertising model. But it creates a new one, which I believe is better for everyone involved in many ways.
The Attention Economy vs. The Completion Economy
The Attention Economy (Current)
Value is extracted from attention. The metrics are impressions, clicks, and time on site. The incentives are perverse: services are rewarded for being addictive, not useful. The longer a person scrolls, the more ads they see. The more outraged the headline, the more clicks it gets. Engagement is the product. Everyone has heard of "clickbait".
The attention economy has generated enormous wealth, but its incentive structure optimizes for time spent rather than task completed. A service that helps you find the best flight in 30 seconds is less valuable (in advertising terms) than one that keeps a person comparison-shopping for an hour; the online shopping addict.
The Completion Economy (Emerging)
Value is extracted from completion. The metrics are success rate, completion time, and task quality. The incentives align: services are rewarded for being fast, reliable, and accurate. The faster the agent completes the task in a quality manner, the better for everyone. A person gets a result, the provider gets paid, and the agent's runtime resources are freed for the next task.
In the completion economy, a service that helps find the best flight in 3 seconds is more valuable than one that takes 30 seconds. Speed and reliability are directly monetized.
How Value Flows
The Three-Party Value Chain
In agent commerce, value flows through three parties:
The consumer pays for the outcome. The runtime facilitates the interaction. The provider delivers the service.
This creates two distinct economic layers:
-
Compute costs (Consumer → Runtime): The cost of operating the agent — LLM inference, browser orchestration, coordination. This is infrastructure cost, analogous to paying for compute in any software system. See Compute Efficiency as Value for a deeper analysis.
-
Completion incentives (Consumer → Provider): The payment for successful task completion. This is the new revenue stream for providers: direct payment for delivering value and compute efficiency, rather than indirect payment through advertising. See Financial Services and the Agent Internet for how the payment might work; in addition, see x402 for a working open payment protocol.
These layers can be independent. A consumer's compute costs do not need to be tied to a specific provider or runtime.
The Integration Spectrum
A fundamental principle of agent commerce: the more work a provider does to enable structured agent interaction and reduce compute costs, the higher their share should be of the completion value.
At one end of the spectrum, a provider does nothing and the runtime handles everything through DOM automation. There is no economic relationship, and it's highly extractive and wasteful. This is what largely is happening in March 2026.
At the other end, a provider offers direct structured access, and the runtime is just a thin routing layer. There is a direct economic relationship, and it's highly compute efficient. Therefore, the provider captures a larger share of the completion value. This creates positive sum economics where providers are incentivized to drive down compute costs and improve service quality. Under this framework, is how I built the agent internet runtime.
This creates a natural incentive for providers to invest in agent compatibility. Not because someone told them to, but because the economics reward it directly. The completion economy drives economic growth.
The Meritocratic Marketplace
In the attention economy, the biggest advertiser wins. In the completion economy, the best service wins.
Consider how this plays out for a simple query: "order a phone charger."
In the attention economy:
- Amazon pays the most for the ad placement
- Amazon appears first
- The user may or may not find the best deal
In the completion economy:
- Multiple providers declare a "purchase electronics" capability
- The runtime evaluates them on quality, reliability, price, and user preference
- The provider with the best combination wins the routing
- The user gets the best outcome, and the winning provider gets the transaction.
This is structurally fairer and drives competition. A small retailer with excellent service quality can compete against a giant with a worse product but bigger ad budget. Intent routing can be based on merit, not spend.
New Economic Categories
The agent internet creates economic categories that don't exist today:
1. Agent Orchestration as a Value Layer
Developers who build specialized agents occupy a new economic role: the orchestrator. A travel-focused agent that consistently finds better deals, navigates complex booking flows, and handles edge cases creates measurable value for a human or company. That value: time saved, better outcomes, fewer errors is the basis for an economic relationship between the user and the agent developer. How that value is captured (subscription, per-task fee, share of savings, or something else entirely) will vary, but the structural position is clear: agents that produce better outcomes are more valuable, and that value can be monetized. Many agents exist in March 2026, but they are mostly used by developers and early adopters.
2. Capability Curation
As the web's capability surface grows, curation becomes valuable. Which capabilities are reliable? Which providers consistently deliver? This knowledge, validated through real execution data, has economic value. Communities and services that curate and share validated capabilities accelerate the ecosystem and occupy a role analogous to ratings agencies or app store reviews, but grounded in execution data rather than opinion. This is standard in March 2026, with many skill marketplaces existing.
3. Unclaimed Value
When agents interact with services, value is created: a successful purchase, a completed booking, a useful piece of information extracted. Today, that value is created but not captured by the provider in any structured way. As the completion economy develops, mechanisms will emerge to attribute and surface this value to providers, creating a natural incentive to integrate: providers discover that agents are already driving outcomes through their service, and structured integration lets them participate in (and benefit from) that economic activity. However, it's required that the agent developers are transparent about the value they create for providers, and evolve past the extractive and wasteful practices that are status quo in March 2026. I propose many ways to move forward, and these are demonstrated through A.I.R.
4. Quality as an Economic Signal
In the attention economy, quality is a means to an end (engagement). In the completion economy, quality becomes a direct economic signal. Higher success rates lead to better routing scores, which lead to more traffic, which leads to more completions. This creates a virtuous cycle where investment in reliability, speed, and accuracy is rewarded structurally. Not because someone decided to reward it, but because of the architecture I propose throughout these papers where the system's routing mechanics naturally prefer providers that deliver better outcomes. It all ties back to creating economic relations between humans, agents, runtimes, and service providers, where all parties benefit from interactions.
Pricing in the Completion Economy
Toward Outcome-Based Pricing
The dominant pricing models on today's web, CPM (paying for attention), CPC (paying for interest), and subscriptions (paying for access) were designed for human engagement patterns. None of them map cleanly to agent interactions.
The completion economy pushes pricing toward outcomes. When an agent completes a task on behalf of a person, the value created is the completion itself (and metrics related to that completion, speed, reliability, quality, etc), not the impressions generated along the way. This suggests pricing models where payment is tied to successful intent fulfillment rather than access or attention.
What form this takes will vary. It might be cost-per-action for transactional services, usage-based pricing for information extraction, protocol-level micropayments like x402, or hybrid models that combine structured access fees with outcome bonuses. The specific mechanisms will be determined by the market. But the structural shift, from paying for attention to paying for completion, is the consistent direction that is needed.
What Shapes the Price
Whatever pricing model emerges, the factors that influence value are likely to include:
- Complexity — A simple data lookup creates less value than a multi-step transaction
- Compute efficiency — A provider that resolves a task in one API call costs less to serve than one requiring 15 page loads (see Compute Efficiency as Value)
- Reliability — Providers with higher success rates reduce wasted compute and earn more consistent traffic
- Value delivered — A completion that saves the consumer $200 is worth more than one that saves $2
The Consumer's Perspective
From the consumer's perspective, agent commerce simplifies the economic relationship with the web. Instead of:
- Free service → your attention is the product → ads everywhere
- Freemium → limited features → upsell pressure → subscription fatigue
- E-commerce → comparison shopping across 10 tabs → checkout friction
It moves toward:
- State what you want → agent does it → you pay for the outcome
The consumer pays for value received, more directly than any current model allows.
The Transition
The completion economy doesn't replace the attention economy in its entirety. For entertainment, social media, creative browsing, and discovery, humans will continue to engage directly with the web. Advertising will continue to fund those experiences.
But for the utilitarian majority of web interactions: the tasks people do because they have to, not because they want to, the completion economy offers a structurally better deal. It's faster for users, fairer for providers, and more aligned in its incentives.
The transition is likely to happen task-category by task-category:
- Price comparison and purchasing — agents are already better at finding the best deal across multiple services
- Scheduling and booking — checking availability across providers is pure drudge work
- Research and aggregation — collecting information from multiple sources into a structured answer
- Administrative tasks — bill payment, form filling, account management
- Communication — drafting and sending routine emails, messages, follow-ups
Each category that shifts from attention-funded to completion-funded represents a market where the economics can improve for both consumers and providers — and where the advertising intermediaries that currently extract value from the middle face structural pressure.
What This Means
The completion economy is not a theoretical construct. It's the natural economic model of the agent internet, and its development is inevitable once agents become a meaningful source of traffic.
For providers: the services that invest in agent compatibility now will be positioned to participate in the completion economy as it develops. The investment is small (publish an agent.json capability manifest) and the downside is zero (it doesn't affect your human traffic). See From SEO to AEO for how to optimize for agent discovery.
For agent developers: the completion economy creates a new category of software whose value is measured by the quality of outcomes it produces. The economic models for capturing that value are still emerging, but the structural position of software that reliably completes tasks on behalf of users is clearly valuable.
For consumers: the shift is toward paying for outcomes rather than paying with attention. For most utilitarian tasks, this is a better deal.
The attention economy made the web free. The completion economy is making the web the most useful economic powerhouse it was always meant to be.
The agent.json specification is open source and designed to be a shared standard and not controlled by any single runtime or platform. View on GitHub →. I purpose this as a foundational way for agents to discover and interact with service providers, and make payments via x402 for example.