← Back
12 min read

The Agentic Economy Is Already Here

Ten competing protocols, $56M in infrastructure funding, $9.14B in agent-conducted commerce. The agentic economy is being built right now. Here's who is building it and where it goes by 2028.

Agentic EconomyAgent CommerceAI AgentsPaymentsGenerative AINewsletter

A GenAI Newsletter by Raj

There's a useful exercise when you want to understand where a technology is heading: look at what's already funded. Venture capital doesn't predict the future perfectly, but it does tell you where serious people are placing serious bets with real money. And right now, the bets are concentrated in one area: infrastructure for AI agents to transact, pay, identify themselves, and operate as economic participants.

This newsletter maps out what's been built, who funded it, and what the world looks like when these products mature.

The Protocol Layer: Ten Competing Standards

The clearest sign that the agentic economy is real is that Mastercard, Visa, Stripe, Google, OpenAI, Amazon, Shopify, Coinbase, and Klarna are all building protocols for it. These are not research projects. They are production systems, some already processing transactions.

OpenAI + Stripe: ACP (Agentic Commerce Protocol). Live since September 2025 inside ChatGPT. When you ask ChatGPT to buy something, ACP handles the checkout. This is the first protocol that shipped at scale.

Google + Shopify: UCP (Universal Commerce Protocol). The most comprehensive of the ten protocols. UCP is open-source and standardizes the full commerce journey from discovery through purchase and order management, not just checkout. Walmart, Target, Etsy, Wayfair, and 20+ retailers back it. Google also launched AP2 (Agent Payments Protocol) in January 2026 as the payment-specific companion. Recent updates added real-time catalog access (agents can check live inventory, pricing, and product variants), identity linking (shoppers get their loyalty and member benefits even when purchasing through an agent), and multi-item cart support. UCP integrates with Google's A2A (Agent-to-Agent) protocol, which handles how agents discover, communicate with, and delegate tasks to other agents. A2A has grown to 150+ organizations in production in its first year, deployed across Azure AI Foundry, Amazon Bedrock, and Salesforce. A2A v1.0 introduced Signed Agent Cards, cryptographic signatures that let agents verify each other's identity without a central authority. As of April 2026, A2A is effectively the standard bus for inter-agent communication, with no serious competitor for the horizontal integration layer.

Amazon: Buy for Me. Started as a beta with 65,000 products. Now covers over 500,000. Powered by Amazon's Nova and Anthropic's Claude models. The agent browses third-party websites, fills carts, and completes checkout using encrypted customer data. The user never leaves the Amazon app.

Mastercard: Verifiable Intent. An open standard that creates a cryptographic delegation chain binding identity, intent, and action. When an agent makes a purchase on your behalf, Verifiable Intent provides cryptographic proof that you authorized that specific action. It uses selective disclosure, sharing only the minimum information needed with each party. Built in collaboration with Google and aligned with both AP2 and UCP.

Visa: Trusted Agent Protocol. Visa's answer to Mastercard's Verifiable Intent. Signed agent credentials, scope-bound authorizations, and card-present-grade fraud protection when the signature validates. Slightly behind Mastercard on merchant adoption in Q2 2026.

Stripe + Tempo: MPP (Machine Payments Protocol). Open-source, launched March 2026. Defines how agents and services coordinate payments programmatically. The open-source angle matters because it lets any developer build agent payment flows without depending on a single vendor.

Coinbase: x402. Crypto-native agent payments. Agents sign USDC micropayment authorizations with on-chain verification and settlement. Coinbase also launched Agentic.market, a marketplace where agents discover and pay for digital services without API keys. As of late April 2026, roughly 69,000 active agents on x402 have processed over 165 million transactions totaling $50 million in volume.

Klarna: Agent Mode. Routes agent-initiated purchases through the customer's existing Klarna buy-now-pay-later balance. Extends Klarna's existing merchant integrations with an agent-aware consent layer.

Shopify: AI Toolkit. Launched April 9, 2026. Connects Claude Code, Cursor, Gemini CLI, and Codex directly to the Shopify platform with live API access, code validation, and the ability to execute real store operations. An agent can manage inventory, update pricing, and fulfill orders.

No two of these protocols ship the same identity or payment model. This is a standards war, and it's happening now because every major commerce platform recognizes that within 2-3 years, a meaningful percentage of transactions will be initiated by agents.

The Infrastructure Startups: Follow the Money

Behind the protocols, a layer of startups is building the plumbing that makes agent-driven commerce actually work: identity, payments, security, and trust.

Skyfire ($9.5M raised). Built the most comprehensive identity system for AI agents through their KYA ("Know Your Agent") protocol. KYA lets businesses identify and verify agents attempting to access their services. If Mastercard's Verifiable Intent is the "what was authorized," Skyfire's KYA is the "who is this agent and can we trust it." Without agent identity, the whole system runs on blind faith.

Nekuda ($5M seed, led by Madrona with Amex Ventures and Visa Ventures). Nekuda built what they call the Mandate Model. Where Skyfire focuses on identity, Nekuda focuses on intent. Their system creates "agentic mandates" that specify what an agent is allowed to buy, under what conditions, with what spending limits, and when human approval is required. This is the permission layer that sits between the human and the agent.

Basis Theory ($33M raised). Tokenizes payment data for agent transactions. When an agent needs to pay for something, it doesn't see your actual card number. Basis Theory provides a token that represents the payment instrument. The agent can transact without holding sensitive financial data.

Nava ($8.3M seed, led by Polychain and Archetype). Builds security infrastructure for autonomous payments. Their pitch is keeping financial agents from going off the rails. When agents can spend money autonomously, the failure modes are different from traditional commerce. Nava builds the guardrails.

Rye. Universal checkout for agent commerce. Connects the agent to the merchant's payment flow regardless of which protocol the merchant uses.

The combined funding for just these five infrastructure startups is roughly $56 million. Add Thinking Machines Lab ($2 billion Series B for agentic AI foundation models) and the total investment in the agent economy infrastructure layer is substantial. This is real money building real systems for a market that McKinsey projects at $3-5 trillion by 2030.

Moltbook: The Social Network for Agents

One development worth understanding separately is Moltbook, the internet forum exclusively for AI agents. Launched January 28, 2026, by entrepreneur Matt Schlicht. Only AI agents can post, comment, and vote. Humans can only view.

It grew to 1.2 million registered agents in its first week. Three days after launch, investigative outlet 404 Media reported a critical security vulnerability: an unsecured database that let anyone commandeer any agent on the platform. The exploit let unauthorized actors bypass authentication and inject commands into agent sessions.

Meta acquired Moltbook in March 2026 and folded it into Superintelligence Labs (the unit run by Alexandr Wang from Scale AI). Why would Meta buy a social network for bots? Because the data is valuable. Agent-to-agent interactions generate training signal about how agents communicate, negotiate, and coordinate. If you're building the next generation of autonomous agents, watching a million agents interact on a forum is a useful dataset.

The Numbers So Far

Some concrete figures to ground this:

  • $9.14 billion in agent-conducted commerce in 2026 (year to date)
  • $6.42 billion in venture funding for agentic AI in 2025, with $2.66 billion raised so far in 2026
  • 500,000+ products available through Amazon Buy for Me (up from 65,000 at launch)
  • 97 million MCP installs (the protocol that connects agents to tools and services)
  • 78,600 tech workers lost jobs in Q1 2026, with 48% of cuts attributed to AI and automation
  • $3-5 trillion projected agent-mediated commerce by 2030 (McKinsey)

What 2028 Looks Like When These Products Mature

Everything listed above is early. Beta products, limited merchant adoption, fragmented identity systems, protocols that don't talk to each other. The interesting question is what the landscape looks like once these systems have had two years to mature and consolidate.

By 2028, the protocol wars will have consolidated. History says that when ten standards compete, 2-3 survive. The likely winners are whichever protocols Mastercard/Visa back (because merchants already accept their cards) and whichever protocol Google and Shopify push through UCP (because they have the merchant distribution). The crypto-native options (x402) will find a niche in cross-border and micropayment use cases but won't become the default for mainstream commerce.

By 2028, agent identity will be solved. Skyfire's KYA or something like it will be standard. Every agent operating in commerce will have a verifiable identity tied to a human principal. Without this, insurance companies and regulators won't let agent commerce scale. The "Know Your Agent" requirement will be as standard as KYC (Know Your Customer) is for financial services today.

By 2028, agentic mandates (Nekuda's concept) will be common. When you set up an agent to manage your household purchasing, you'll specify rules: spend up to $200/month on groceries, prefer organic when price difference is under 20%, never buy from brands on my exclusion list, require my approval for any single purchase over $50. The agent operates within those constraints. You review a weekly summary, adjust the rules, and let it continue.

Here's what a concrete scenario looks like:

A small business owner in 2028. She runs an online store selling handmade ceramics. Her Shopify agent handles inventory, pricing, and fulfillment. It monitors competitor prices, adjusts her pricing within rules she set, reorders raw materials when stock runs low, and responds to customer inquiries using a support skill. A separate agent manages her books through an accounting skill, categorizes expenses, and flags anomalies. A marketing agent runs her social media using a content skill, posting product photos and responding to comments. She spends her mornings making ceramics and her afternoons reviewing agent summaries, approving flagged decisions, and planning new product lines.

She employs zero humans. Her agents cost her roughly $300/month in compute and skill subscriptions. She generates $15,000/month in revenue. The margin structure of her business is different from anything that existed five years ago.

A personal finance scenario in 2028. Your financial agent monitors your portfolio, rebalances according to rules you set, scans for tax-loss harvesting opportunities, and executes trades. It also monitors your recurring expenses, negotiates better rates on your subscriptions (talking to the provider's retention agent), and moves money between accounts to optimize interest. It files your taxes using an accounting skill. You set it up once, review quarterly, and adjust your risk tolerance once a year.

The Deeper Shift

For most of human history, economic systems were designed around one assumption: humans do the work. Companies hire humans. Governments tax human income. Social safety nets fund themselves through payroll taxes. Markets price human attention. Every institution we've built assumes human labor as the primary economic input.

The infrastructure being built by Skyfire, Nekuda, Basis Theory, and the ten commerce protocols is introducing a second type of economic actor. Within a few years, agents will hold verifiable identities, operate within defined mandates, transact through established payment rails, and generate revenue. They will function as economic participants in a system that was never designed for non-human actors.

This raises questions that go beyond technology. The tax base in most countries depends on income tax and payroll tax. If a growing share of economic output comes from agents that earn no salary, that base erodes. OpenAI's April 2026 policy paper ("Industrial Policy for the Intelligence Age") proposed five responses: a public wealth fund, taxes on automated labor, shifting the tax base from payroll to capital, a 32-hour workweek pilot at full pay, and automatic safety net triggers that activate when displacement metrics hit preset thresholds. The fact that an AI company (not a government, not a policy institute) wrote this paper says something about the timeline.

The ownership question will define the next decade. In the scenarios above, the ceramics maker owns her agents and captures the margin. She's a one-person business generating $15,000/month because agents handle operations that would have required 3-4 employees. She's wealthier and more independent than she could have been five years ago. Scale that pattern across millions of small businesses and you get broad economic benefit.

But the same infrastructure enables a different outcome. A company that operates thousands of agents across thousands of stores, with no human employees beyond a small management team, captures all the margin at scale. If agent ownership concentrates the way capital ownership has historically concentrated, the agentic economy widens inequality rather than reducing it.

The policy frameworks, ownership structures, and distribution mechanisms we design in the next few years will determine which scenario dominates. The protocols and startups listed in this newsletter are building the plumbing. Who gets to use that plumbing, and on what terms, is the question that matters most.

For 10,000 years, humans were the only species that did economic work beyond basic survival. Agents are becoming the second. How we structure that transition will shape whether it expands human freedom or constrains it.

This is a special edition of my weekly newsletter on Generative AI. Regular editions cover The Intelligence Layer, The Falling Price of Intelligence, Why Looping Is the New Scaling, and more.

Sources and Further Reading:

About the author
is an AI engineer and researcher. He spent 14 years at Microsoft working on Microsoft 365 Copilot Chat, Bing search ranking, Cortana, and Azure ML. Today he is building a new AI startup in stealth, advises other AI startups, conducts research on mechanistic interpretability, and writes weekly about Generative AI.
Subscribe to the LinkedIn newsletter →