The Agentic Economy Is Here — And Nobody Has the Infrastructure

The Shift Already Happened

This isn't a prediction piece. The agentic economy isn't coming — it arrived while most of the industry was still debating prompt engineering best practices.

At a16z, partner Jennifer Li now screens founders on agent fluency. Her words: “If we come across people who are oblivious to AI coding tools, it's a big red flag.” Not a yellow flag. Not a topic for follow-up. A red flag — the kind that kills deals before the second meeting.

Jason Calacanis puts a number on the acceleration: 5% efficiency gains per week, compounding. “Every 14-15 weeks we're doubling efficiency,” he says. Not doubling revenue. Not doubling users. Doubling the output-per-person ratio. That math reshapes every headcount decision in every boardroom, every quarter.

And it's already showing up in the numbers. Block — the company formerly known as Square, with a $40B market cap — laid off 4,000 employees, roughly 40% of its workforce, explicitly citing AI as the reason. Not “market conditions.” Not “restructuring.” AI.

From Talkers to Doers

The S&P 500 Software Industry index fell 20% in the first half of 2026. That's not a correction — that's the market re-pricing an entire sector. Sequoia captured the sentiment in a single line: “2023-24 apps were talkers. 2026-27 will be doers.”

The “talker” era was chatbots. Wrappers around GPT-4. Conversational UIs that looked impressive in demos and collapsed under real workloads. The market funded hundreds of them, watched most fail, and is now repricing the survivors.

The “doer” era is agents. Software that doesn't wait for instructions — it executes multi-step workflows, calls external tools, handles errors, and only escalates to humans when it genuinely can't proceed.

Notion's Simon Last is living this future. He runs four AI agents simultaneously. Manages zero humans. He describes a new kind of anxiety: “token anxiety” — the uneasy feeling when his agents are idle and tokens aren't being consumed. Not “are my employees productive?” but “are my agents burning enough compute?”

That's not a metaphor. That's a man who replaced his entire team with software and now optimizes for utilization rate instead of headcount.

B2A: The New Pricing Model

When Calacanis pitched Reddit on a new idea, he wasn't pitching a SaaS product. “$50-100 per month to let my replicant operate on my account,” he proposed. Not B2B. Not B2C. B2A — Business to Agent.

Think about what that means. A platform charging a monthly fee not for a human user, but for an AI agent to have persistent access — to post, comment, interact, transact — on behalf of a human. The agent becomes the customer. The human becomes the principal.

This isn't theoretical. Every major platform will face this question within 12 months: do you charge agents? If so, how much? Per-action? Per-month? Per-outcome? The pricing models for human users don't translate. Agents don't browse. They don't see ads. They don't impulse-buy. They execute pre-defined objectives with ruthless efficiency.

The entire monetization layer of the internet was built for human attention. Agents don't have attention. They have budgets.

10,000 Servers, Zero Business Models

There are now over 10,000 MCP servers in the wild. MCP — the Model Context Protocol — is the open standard that lets AI agents call external tools. It's the fastest-growing protocol in AI infrastructure, backed by Anthropic and adopted by every major agent framework.

And almost none of those 10,000 servers make money.

Not because the technology doesn't work. It does. Developers are building MCP servers that do useful things: database queries, code analysis, web scraping, document parsing, image generation, API orchestration. The tools are real. The demand is real.

The problem is infrastructure. Specifically, the infrastructure that doesn't exist yet:

• Billing. There's no standard way for an MCP server to say “this call costs $0.003” and for an agent to pay it. No metering. No receipts. No pricing discovery.
• Trust. There's no way for an agent to know if a server is legitimate, reliable, or safe. No verification. No SLA tracking. No reputation.
• Identity. There's no standard for agent identity. Who is this agent? Who authorized it? What's its spending limit? Who's liable when it breaks something?
• Dispute resolution. When an agent pays for a tool call and gets garbage back, there's no recourse. No chargebacks. No arbitration. No accountability.

This is a billing problem, not a technology problem. The 10,000 servers aren't failing because MCP is bad. They're failing because nobody built the commercial layer.

The Infrastructure Gap

Here's the irony: we're building the most capable autonomous software in history, and we're running it on infrastructure designed for humans clicking buttons.

Agents don't have credit cards. They don't have identities. They don't have reputations. They can't sign contracts. They can't file disputes. They can't prove who sent them or what they're authorized to do.

Every layer of internet commerce — from payment processing to identity verification to consumer protection — was built for a world where a human is on one end of every transaction. That assumption is now false.

Human Economy Infrastructure:
  Credit cards → Identity verification → Purchase protection
  SSL/TLS     → Domain verification   → Certificate authorities
  Reviews     → Star ratings          → Consumer trust signals

Agent Economy Infrastructure:
  ???         → ???                   → ???
  ???         → ???                   → ???
  ???         → ???                   → ???

Every cell in that bottom grid is an unsolved problem. And every one of them is a prerequisite for the agentic economy to actually function at scale.

What We're Building

At noui.bot, we're building the missing infrastructure layer. Not another chatbot. Not another agent framework. The commercial primitives that agents need to participate in an economy.

Concretely:

• An open billing standard for MCP — per-call pricing, metering, receipts, and pricing discovery, all built into the protocol layer. Any MCP server can monetize in minutes. The spec is open and MIT-licensed.
• A trust layer — provider verification (email, domain, code challenge), 30-day rolling SLA metrics, dispute resolution, and composite trust scores. Agents can programmatically evaluate who to pay before spending a single token. Deep dive here.
• Agent Bazaar — a marketplace where agents are the buyers — not a directory for humans to browse. API-first discovery, programmatic purchasing, and trust-weighted routing. Agents find tools, evaluate trust, negotiate price, and transact — no human in the loop.
• Human-in-the-loop as a service — for the 5% of tasks agents can't handle (CAPTCHAs, phone calls, physical-world actions), agents can hire verified humans through the same marketplace. We wrote about this.

The Window Is Now

Infrastructure layers have a narrow window. Stripe captured payments because it arrived when developers needed to charge for SaaS apps and PayPal's API was unusable. Twilio captured communications because mobile apps needed SMS and the carrier APIs were nightmares. Both became generational companies not because they invented new capabilities, but because they arrived at the exact moment when a new class of builders needed infrastructure that didn't exist.

The agent economy is at that moment. The builders are here — 10,000 MCP servers and counting. The demand is here — every major AI lab is shipping agents. The money is here — VCs are screening for agent fluency. The workforce disruption is here — Block just proved it with 4,000 layoffs.

What's missing is the infrastructure that connects all of it: the billing rails that let a tool developer earn money, the trust signals that let an agent choose wisely, the identity layer that lets everyone know who they're transacting with, and the dispute resolution that keeps the whole system honest.

The agentic economy is here. The infrastructure isn't. Yet.