21,000 Agents, Zero Code: Agent Building Goes Mainstream

Notion shipped Custom Agents on February 24. Within the first week, early testers built 21,000 of them—autonomous digital workers that triage tickets, draft reports, and resolve support issues around the clock. Not one required a line of code. The same day, Intuit announced its 100 million users would soon build custom AI agents through a partnership with Anthropic's Claude Agent SDK, also without writing code. The barrier to building AI agents didn't lower gradually. It collapsed in a single news cycle.

Gartner predicted 40% of enterprise applications would embed AI agents by the end of 2026, up from under 5% in 2025. That forecast assumed a gradual rollout. Instead, February 24 delivered a concentrated burst: Notion shipped autonomous Custom Agents, Intuit partnered with Anthropic for no-code agent creation at massive scale, Anthropic launched 13 enterprise plugins spanning finance, HR, and legal, Cursor shipped background agents that iterate autonomously on their own virtual machines, and New Relic launched a no-code agentic observability platform. The common thread: none of these require developers. Office managers, accountants, and support leads can now build agents that run 24/7.

This is the moment AgentPMT's drag-and-drop workflow and skills builder was designed for—with a critical difference these platforms haven't addressed: portability. Notion's agents only work inside Notion. Intuit's agents only work inside QuickBooks. AgentPMT workflows and skills run on Claude, GPT, Gemini, Cursor, VS Code, or any MCP-compatible agent. One build, every platform. When the scale of agent creation explodes—and 21,000 agents in a single week says it already has—the infrastructure that connects them matters more than any single platform's agent layer.

What Happened on February 24

The announcements landed within hours of each other. Notion 3.3 introduced Custom Agents: autonomous AI teammates that handle recurring work across Notion, Slack, Mail, Calendar, Figma, Linear, and HubSpot via MCP (Model Context Protocol) integrations. These aren't chatbots waiting for prompts. They operate on triggers and schedules, running continuously without manual input.

The adoption numbers tell the story. Notion co-founder Akshay Kothari noted the company now runs more agents than employees internally. Early testers outside Notion built over 21,000 custom agents before the public launch. Remote, a global HR platform, reported saving 20 hours per week with a single agent that triages support tickets with over 95% accuracy and resolves more than 25% of tickets autonomously—completely replacing their IT help desk. Ramp deployed over 300 agents, including a "Product Oracle" that fields dozens of daily questions about the product roadmap and feature set.

The same day, Intuit and Anthropic announced a partnership embedding the Claude Agent SDK directly into Intuit's platform. The target: enabling mid-market businesses to create industry-specific AI agents with no technical expertise, with a rollout to Intuit's 100 million-plus users planned for spring 2026. Anthropic simultaneously launched 13 enterprise plugins for Claude Cowork, with connectors for Google Workspace, DocuSign, FactSet, Harvey, and LegalZoom covering finance, HR, and legal workflows. Bloomberg reported the move as institutional validation that enterprise agents have reached production readiness.

Cursor shipped background agents that run 10 to 20 tasks in parallel, each on its own virtual machine, triggered from web, mobile, Slack, or GitHub. New Relic launched a no-code agentic platform for data observability. Five companies, five agent platforms, one 72-hour window. This was a coordinated land grab—every major SaaS product decided agents were table stakes on the same day. The speed suggests competitive pressure, not organic demand.

For teams evaluating where to invest, the critical question isn't whether to build agents. These platforms will push that decision for you. The question is whether you build on a single vendor's agent layer—one that only works inside their product—or on infrastructure that works everywhere. AgentPMT's workflow and skills builder addresses the same no-code agent-building need, but without locking you to a single platform. Skills created in the builder deploy across Claude Desktop, ChatGPT, Gemini CLI, Cursor, VS Code, Windsurf, Zed, and any other MCP-compatible agent.

The Lock-In Problem

Each platform's agents only work inside their own ecosystem. Notion agents operate in Notion. Intuit agents operate in QuickBooks. Cursor agents operate in Cursor. A business using all three now has three separate agent platforms that don't communicate with each other.

The math scales badly. Gartner's prediction that 40% of enterprise apps will embed agents by end of 2026 means 40% of your software stack will each run independent agent layers. Every skill built inside Notion's Custom Agents is a skill that doesn't exist anywhere else. Every workflow automated inside Intuit's agent layer is a workflow locked to QuickBooks. Switch tools or add new ones, and you start over.

MCP is the shared protocol underneath many of these platforms, but each implementation is proprietary above the protocol layer. The SaaS industry just reinvented the walled garden. Except this time, the walls trap not just your data but your automated workflows and agent intelligence.

AgentPMT's Dynamic MCP eliminates the walled-garden problem entirely. One integration gives agents access to the largest marketplace of AI tools and skills, regardless of which LLM or platform the agent runs on. Tools are fetched on demand—nothing loads into your agent's context window until it's actually needed, keeping context clean and token costs low. The server costs $0 to run, ships as a 5MB binary, and auto-detects platforms. New tools and skills become available automatically with no reinstalls, no config changes, and no restarts. Your tool catalog refreshes every 30 minutes without any action from you.

The pricing fragmentation highlights how real this problem is. Notion charges credits. Intuit's model is yet to be announced. Enterprise plugins bill per seat. Each vendor's billing operates independently, making cost prediction across an agent portfolio nearly impossible. Businesses that build on agent-agnostic infrastructure sidestep both the portability trap and the pricing unpredictability.

The Commerce Layer Catches Up

As non-developers build agents that take actions, those agents need to handle money. The commerce infrastructure is racing to keep up.

Stripe launched the Agentic Commerce Suite, enabling businesses to sell to AI agents through a single integration. The centerpiece is Shared Payment Tokens (SPTs), a new payment primitive for agentic commerce. SPTs let AI agents initiate payments using a buyer's saved payment method without exposing payment credentials. Each token is scoped to a specific seller, bounded by time and amount, and observable throughout its lifecycle. Stripe's Radar fraud detection powers the underlying risk signals, differentiating between high-intent agents and automated bots.

Major brands are already onboarding: URBN (including Anthropologie, Free People, and Urban Outfitters), Etsy, Ashley Furniture, Coach, Kate Spade, and Revolve, among others. Etsy's chief product and technology officer Rafe Colburn said the suite enables Etsy to ensure sellers' work gets discovered wherever buyers choose to shop. Stripe co-founder John Collison predicted a "torrent" of AI agent commerce powered by stablecoins and high-speed chains, according to reporting by CryptoTimes.

The commerce layer is fragmenting just as fast as the agent layer. Stripe's SPTs, Google's Universal Checkout Protocol (UCP), and x402 stablecoin payments represent different architectures with different assumptions and limited interoperability between them.

AgentPMT agents come with blockchain wallets on Base, enabling x402 and x402Direct payments natively. x402Direct is a smart contract system built on the x402 protocol that provides on-chain guarantees for every agent transaction. Agent credits—100 credits equals $1 USD, charged only on successful tool calls—deliver simple, transparent pricing. Budget controls let non-technical users set daily, weekly, or monthly spending limits, per-transaction caps, and vendor whitelists. When agents are being built by people who aren't developers, controllable payment infrastructure isn't a nice-to-have. It's the difference between automation and liability.

What This Means For You

The no-code agent wave means millions of new agents will come online in 2026, built by people who've never opened a code editor. The risks are immediate: platform lock-in traps your agent workflows in single ecosystems, coordination failure means agents across different tools can't work together, and cost spiraling accelerates when agents spend without visibility or guardrails.

The businesses that thrive will build on agent-agnostic infrastructure from the start—not lock into whichever SaaS vendor happened to ship agents first. AgentPMT was built for this moment: a marketplace with hundreds of production-ready tools, a drag-and-drop builder that works across every LLM, autonomous agent wallets with transparent pricing, and budget controls that keep spending visible and accountable. Build once, run anywhere—before your agent workflows get locked into platforms that don't talk to each other.

What to Watch

Notion's 21,000 agents were built during early testing. Full general availability numbers will reveal whether mainstream adoption is linear or exponential. Intuit's spring 2026 rollout will test whether "no technical expertise required" holds true for 100 million-plus users accustomed to accounting software. Stripe's SPT adoption among the major brands already onboarded will indicate whether agent commerce volume is months or quarters away.

The deeper question: whether MCP becomes sufficient as an interoperability standard or whether each platform's proprietary agent layer fragments the ecosystem into another generation of walled gardens. Watch for agent-building announcements from Salesforce, HubSpot, and Asana in the next 90 days. The land grab has started.

The 21,000 agents built in Notion's first week are a preview. Every productivity tool, every financial platform, every developer environment will ship its own agent-building capability within months. The choice that compounds isn't whether to build agents—that decision is being made for you. It's whether to build on infrastructure that works across all of them.

Start building portable agent workflows now. Explore the marketplace and the drag-and-drop workflow builder at AgentPMT.

Key Takeaways

• Non-developers built 21,000 autonomous AI agents in Notion's first week with zero code—mainstream agent adoption has arrived
• Five major platforms shipped agent-building capabilities within 72 hours, creating immediate platform lock-in risk for businesses building agents inside single tools
• Agent-agnostic infrastructure that works across every LLM and platform is the most durable investment as the agent ecosystem fragments

Sources

• Introducing Custom Agents — Notion Blog
• Notion 3.3: Custom Agents — Notion
• Anthropic launches new push for enterprise agents with plug-ins — TechCrunch
• Intuit and Anthropic Partner to Bring Trusted Financial Intelligence and Custom AI Agents — BusinessWire
• Anthropic Adds 13 Enterprise Plugins to Claude Cowork — WinBuzzer
• Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026 — Gartner
• Cursor announces major update to AI agents — CNBC
• New Relic launches new AI agent platform — TechCrunch
• Introducing the Agentic Commerce Suite — Stripe Blog
• Stripe Co-founder: AI Agents to Flood Commerce With Stablecoins — CryptoTimes
• How enterprises are building AI agents in 2026 — Claude Blog
• Anthropic Links AI Agent With Tools for Banking, HR — Bloomberg
• Intuit stock price jumps on Anthropic AI deal — TS2 Tech