The Week Agent Infrastructure Went Federal

NIST launched the AI Agent Standards Initiative on February 17, 2026. Four days later, Docker published survey data from over 800 global technology leaders showing that 60% of organizations already run AI agents in production — but 40% can't scale past their current deployments because security and governance remain unsolved. The federal government and the private sector arrived at the same conclusion in the same week: the model race is over. The infrastructure race is starting.

This wasn't a coincidence. In a single seven-day window, the U.S. government created its first-ever dedicated standards body for AI agents, Slack and Atlassian shipped production MCP (Model Context Protocol) servers, Cisco announced purpose-built agent observability, Samsung expanded its multi-agent ecosystem, and Microsoft previewed cloud PCs built specifically for autonomous agent workloads. Every signal pointed in one direction: the bottleneck is no longer model capability.

It's identity, interoperability, security, and governance — the infrastructure layer that determines whether agents can actually operate at enterprise scale. This is exactly the problem set AgentPMT was built to solve. Dynamic MCP for interoperability with zero context bloat, AgentAddress for wallet-based agent identity, agent wallets on Base with budget enforcement, and complete audit trails — every pillar NIST is now standardizing around maps to infrastructure we're already shipping.

The gap between deployment and governance is the defining challenge of 2026. Docker's numbers make it visceral: 60% deployed, 40% blocked. CrewAI's survey of 500 C-level executives at companies above $100M in revenue found that 100% plan to expand agent adoption this year, with security and governance rated the top priority by 34% of respondents. The agents are running. The controls aren't keeping up.

The Federal Starting Gun: NIST's Three Pillars

The U.S. government doesn't launch standards initiatives for speculative technology. It does it for technology that's already deployed and causing problems that need structured answers. NIST's Center for AI Standards and Innovation (CAISI) built its AI Agent Standards Initiative around three pillars: facilitating industry-led standards development with U.S. leadership in international bodies, fostering community-led open-source protocol development, and advancing research in AI agent security and identity.

That third pillar is the one to watch. NIST's National Cybersecurity Center of Excellence (NCCoE) separately published a concept paper titled "Accelerating the Adoption of Software and AI Agent Identity and Authorization" on February 5, with comments due April 2. The paper specifically names MCP, OAuth 2.0/2.1, SPIFFE/SPIRE, and OpenID Connect as standards under active consideration. A separate Request for Information on AI Agent Security has a March 9 deadline. The federal government is collecting input right now — not in some abstract future.

Michael Kratsios, director of the White House Office of Science and Technology Policy, framed the initiative as essential to trust: standards that "will ultimately empower the proliferation of this technology across many industries." But the sharper assessment came from Gunter Ollmann, CTO of Cobalt Labs, who cautioned that "standards alone will not prevent abuse. Security validation, continuous testing, and adversarial simulation must evolve in parallel so organizations can understand how agents behave under attack conditions before those weaknesses are exploited in the wild."

He's right. Standards set the floor, not the ceiling. And the floor matters because right now most organizations don't even have one. According to Microsoft's Cyber Pulse report, more than 80% of Fortune 500 companies are actively deploying AI agents — while UC Berkeley's Center for Long-Term Cybersecurity published a 67-page agentic AI risk management profile building on NIST's existing AI Risk Management Framework with four core functions: Govern, Map, Measure, Manage.

The academic community, the federal government, and enterprise IT are all scrambling to define the same thing: how do you give an autonomous system an identity, scope its permissions, and audit what it does?

AgentPMT's answer is AgentAddress — an open-source, wallet-based identity system where the agent's blockchain wallet is its identity. EIP-191 wallet signatures, no accounts, no API keys, no OAuth flows. Compatible across 55+ EVM chains. NIST is calling for agent identity standards. This is one that already exists in production.

The 60% Production Line

Docker's State of Agentic AI report, published February 20 and based on surveys of 800+ global technology leaders, delivered the most granular picture yet of where enterprise agent deployment actually stands. The headline — 60% already in production, 94% viewing agents as a strategic priority — gets quoted. The more important numbers are buried deeper.

Forty percent of organizations cite security and compliance as the primary barrier to scaling their agent deployments. Forty-eight percent cite orchestration complexity. Eighty-five percent are familiar with MCP but report "significant security, configuration, and manageability issues" preventing production-scale deployment. Nearly half of organizations use four to six different models, and 61% combine cloud and local execution, creating coordination overhead that compounds with every new integration.

Seventy-six percent are concerned about platform lock-in. That number should concern every vendor in this space.

The parallel data from CrewAI's enterprise survey reinforces the picture. Of 500 C-level executives at companies with $100M+ in revenue, every single one plans to expand agent adoption in 2026. Sixty-five percent already use agents, with 81% at full scale or actively expanding.

Organizations have automated 31% of their workflows and expect to add another 33% this year. Seventy-five percent report high impact on time savings. Sixty-nine percent report significant cost reductions.

The demand side is solved. Everyone wants agents, everyone's deploying agents, and the ones who've deployed them are seeing returns. What's not solved is the infrastructure that makes deployment safe, auditable, and scalable.

AgentPMT's Dynamic MCP addresses the exact pain points Docker surfaced. For security: Dynamic MCP runs 100% in the cloud with encrypted credential storage — agents never see your secrets. For configuration: one command installs and auto-detects your AI platforms. For manageability: centralized dashboard and mobile app for real-time monitoring across every agent.

Platform lock-in? Build once, run on every LLM — Claude, ChatGPT, Cursor, Gemini, Codex, local models, all of it. Orchestration complexity? A drag-and-drop workflow builder with clear task definitions, auditable failures, and per-step cost tracking. Docker found the problems. The solutions exist today.

The Enterprise MCP Tipping Point

When Slack, Atlassian, Cisco, Samsung, and Microsoft all ship agent infrastructure within the same month, that's a market declaring its priorities.

Slack's MCP Server and Real-Time Search API hit general availability on February 17. More than 50 industry partners — including Anthropic, Google, OpenAI, Perplexity, and Cursor — built Slack-powered agents, driving a 25x increase in both real-time search queries and MCP tool calls since limited release. Scott White from Anthropic noted that "customers are increasingly seeing the value of this bi-directional integration, and usage is accelerating as a result."

Atlassian's Rovo MCP Server went GA on February 4, connecting Jira and Confluence to 15+ AI clients with enterprise-grade security: admin controls for approving which clients connect, MCP usage logs for visibility into AI interactions, and permissions aligned with existing team structures. Nearly 40% of Atlassian's monthly active users are enterprise customers — the audience that cares most about governance.

Cisco announced AgenticOps innovations across its portfolio at Cisco Live EMEA on February 10, with Splunk AI Agent Monitoring going live on February 25. It's the first major enterprise observability tool purpose-built for tracking agent performance, cost, quality, and behavior — with forthcoming integration into Cisco AI Defense for mitigating hallucination, bias, data leakage, and prompt injection. Jeetu Patel, Cisco's president and chief product officer, called AgenticOps "a profound and fundamental shift away from complexity."

Samsung expanded its Galaxy AI multi-agent ecosystem on February 22, adding Perplexity as an agent. Eighty percent of Samsung users now use two or more AI agents, signaling that multi-agent environments aren't just an enterprise pattern — they're going consumer.

And Microsoft previewed Windows 365 for Agents: cloud PCs purpose-built for autonomous agent workloads, with Microsoft Entra Agent ID providing cryptographic identity credentials and audit logs that distinguish agent activity from human activity.

Each of these solves one slice of the infrastructure problem. Slack provides conversational context. Atlassian provides project data. Cisco provides observability. Microsoft provides compute and centralized identity.

But no single vendor in this list offers tool discovery, payment capability, budget controls, cross-platform governance, and identity in one integration. That's the gap AgentPMT fills. Dynamic MCP works alongside Slack's MCP server, Atlassian's MCP server, and any other MCP implementation. What AgentPMT adds is the marketplace — the largest marketplace of AI tools and AI skills — plus the wallets, the budgets, and the governance layer that connects all these data sources to actual tool execution with cost controls.

The Identity Problem NIST Can See Coming

Strip away the product announcements and vendor positioning, and one problem sits underneath everything else: agent identity.

NIST's NCCoE concept paper frames it precisely. AI agents "operate continuously, trigger downstream actions, access multiple systems in sequence." Traditional authentication — built around humans logging in, approving sessions, and managing credentials — breaks when the entity doing the work never sleeps, never takes a break, and can make thousands of decisions per hour. The question isn't just "who is this agent?" It's "what can it access, who authorized it, how much can it spend, and is there a record of everything it did?"

Microsoft's approach with Entra Agent ID uses centralized cryptographic credentials tied to Azure infrastructure. It's a natural extension of their identity stack for organizations already deep in the Microsoft ecosystem. But it's also cloud-vendor-locked by design.

AgentPMT takes a different architectural approach. AgentAddress provides decentralized, wallet-based identity where the agent's blockchain wallet is the credential. No centralized provider dependency. No accounts to compromise. The agent signs with its wallet, the signature is verified, and the identity is established — across any service that accepts wallet authentication, not just AgentPMT.

Pair that with agent wallets on Base that enforce budget limits on-chain through x402Direct smart contracts, and you get identity, payment capability, and spending governance in a single primitive. Every transaction is recorded on-chain and mirrored in the dashboard. Compliance-ready by default.

This is the infrastructure the industry will need as agents move from single-task execution to persistent, multi-system operation. Identity tied to authorization tied to auditability. NIST is writing the standards. The architecture already exists.

What This Means for You

The infrastructure stack is forming right now, and the choices made in the next six months will lock organizations in for years.

If you're in the 60% already running agents in production, your immediate risk is the governance gap. Docker's data confirms what incident reports have been showing: the number-one scaling barrier isn't model capability — it's security, compliance, and orchestration complexity. If your agents don't have per-transaction budget limits, credential isolation, and audit trails, you're operating on assumptions that won't survive an audit or a breach.

If you're in the 40% blocked by security concerns, the infrastructure solutions exist today. You don't have to wait for NIST to finish its standards process. Platforms like AgentPMT already ship the controls those standards will codify: agent identity without shared credentials, budget enforcement at the wallet level, encrypted credential storage where agents never access your secrets, and full audit trails on every tool call.

Federal standards are being shaped right now. The NIST RFI deadline is March 9. The identity concept paper deadline is April 2. CAISI listening sessions on sector-specific barriers begin in April.

Companies that participate help shape the standards. Companies that don't will comply with standards they had no input on.

What to Watch

March 9, 2026: NIST CAISI Request for Information on AI Agent Security — deadline for public input that will shape federal technical guidelines.

April 2, 2026: NIST NCCoE Agent Identity and Authorization concept paper — comment deadline that will inform the practical guide for agent identity implementation.

February 25, 2026: Cisco AgenticOps and Splunk AI Agent Monitoring go generally available — the first purpose-built enterprise agent observability tool. Watch for early adoption data.

Enterprise MCP adoption metrics: Slack reports 50+ partners and 25x activity growth. Atlassian connects 15+ AI clients. Track whether Salesforce, ServiceNow, and Workday announce MCP servers in the coming months — that's the signal that MCP has locked in as the enterprise standard.

The 60/40 benchmark: Docker's production-to-security-wall ratio is now the number to beat. Watch for movement on that 40% as infrastructure solutions mature.

Key Takeaways

• NIST launched the first federal standards initiative for AI agents, with active comment periods (March 9 and April 2) shaping the rules around agent identity, interoperability, and security.
• Docker's data shows 60% of enterprises run agents in production, but 40% can't scale past security and governance barriers — making infrastructure, not model capability, the defining bottleneck of 2026.
• Slack, Atlassian, Cisco, Samsung, and Microsoft all shipped agent infrastructure in the same month, signaling MCP's shift from developer tool to enterprise standard.
• The infrastructure stack forming now — identity, governance, payments, tool access, auditability — will lock organizations in for years. The time to evaluate and adopt is before standards are finalized, not after.

The model wars made the headlines. The infrastructure war will make the money. Sixty percent of enterprises already run agents in production. The other forty percent aren't waiting for better models — they're waiting for better infrastructure. Identity, governance, payments, tool access, auditability.

That's the stack. That's what decides who scales and who stalls. AgentPMT built this stack — Dynamic MCP for tool access, AgentAddress for identity, agent wallets for payments and budget enforcement, complete audit trails for compliance. The infrastructure the industry is standardizing around? We're already shipping it.

Sources

• Announcing the AI Agent Standards Initiative for Interoperable and Secure Innovation — NIST
• NIST launches AI Agent Standards Initiative as autonomous AI moves into production — SiliconANGLE
• NIST agentic AI initiative looks to get handle on security — Federal News Network
• New Concept Paper on Identity and Authority of Software Agents — NIST NCCoE
• State of Agentic AI Report: Key Findings — Docker
• Security and complexity slow the next phase of enterprise AI agent adoption — Help Net Security
• Agentic AI Reaches Tipping Point: 100% of Enterprises Plan to Expand Adoption in 2026 — BusinessWire / CrewAI
• Slack Securely Powers Your Third-Party Agents With Your Business Context — Slack
• Atlassian Rovo MCP Server is now GA — Atlassian
• Cisco Expands AgenticOps Innovations Across Portfolio — Cisco Newsroom
• Galaxy AI Expands Multi-Agent Ecosystem To Give Users More Choice and Flexibility — Samsung Global Newsroom
• Windows 365 for Agents brings managed cloud PCs to autonomous workflows — Help Net Security
• New CLTC Report Provides Framework for Managing Risks of Agentic AI — UC Berkeley CLTC