AI Agent Systems Hit Enterprise Scale as Anthropic

AI agent systems moved decisively into enterprise production this week as Anthropic launched “dreaming” capabilities for self-improving agents and Microsoft took Agent 365 out of preview to address what it calls “shadow AI” — autonomous tools spreading across organizations without IT oversight.

The developments signal a shift from experimental chatbots to persistent, learning systems that can execute complex workflows autonomously. Anthropic reported that early adopters saw task completion rates increase 6x using its new dreaming feature, while Microsoft positioned Agent 365 as a unified control plane for governing AI agents across cloud platforms and employee endpoints.

Anthropic Introduces Self-Learning AI Agents

Anthropic on Tuesday unveiled “dreaming” at its Code with Claude developer conference in San Francisco — a capability that lets AI agents learn from their own past sessions and improve performance over time. The feature addresses what the company identifies as the core challenge in running AI agents at scale: maintaining accuracy while enabling autonomous operation.

According to VentureBeat, legal AI company Harvey saw task completion rates increase roughly 6x after implementing dreaming. Medical document review company Wisedocs cut document review time by 50% using Anthropic’s outcomes feature, which moved from research preview to public beta alongside multi-agent orchestration capabilities.

The announcements come as Anthropic reports growth outpacing its internal projections. CEO Dario Amodei disclosed during the conference that the company’s momentum has exceeded even aggressive internal forecasts, though specific revenue figures were not provided.

Key features now in public beta:

• Dreaming: Agents learn from past sessions to improve future performance
• Outcomes: Define specific goals and success metrics for agent tasks
• Multi-agent orchestration: Coordinate multiple agents on complex workflows

Microsoft Tackles “Shadow AI” with Agent 365

Microsoft last week moved Agent 365 from preview to general availability, positioning the platform as a governance solution for what it terms “shadow AI” — autonomous tools that employees install without IT knowledge or approval. The platform provides a unified control plane for observing and securing AI agents across Microsoft’s ecosystem, third-party clouds like AWS Bedrock and Google Cloud, and employee endpoints.

“Most enterprises are trying to figure out how to harness the potential of autonomous agents,” David Weston, Corporate Vice President of AI Security at Microsoft, told VentureBeat. “They’re trying to find a balance between what we call YOLO — just let anything run — and completely locking everything down.”

The platform addresses a new category of enterprise security risk as coding assistants, productivity tools, and autonomous workflows proliferate across organizations. Microsoft’s aggressive push into discovering and managing local AI agents reflects the urgency IT departments face in maintaining visibility and control over AI deployments.

Enterprise Automation Evolves Beyond Bot Deployment

The shift toward intelligent agent systems represents an evolution beyond traditional robotic process automation, according to enterprise automation experts. Forbes reports that many organizations have experienced “automation sprawl” as bot deployments scaled without cohesive architecture or governance.

Sanjoy Sarkar, SVP of Application Development at First Citizens Bank, argues that the next phase of enterprise transformation will be defined not by deploying more bots, but by how intelligently automation is architected and orchestrated across the enterprise — what he terms the “agentic enterprise.”

The pattern is familiar: organizations identify repetitive processes, deploy automation, scale quickly, and celebrate efficiency gains. However, the underlying architecture often becomes less cohesive over time as different departments adopt tools independently and governance practices vary across business units.

Security Applications Drive Investment

Autonomous offensive security firm XBOW raised $35 million in a Series C extension, bringing total funding to over $270 million at a valuation exceeding $1 billion. According to SecurityWeek, the platform leverages AI reasoning and adversarial workflows to continuously test applications for vulnerabilities, operating autonomously to identify and validate security holes.

“Each XBOW agent operates like an extension of our in-house red team, allowing us to scale offensive testing with speed and depth that was previously out of reach,” said Alex Krongold, director of Corporate Development & Ventures at SentinelOne.

The funding came from Accenture Ventures, DNX Ventures, Liberty Global Tech Ventures, NVentures, Samsung Ventures, and SentinelOne S Ventures, with proceeds earmarked for go-to-market and international expansion efforts.

Open Source Agent Platforms Gain Traction

The open source project OpenClaw has emerged as a significant force in AI agent development, crossing 250,000 GitHub stars by March 2026 to become the most-starred software project on the platform in just 60 days. According to NVIDIA’s AI blog, the project attracted more than 2 million visitors in a single week as developer interest surged.

Created by Peter Steinberger, OpenClaw is a self-hosted, persistent AI assistant designed to run locally or on private servers. The project gained attention for its accessibility and unbounded autonomy, allowing users to deploy AI models locally without depending on cloud infrastructure or external APIs.

The rapid adoption of OpenClaw demonstrates growing demand for AI agent systems that organizations can control and customize without relying on third-party cloud services. This trend aligns with enterprise requirements for data sovereignty and reduced dependency on external platforms.

What This Means

The convergence of enterprise-grade AI agent platforms from major vendors, significant venture investment in specialized applications, and explosive growth in open source alternatives signals that autonomous AI systems have moved beyond experimental phase into production deployment.

The emergence of “shadow AI” as a governance challenge reflects the reality that AI agents are proliferating faster than traditional IT oversight mechanisms can adapt. Microsoft’s focus on discovery and management of local agents acknowledges that the genie is already out of the bottle — the question is no longer whether AI agents will be deployed, but how organizations will maintain visibility and control over them.

Anthropic’s dreaming capability represents a technical milestone toward truly autonomous systems that improve through experience rather than requiring constant human oversight. The 6x improvement in task completion rates reported by Harvey suggests that self-learning agents may soon become competitive advantages rather than experimental tools.

For enterprises, the choice is increasingly between proactive governance of AI agent deployment or reactive management of systems that employees will deploy regardless of official policy.

FAQ

What are AI agents and how do they differ from chatbots?
AI agents are autonomous systems that can execute complex workflows, use tools, and learn from experience over time. Unlike chatbots that respond to individual queries, agents can maintain persistent sessions, coordinate with other agents, and complete multi-step tasks without constant human supervision.

What is “shadow AI” and why is it a concern for enterprises?
Shadow AI refers to autonomous AI tools that employees install and use without IT department knowledge or approval. This creates security, compliance, and governance risks because organizations lack visibility into what AI systems are processing their data and making decisions on their behalf.

How significant is the $35 million XBOW funding in the context of AI agent investments?
XBOW’s Series C extension brings its total funding to over $270 million at a $1+ billion valuation, indicating substantial investor confidence in autonomous AI applications for cybersecurity. The funding round included strategic investors like Accenture Ventures and SentinelOne, suggesting enterprise demand for AI-powered security testing at scale.