AI Claws Signal the Next Phase of Enterprise AI: Persistent, Collaborative Agents

The next phase of enterprise AI innovation will be defined less by standalone model performance and more by how autonomous agents are architected, deployed, and governed, according to global technology intelligence firm ABI Research. In its latest analysis, ABI Research identifies AI claws as an emerging class of persistent, long-running agents designed to move enterprises beyond session-based assistants and toward more continuous, autonomous knowledge-work systems.

“Enterprise AI is entering a new phase in which the real innovation is happening around the agentic harness—memory, tool orchestration, model routing, and workflow execution—not just inside the model itself,” says Larbi Belkhit, Principal Analyst at ABI Research. “AI claws and autonomous agents reflect that shift because they are designed to persist, operate across longer time horizons, and interact more deeply with local systems, which makes them far more relevant to real enterprise productivity workflows.”

This transition is being accelerated by several technology developments, including the maturity of agent harnesses, the rising adoption of open-weight models, and growing enterprise demand for lower-cost, lower-latency, and more private AI execution. ABI Research notes that as agentic workloads expand and token consumption rises, enterprises are increasingly exploring hybrid deployment strategies that balance device, edge, and cloud inference, with inference orchestration becoming a central capability for production-scale agent systems. While the firm forecasts enterprise spend on Generative AI software at the edge will exceed US$70 billion by 2030, the bigger story is the architectural shift now underway as AI becomes more distributed across enterprise environments.

The market is already showing how quickly that innovation trajectory is taking shape. Anthropic has introduced Claude Cowork and Claude Tag to support both personal and shared team-based agent experiences; Microsoft has positioned Scout within its Microsoft 365 ecosystem as part of its Autopilot category; Perplexity has added hybrid local-cloud inference orchestration to Perplexity Computer; NVIDIA is advancing open agent blueprints and secure runtimes through NemoClaw and OpenShell; and Intel’s SuperClaw emphasizes local inference and intelligent routing across AI PCs, workstations, and edge devices. Together, these developments point to a market evolving from isolated assistants toward modular agent ecosystems built for longer-running, enterprise-grade automation.

“The long-term opportunity is not just the personal AI assistant, but the AI colleague,” Belkhit concludes. “As autonomous systems become provisioned to teams and departments, vendors that can combine open frameworks, secure runtimes, and intelligent orchestration across device, edge, and cloud infrastructure will be best positioned to shape the next wave of enterprise AI adoption.”

These findings are from ABI Research’s AI Claws: Market Developments & Commercial Opportunities report, part of the company’s AI & Machine Learning research service, which includes research, data, and ABI Insights.