AI agents are moving into a new phase, and the focus is shifting from what they can do to how safely they can act. Three announcements between August 26 and September 1, 2026, show that shift from different angles: fixing cloud failures, controlling agent activity, and training agents inside realistic software environments.
DataAgent has emerged from stealth with $10 million in pre-seed funding to build an AI-native platform for cloud infrastructure management. The round was led by MizMaa Ventures and Alicorn Venture Partners, and the company plans to use the money to bring its platform to market and expand adoption in North America.
DataAgent wants infrastructure systems to fix themselves
Founded in January 2026, DataAgent is based in Tel Aviv and has 15 employees. CEO Ishay Yaari and CTO Nati Shalom are building a platform that operates inside Kubernetes and connected cloud infrastructure, where it can take action when failures occur.
Most modern observability platforms watch systems, detect problems, and alert engineers. DataAgent wants to reverse that workflow by reading live system state, topology, and configuration information, then applying verified remediations first. The company describes this approach with a short slogan: “Remediation-First, the rest is noise.”
That does not mean the system gets unlimited control. Customers can define guardrails around the actions the platform is allowed to perform, creating boundaries for its automated work. DataAgent describes the platform as an autonomous site reliability engineer, or SRE, aimed at handling infrastructure problems instead of stopping at notifications.
The difference matters because detecting an outage and resolving one are separate tasks. DataAgent’s platform is designed to connect those steps by using the current condition of a system, its structure, and its configuration to choose a verified response. The company’s funding gives it capital to turn that approach into a product and pursue customers in North America.
Broadcom adds control around autonomous agents
Broadcom announced AgentMinder on August 31, 2026, as a solution for AI agent governance and runtime control. The product is generally available as of that date and takes a different approach from DataAgent: instead of repairing infrastructure failures, it checks whether autonomous agents are acting within approved limits.
Clayton Donley, vice president and general manager at Broadcom, described the need this way: “AI has moved from the pilot stage to becoming a critical part of enterprises’ daily business operations.” AgentMinder acts as a traffic controller for those agents, verifying actions against their declared mission, intent, context, and current risk.
The system treats AI agents as enterprise-grade identities. Their authority connects to a declared mission, permitted intents, approved tools, and authorized resources, which gives organizations a way to define what each agent may access and do.
AgentMinder’s cloud-native AI gateway secures tool calls at runtime, authenticates tokens, and directs traffic to authorized backends. It also provides observability and audit capabilities through OpenTelemetry, giving organizations visibility into agent sessions and actions.
Organizations can deploy AgentMinder alongside existing large language models on-premises, in virtual private clouds, or across public cloud environments. It supports VMware vSphere Kubernetes Service, Google Cloud Platform, and other standards-based, cloud-native Kubernetes platforms.
The architecture supports nearly 36 million customer-related API calls and seven million workforce-related API calls each day, including peak loads at those levels. Alan Davidson, CIO of Broadcom, said, “With AgentMinder powering our platform, we’ve achieved massive global scale paired with zero downtime, even during maintenance and upgrades.”
Donley added, “AgentMinder acts as a traffic controller, verifying exactly what these AI agents are doing, providing guardrails, and tracking their work.” That description captures the product’s role: agents can act, but their authority, tools, and destinations remain subject to checks.
Arga Labs builds realistic places for agents to learn
Arga Labs is working on another piece of the same puzzle: training environments. The company announced a $10 million seed round on August 26, 2026, led by General Catalyst, with participation from Box Group, Emergence, Gradient, and SV Angel.
Arga Labs builds training environments for enterprise software such as Salesforce, Workday, and email clients. Its technology creates full-scale digital twins of those programs while keeping their permission systems and web hooks intact, giving AI agents a setting where they can practice complex tasks.
CEO and co-founder Phillip Li said the tools help agents correctly identify duplicates, check email sends, and identify recipients. The system also makes it easier to reset and modify a digital recreation of enterprise software, allowing teams to repeat training without rebuilding an environment each time.
Arga Labs can run many environments at the same time, which lets agents train on complex interactions between different programs. Yuri Sagalov, managing director of General Catalyst, emphasized the importance of repeatable sandbox environments for enterprise AI agents.
Arga Labs has an event scheduled in San Francisco from October 13 to 15, 2026. Alongside DataAgent and AgentMinder, its work points to a broader direction for enterprise AI: agents need realistic places to learn, clear rules for action, and systems that can verify or correct what happens after deployment.
Based on
- DataAgent Emerges From Stealth With $10M Pre-Seed to Build Self-Healing Cloud Infrastructure — unite.ai
- A DBA’s Guide To Agentic AI In Production — forbes.com
- Broadcom Unveils AgentMinder, An Enterprise Solution for AI Agent Governance and Runtime Control | Markets Insider — markets.businessinsider.com
- Arga Labs is building a better way to train enterprise AI agents | TechCrunch — techcrunch.com
