en.Wedoany.com Reported - Giant Swarm has launched an AI agent platform that treats agents as independent, orchestratable workloads running within users' own Kubernetes infrastructure, rather than as IDE extensions. The company positions the platform as an autonomous, auditable, and curated solution, emphasizing that AI agents should run where the data resides, not on external servers.
The platform treats agents as entities no different from other workloads in a Kubernetes cluster, deployable in the cloud, on-premises data centers, edge environments, or even isolated air-gapped networks, with no proprietary core or data-backhauling SaaS services. Giant Swarm CEO Henning Lange stated that the company has been running its own agent cluster on the same platform infrastructure since mid-2025 for code review, incident triage, and task processing. The agent platform is now available in beta to selected partners and customers, with a full rollout planned for later this year.
The new service aims to help enterprises address the growing problem of shadow AI. A survey by IDC last year showed that approximately 56% of employees used AI tools not provided by their organization, while only 23% used company-owned tools. IBM found that one in five organizations has suffered from shadow AI incidents, with each incident adding an average of $670,000 in costs.
On the technical side, the platform adopts a workflow engine approach, converting recurring patterns of agent operation into deterministic MCP workflows. In its own operations, Giant Swarm has used this to reduce the cost per agent run to 1/2.8 of the original, reduce tool call counts to 1/17, and run an average of approximately 500 agents in parallel. The company's monthly pull request volume in spring 2026 grew by over 300% compared to the previous year, while headcount remained unchanged.
The agent platform follows three guiding principles: autonomy, using open-source, swappable models, tools, and frameworks with full data control; curation, continuously monitoring the agent market and centrally integrating suitable components; and enterprise readiness, supporting multi-tenancy, security, and auditable operations. The architecture is based on Giant Swarm's Kubernetes management experience since 2014. The platform is cloud-agnostic, supporting AWS, Azure, VMware in on-premises data centers, and hybrid combinations. Cluster lifecycle management is based on Cluster API, GitOps uses Flux Operator, the developer portal uses Backstage, and observability relies on the Grafana-LGTM stack. Single sign-on can integrate with Google Workspace or Microsoft Entra ID, access control is managed granularly through Kubernetes RBAC, supplemented by network policies. Reference customers include Adidas, Vodafone, Deutsche Telekom, and the Stuttgart Stock Exchange.
This move aligns with an industry trend shifting from individual programming assistants to the orchestration of entire agent clusters. The framework Gas Town, developed by Steve Yegge in December 2025 and launched in early January 2026, pursues a similar goal, orchestrating multiple programming agents in parallel according to the Kubernetes model and compensating for errors through the orchestration layer. Language models do not automatically become agents—a concept also articulated in the Agent Harness: only the specific software surrounding the model can integrate the environment, tools, rules, and agent loop. Giant Swarm's platform can be seen as a platform-level implementation of this concept, emphasizing that data does not leave the user's environment. For enterprises with strict data protection and compliance requirements, on-premises, air-gapped, or hybrid operation modes become key selling points.
