Premium sportswear brand On used AI agents on Google Cloud to migrate 24 core services, reducing migration time from three months to two weeks per service, according to the companies. The project demonstrates how agentic AI can automate complex cloud modernization tasks while keeping engineers responsible for code reviews, infrastructure changes, and production approvals. On says its internal team completed 15 migrations without external implementation support, with less than five minutes of planned downtime per service.
On Uses Agentic AI to Rethink Cloud Migration
Cloud migration is often a lengthy engineering exercise involving application analysis, infrastructure configuration, data movement, testing, and production cutovers. For companies operating interconnected services across global markets, the challenge is to modernize infrastructure without interrupting business operations or diverting engineers from product development.
On, the Swiss premium performance sportswear brand, is using Google Cloud’s AI capabilities to change how it approaches that work. The company says AI agents helped its engineers migrate 24 core services, reducing the reported migration time per service from three months to two weeks. Fifteen migrations were completed entirely by its internal engineering team.
The deployment illustrates a growing enterprise AI use case: moving beyond tools that assist individual developers toward coordinated agents that execute defined stages of complex technical workflows.
How the Multi-Agent Migration Worked
According to Google Cloud and On, the migration relied on specialized AI agents assigned to discrete tasks. These included analyzing codebases, generating infrastructure configurations, creating translation pipelines, and validating tests.
The agents helped engineers map inconsistent deployment patterns across more than 20 code repositories in hours rather than days, according to the companies. They also handled repetitive activities such as configuration scripting, data replication, and environment validation.
Rather than handing over the entire migration to autonomous software, On adopted a bounded-execution model. Engineers reviewed generated code, executed infrastructure changes, and authorized production cutovers. Defined tasks and verification checks constrained the agents’ work, while consequential decisions remained under human control.
This distinction matters for enterprise adoption. Cloud infrastructure changes can affect availability, data integrity, security, and customer-facing applications. Automating repeatable work while requiring human approval for sensitive actions provides a more controlled route to agentic AI adoption than unrestricted autonomy.
On also cited a commercial incentive. An external vendor had initially quoted $500,000 to migrate a fraction of the microservices using traditional methods. The company says its internal team completed the broader migration without external implementation resources. The comparison suggests potential savings, although the companies did not disclose the project’s total internal cost or a fully comparable cost breakdown.
Google Cloud Becomes an AI Foundation
The migration is part of a broader infrastructure strategy. On selected Google Cloud to consolidate its compute, data, and AI capabilities on a common platform as the business expanded internationally.
That consolidation is intended to reduce architectural fragmentation and make it easier for teams to develop and deploy intelligent workflows. A shared cloud foundation can also simplify the process of connecting enterprise data with AI applications, although the actual benefits depend on data governance, integration quality, access controls, and operational maturity.
On has extended its AI strategy beyond engineering. The company rolled out Gemini Enterprise to employees, enabling them to build custom no-code agents for activities including executive reporting, market research, scheduling, and employee onboarding. It has also served as a strategic partner and tester for Google’s Gemini agent for work.
The wider deployment points to two distinct uses of enterprise AI: agents that automate technical operations and workplace agents that help employees complete business processes. Both depend on reliable access to organizational information and clear boundaries around what an agent can do.
Human Oversight Remains Central
On’s results offer a practical example of how businesses can structure agent-led automation. Instead of treating AI as a replacement for engineering judgment, the company used agents to accelerate execution while retaining human responsibility for consequential changes.
However, the reported timelines are company-provided results, not an independently audited benchmark. Migration complexity, application dependencies, testing requirements, compliance obligations, and rollback procedures can vary significantly between environments. The two-week figure should therefore not be interpreted as a universal timeline for enterprise cloud migrations.
For technology leaders, the more useful lesson is architectural and operational: identify repeatable work, break it into verifiable tasks, give agents narrowly defined responsibilities, and preserve approval gates for high-impact actions.
What It Means for Enterprise AI
The project reflects a broader shift in enterprise AI from experimentation toward workflow execution. Organizations are exploring agents that can coordinate multiple steps across software development, IT operations, data management, and business functions.
Google Cloud benefits when customers use its infrastructure as a shared foundation for data and AI workloads. For enterprises, however, platform consolidation is only one part of the decision. Teams must also assess portability, integration with existing systems, model performance, security, governance, and the long-term cost of operating agent-driven workflows.
On’s migration provides a concrete example of agentic AI applied to a traditionally labor-intensive technology project. Whether similar approaches deliver comparable results elsewhere will depend on the quality of the underlying architecture and the controls surrounding automated execution.
Market Landscape
Enterprise cloud modernization is increasingly intersecting with generative AI and agentic automation. Traditional migration tools help assess applications, move workloads, and validate environments; agent-based systems aim to coordinate these activities with less manual intervention.
Google Cloud’s Gemini ecosystem is part of a competitive enterprise AI market that also includes Microsoft Azure AI, Amazon Web Services, and other platforms offering cloud infrastructure, foundation models, and developer tools. The differentiation increasingly involves more than model access: enterprises need integration with business data, workflow orchestration, identity controls, and mechanisms for supervising agent actions.
On’s case highlights a potential advantage of consolidating compute, data, and AI services, but it does not establish that a single-cloud strategy is optimal for every organization. Hybrid environments, legacy dependencies, regulatory requirements, and existing vendor commitments remain important considerations.
The key measure of agentic AI maturity will be whether organizations can reproduce productivity gains safely across different workflows—not simply how many tasks an agent can perform, but how reliably it completes them under appropriate human oversight.
Top Insights
- On says AI agents helped migrate 24 core services, reducing reported migration time from three months to two weeks per service.
- Fifteen migrations were completed in-house, with engineers retaining approval authority over infrastructure changes and production cutovers.
- The agentic workflow automated code analysis, configuration generation, data replication, and testing across more than 20 code repositories.
- On deployed Gemini Enterprise across its workforce to support no-code agents for reporting, research, scheduling, and onboarding.
- The project reflects a wider enterprise trend: moving from AI experimentation toward supervised, multi-step automation integrated with cloud infrastructure.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
