How companies are modernizing infrastructure for AI agents
LinkedIn, Walmart and Zendesk share their experiences moving AI agents to production.
1The infrastructure challenge for AI agents
Leaders from LinkedIn, Walmart and Zendesk revealed at VB Transform 2026 that legacy infrastructure - not the AI models themselves - was the main bottleneck when moving agents from pilot to production. Systems designed for human speed can't keep up with agents that think in milliseconds.
2LinkedIn's Kubernetes solution
LinkedIn's first bottleneck was Kubernetes' container spin-up time taking seconds - too slow for agents. Their fix: shifting from on-demand provisioning to pre-provisioned container pools where agent workloads swap in real-time. This reduced latency significantly for their AI applications.
3Walmart's citizen developers
When Walmart put AI agent tools directly in employees' hands, usage went viral internally. 'Citizen developers' began building agents to solve problems that previously required engineering roadmaps. The challenge became managing dozens of overlapping agents without coordination, requiring new governance approaches.
4Zendesk's data pipeline investment
Zendesk discovered that simply feeding 20 billion customer conversations to LLMs didn't work. Instead, they invested heavily in underlying data pipelines and infrastructure. Modernizing their data processing and storage systems became key to making AI agents effective with their massive datasets.
5Open source and independence
All three companies converged on similar open source strategies: own what you can, use frontier models only where needed. LinkedIn built an AI gateway and independent memory subsystem. Walmart created an internal gateway to remain vendor-agnostic across different workload types while ensuring compliance.
6Key lessons for AI modernization
Three critical lessons emerged: 1) Invest early in evaluation systems 2) Put AI tools directly in employees' hands quickly 3) Design for model and context independence. These principles enable scalable deployment of AI agents while maintaining flexibility as technology evolves.