# How companies are modernizing infrastructure for AI agents Lessons from LinkedIn, Walmart and Zendesk on building AI agent infrastructure. - **Source:** [VentureBeat](https://venturebeat.com) - **Published:** 2026-07-17 - **Topic:** [Data & AI](https://brifing.si/c/data-ai) - **Original article:** - **On Brifing:** https://brifing.si/s/af981beb-de60-4c51-bde3-e08c435daa54 ## The bottleneck for AI agents Infrastructure leaders from LinkedIn, Walmart and Zendesk revealed at VB Transform 2026 that legacy systems designed for humans - not the AI models themselves - were slowing down agent deployment. Container spin-up times measured in seconds proved inadequate for agents that think in milliseconds. ## LinkedIn's container solution LinkedIn moved from on-demand provisioning to pre-provisioned container pools for real-time workload switching. They also reduced LLM orchestration by scripting 80% of workflows, using reasoning only where necessary. This structural change addressed hallucination issues in their evaluation systems. ## Walmart's citizen developers When Walmart gave employees direct access to AI tools, it sparked an internal movement of 'citizen developers' creating their own agents. While this drove innovation, it also created duplication. The solution was governance infrastructure to identify the best agents without making engineering a bottleneck. ## Zendesk's data-first approach With 20 billion customer conversations in its repository, Zendesk chose to invest in foundational data pipelines rather than dumping raw data into large context windows. Ghoche emphasized that robust data infrastructure is more critical than model size for effective agent deployment. ## Open source and independence All three companies adopted similar strategies: own what you can, use open source where practical. LinkedIn built provider-agnostic AI gateways and memory subsystems. Walmart created flexible systems to choose between frontier and open-weight models based on specific workload needs. ## Modernization advice Key recommendations from the panel: 1) Invest in evaluation systems early, 2) Empower employees with AI tools to drive innovation, 3) Build infrastructure for model and context independence. These approaches are critical for unlocking AI agents' full potential while maintaining flexibility for future advancements. ## Sources covering this story - https://venturebeat.com/data/agents-think-in-milliseconds-legacy-infrastructure-doesnt-linkedin-walmart-and-zendesk-shared-how-they-closed-the-gap-at-vb-transform-2026 — --- _Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._