Home / Technology / AI Agents Slowed by Old Tech, Not Models
AI Agents Slowed by Old Tech, Not Models
18 Jul
Summary
- Enterprise infrastructure, not AI models, causes slowdowns.
- Legacy systems struggle with agent work speeds.
- Infrastructure leaders shared insights at VB Transform 2026.
Infrastructure leaders from LinkedIn, Walmart, and Zendesk convened at VB Transform 2026 to address the primary obstacles hindering AI agent deployment. A consensus emerged that legacy infrastructure, not the AI models, is the principal cause of slowdowns. These systems, originally designed for human operational tempos, struggle to keep pace with the demands of AI agents.
Each company encountered unique infrastructure-related issues. LinkedIn faced delays with Kubernetes provisioning, necessitating a shift to pre-provisioned container pools. Walmart grappled with managing an influx of internally developed agents, requiring new governance structures. Zendesk identified data pipeline and infrastructure investment as crucial for handling vast customer conversation data effectively.
These experts advised a strategic approach to AI modernization. Key recommendations include prioritizing evaluation systems, taking ownership of agent harnesses from the outset, and building for independence from specific AI models and contexts. This focus on robust infrastructure and adaptable systems is paramount for successful AI agent implementation.
