AI Agents Stumble Over Outdated Infrastructure, Panelists Say

Modernizing the Stack: Practical Steps from the Frontlines

At VB Transform 2026 in Las Vegas, senior AI infrastructure leaders from LinkedIn, Walmart and Zendesk warned that legacy systems, not the AI models themselves, are the primary bottleneck slowing down intelligent agents. Animesh Singh of LinkedIn, Desiree Gosby of Walmart and a senior engineer from Zendesk shared their experiences and solutions during a three‑hour discussion on June 12.

The panel explained that many enterprises still rely on on‑premise servers and monolithic pipelines built years ago. These environments cannot handle the millisecond‑level response times modern AI agents require. According to Singh, „Our models can process a request in under ten milliseconds, but the surrounding data‑fetch and routing layers add hundreds of milliseconds.” The speakers highlighted how outdated networking, storage latency, and lack of containerization force companies to compromise on performance and user experience.

Participants described concrete upgrades that have begun to close the performance gap. Walmart invested in a hybrid cloud architecture, moving critical inference workloads to low‑latency edge nodes while keeping legacy databases in place. Gosby noted, „By offloading the most time‑sensitive calls to the edge, we reduced end‑to‑end latency by 40 percent.” Zendesk’s engineer recounted a migration to a microservices framework that allowed independent scaling of language‑model components, cutting processing time from 250 ms to 80 ms. LinkedIn’s team emphasized the importance of adopting high‑throughput networking fabrics and software‑defined storage to eliminate bottlenecks.

Can Legacy Systems Keep Up with Real‑Time AI Agents?

The consensus was clear: without a fundamental overhaul, older infrastructures will lag behind the rapid expectations of AI‑driven products. Singh warned that „continuing to patch legacy hardware creates technical debt that grows faster than the value of new features.” He suggested that organizations audit their data pipelines, replace synchronous calls with asynchronous messaging, and adopt container orchestration platforms such as Kubernetes. Gosby added that a phased approach—starting with high‑impact services—helps manage costs while delivering measurable speed gains.

The panel’s insights point to a broader industry shift. Companies that modernize their underlying stacks can unlock the full potential of AI agents, delivering smoother interactions and higher conversion rates. As more firms recognize the hidden cost of outdated infrastructure, investment in cloud‑native, low‑latency solutions is likely to accelerate, reshaping the competitive landscape for AI‑enabled services.

Frequently Asked Questions

What specific latency issues do legacy systems cause for AI agents? Older hardware and monolithic pipelines often add 100‑300 ms of delay, dwarfing the sub‑10 ms processing time of modern models.

How can businesses begin modernizing without a full overhaul? Start by identifying the most latency‑sensitive services, migrate them to edge or cloud environments, and replace synchronous calls with asynchronous messaging.

Will the shift to cloud‑native infrastructure affect data security? Properly configured hybrid solutions can maintain or even improve security, as modern platforms offer granular access controls and encryption by default.