{
  "id": 8253920,
  "title": "Your agent is only as good as your infrastructure",
  "url": "https://urgent.news/2026/09/18/your-agent-is-only-as-good-as-your-infrastructure",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-18T13:00:00.000Z",
  "source": {
    "name": "The New Stack",
    "slug": "the-new-stack",
    "url": "https://thenewstack.io/ai-agent-infrastructure-performance/"
  },
  "original_language": "en",
  "account": "Your agent's performance is heavily reliant on the infrastructure it operates on. In testing, the agent swiftly reviewed pull requests independently, but in production, the same steps ran sequentially for every review. Occasionally, due to resource contention, reviews took longer. This illustrates how infrastructure significantly impacts latency, reliability, and cost in AI agents compared to traditional chatbots.\n\nAn AI agent executes a sequence of inference calls and tool executions, unlike a chatbot that makes one inference call per user message. Each reasoning step adds to the model's context, triggering subsequent inference calls. This multi-turn tool call or agentic loop model changes infrastructure requirements as it depends on every component in the chain.\n\nUnlike a simple inference request, an agentic workflow is inherently sequential. A single user request can expand into numerous inference and tool execution steps, each waiting for the previous step to finish. This sequential nature amplifies latency, making even low latency feel like a significant amount of time. Infrastructure must hold the chain together, as its slowest link determines the overall performance.\n\nThe unpredictability of demand patterns in agentic workflows can lead to unpredictable inference bills and performance issues. Most inference services assume steady request patterns, but AI agents often pause while waiting on external systems, then resume rapidly when results come back. This creates short, sharp spikes in GPU demand, leading to inconsistent latency and GPU underutilization. GPU utilization patterns resembling a heartbeat, rather than steady load, indicate the need for provisioning for peak demand rather than average.",
  "summary": "You built a great agent, but something happened when it moved into production. In testing, your agent reviewed pull requests The post Your agent is only as good as your infrastructure appeared first on The New Stack .",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}