Skip to content
The AI-First Web Talking point

HPE says customers start with simple routing rules, as a smart router needs watching

HPE's Sadeepa Wijesekara said his customers begin with fixed routing so they can watch it before automating.

W
WebPulse Newsroom
AI-assisted · 2 min read
Share on X LinkedIn
HPE says customers start with simple routing rules, as a smart router needs watching
In brief
  • Sadeepa Wijesekara of HPE said the enterprise customers he meets begin with fixed routing rules because a smart router becomes one more system to monitor.
  • Our reading is that the work moves from choosing a model to checking whoever or whatever chooses.

Sadeepa Wijesekara heads the engineering team behind HPE's Private Cloud AI product. He told The TWIML AI Podcast that the customers he meets begin with fixed routing rules, not a router that judges each request. His reason: a router that decides for you is one more thing to watch. He described AI agents that make many model calls, and each call can go somewhere different.

What was said

Wijesekara explained that one agent job can involve many model calls. Some calls need a top-tier hosted model. Others can run on a model the company hosts itself. Some involve data that cannot leave the company. He said the choice depends on cost, but also on delay, compliance and data rules.

He said routing products sit in front of the models and pick where each request goes. The main approach he sees is fixed rules: this kind of work goes to the cloud, that kind stays in-house. A step further, a router reads each prompt and judges how demanding it is.

Customers hold back from that step. In his words, "if the model router is super intelligent and deciding where to go, it creates another place of like, now I need to monitor the model router". So they begin with rules, watch how requests are routed, tune it, and only then consider automated routing.

He added that none of this works without knowing where the spend goes. A single total, such as his example of $10 million, is not enough. Firms must split it by agent and use case. Asked if most organisations can already see that split, he said no, and called it a step zero.

Why it matters

Our reading is that the skill here is judging, not building. If routing is a per-call decision, someone must write the rules, read the routing record and spot a wrong choice.

Wijesekara described the groundwork: tag each agent's activity to a business process, then see what each step costs. A team that cannot do that cannot judge any router, simple or clever. For managers, the first purchase may be visibility, not a smarter model picker.

The other side

Wijesekara works for a company that sells in-house AI systems. He said running on-premises could save up to 60% of token costs. That is his claim, and the excerpts give no evidence behind it.

He did not say smart routers fail. He described rules as the starting point, with automated routing coming later for customers who build confidence. He also said, "right now we are maybe worrying too much about cost," and that risk questions deserve more attention.

The host raised a further objection near the end: hosted-model labs offer wide capabilities and operate at huge scale. The excerpt stops before the answer, so that question stays open here.

Written by the WebPulse Newsroom with AI assistance, and checked by our editorial review: every quotation was verified against the recording's transcript. How we use AI.

The conversation this talking point comes from