- Sadeepa Wijesekara of HPE likened AI token use to early company purchasing cards, with free use first and controls later.
- She said customers start with routing rules that people write, then watch and tune them before trusting an automatic router.
Sadeepa Wijesekara, who heads the engineering team behind HPE Private Cloud AI, compared today's AI spending with the early days of company purchasing cards. She spoke on The TWIML AI Podcast with host Sam Charrington. In her telling, free use came first, then possible waste, then controls. WebPulse adds a point of its own: choosing the right model for each task is becoming a skill.
What was said
Wijesekara said staff who cannot see what a model costs their employer tend to pick the best one, or never change the default. She likened it to a $20 chatbot subscription. People also experiment, she said, simply because they can.
Then came the comparison: "early days of P cards." Purchasing cards let staff buy freely. Later, companies added rules on what could be bought. She said firms now see maybe some waste in AI tokens, the units AI services charge for. Making the individual responsible, she said, at least creates a control point.
Routing is the next step: sending each request to a hosted or in-house model. Customers mostly begin with rules that people write, such as sending one type of work to the cloud. A later level reads each prompt and chooses automatically. Customers are cautious there, she said, because "now I need to monitor the model router." They watch the rule-based routing and tune it first. She added that a total such as $10 million in token spend should be split by agent and use case.
Why it matters
Our reading: the judging moves from the individual user to whoever writes the rules. An agent can make many model calls to finish one job, so a single bill hides which calls needed a top model. Rule-based routing is easy to read, because a given type of work goes to a given place. Automatic routing decides from the prompt's content, so it adds a decision-maker that itself needs checking.
A budget signer can ask three questions. Who owns the routing rules? What is logged for each agent and use case? Who reviews the router's decisions? Wijesekara did not list these questions. They are our inference from her account.
The other side
One might say that routing means people no longer need to choose a model. That is a counterpoint WebPulse is raising, not something her examples show. In her account, people stay in the loop, writing rules and watching the routing. The autonomous router is the step customers hold back from.
She also questioned the cost focus: "right now we are maybe worrying too much about cost." She said the value side needs more work. Her purchasing card story is one leader's analogy, not a measured result. She describes observing and tuning rule-based routing before automating it. The excerpts give no detail on how that checking is done or what counts as a correct choice.
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
- The TWIML AI Podcast: Buying Tokens vs Building an AI Factory with Sadeepa Wijesekara - #785 (2026-11-17)





