IBM's pitch for feeding AI in the agentic era rests on a new architecture it calls the "stream house." It's built to hand a large language model a business's real-time state, not its archived history. The old mantra held that data was "the new oil," mined once and stored. Shaun Clowes, General Manager of Confluent and Data at IBM, told John Furrier and Dave Vellante on theCUBE that in the agentic era, data instead has to behave like continuous fuel for a system that never shuts off.
"An LLM is like a smart brain in a computer, but it is limited entirely by what it can see about the world."
Shaun is the one responsible for making IBM's stream-house bet pay off: Confluent's streaming technology, now part of IBM's data business, is the infrastructure he is betting enterprise AI will run on.
The full interview is covered here so you can skip it. 20 minutes of audio, 7 minutes of reading.
Here are the 5 takeaways that matter.
SiliconANGLE theCUBE · Oct 22, 2026 · 20m · 13m saved
With Shaun Clowes, General Manager of Confluent and Data, IBM
This Scuttlebutt Summary is available in full to premium subscribers. If you're not one already, please consider subscribing to keep reading.
Key Takeaways
An LLM is "a smart brain in a computer" that's blind without a constant feed of data, Shaun says
Data has to behave like continuous fuel for an always-on system, not oil mined once and stored
IBM's "stream house" architecture hands an LLM a business's real-time state, not its stored history
Getting the wrong data to an LLM fast is as useless as getting the right data too slowly
By the time slow data arrives, the decision is usually already made
You wouldn't hand a nuclear physicist your accounts receivable — so don't hand an LLM everything and expect it to reason well
RPA failed because it couldn't reason, not because it was too narrow
The opposite failure now: giving a model the whole enterprise and hoping it finds a bounded task on its own
Complaints about agents making "too many tool calls" really mean wasted tokens, Shaun says
1. The Stream House
Shaun framed his whole pitch around one image: an LLM reasons well only if it can see the world, and in the agentic era that means data has to arrive continuously, not get archived for later.
Enterprise data is "the electrons that feed into the LLMs," not a resource to mine and store, Shaun says
"Data is quite literally the electrons that feed into the LLMs, and without them nothing good happens." — Shaun Clowes
That's a break from the old cliché that data is "the new oil" — oil gets extracted once and refined later. Shaun's point is that an agent is always running, so the data feeding it can't go stale between uses.
IBM calls its answer the "stream house" — built to hand an LLM a business's current state directly, not its archived history
"We need a new type of architecture, a type of architecture that can bring together all the state of our business right now and present that directly to LLM. So we call that the stream house architecture." — Shaun Clowes
John Furrier, who co-hosts theCUBE Research with Dave Vellante, traced the arc with Shaun: from Hadoop's batch processing, through Amazon S3-based data lakes, to an architecture built specifically to keep an AI model current instead of just archiving records for a person to query later.
