Don't Panic! It's Just Data

There are a few genuinely interesting ideas - the knowledge/memory duality for agents, the problem of non-deterministic LLM data modeling creating downstream inefficiencies, and the concept of decision traces - but they are buried under a lengthy database history recap and extended vendor pitch for MeKo. The ratio of novel-per-minute to filler is low.

“agents and like LLMs in specific are notoriously non deterministic. If you run two times, they give you different answers and they're not exactly the experts in how to do the best Data modeling”

“you're burning tokens to get a different agent to guess why you would have come up with something instead of just sharing what you did”

The knowledge-versus-memory framing applied to agentic infrastructure is moderately fresh, and the decision-trace argument has some novelty, but the episode repeatedly falls back on recycled takes: 'AI is only as good as the data,' 'silos are bad,' and a well-worn database genealogy tour. The host even coins 'context is the new oil' and claims it as original.

“context is the new oil. You heard it here first”

“your AI is only as good as the data”

Karthik has genuine practitioner depth - co-building Apache Cassandra at Facebook, working on HBase and RocksDB, co-founding Nutanix before Yugabyte - making him a credible infrastructure operator. However, the conversation functions largely as a product launch vehicle for MeKo, which caps how much raw operator wisdom surfaces.

“my co founder and I and a bunch of others, we ended up building Apache, Cassandra and open sourced it. We worked on HBase, we worked with a team that built RocksDB”

“I was at Nutanix, a company that builds distributed block stores”

There are a handful of concrete anchors - Paramount Plus scaling for Champions League finals, four named internal agents, a rough '5 to 10 million per year' savings figure - but customer metrics are unattributed and approximate, no adoption or benchmark data is provided for MeKo, and most claims about efficiency gains and cost control remain hand-wavy.

“we have customers, um, that are using us. Like for example, a Paramount plus, like one of their workloads is they need to be able to adapt on the fly”

“typically I think the number comes to about you know, 5 to 10 million per year is what we're able to save in a lot of cases”

The host asks one worthwhile follow-up (pressing deeper on knowledge vs. memory) but otherwise lobs softballs, never challenges product claims, and inserts filler asides ('context is the new oil, you heard it here first') that break momentum. The interview functions as an unchallenged press release for a new product launch.

“Can you dive a little bit further into this idea of knowledge versus memory? I think it's fascinating. It's not. I don't hear that a lot”

“Well, this has been great stuff. This is a wrap, though. We're out of time here”