Software Development, Finance and AI

There are genuine technical ideas - hybrid RAG with BM25, knowledge graph contextual bounding, runtime multi-model selection for cost/fit, and markdown over JSON for agent-to-agent communication - but the episode is diluted by a stock-picking segment, a food-ordering segment, and a ten-minute host monologue. Useful signal is clustered in the first 40 minutes; the back third is almost pure filler.

“we capture that in a knowledge graph that's critically important. That's where the plus comes in.”

“at runtime determine the appropriate model for purpose fit, uh, assignment given the task that's needed and also to ensure that we're not overspending or sending a small focus task to a big uh, model unnecessarily”

The framing of 'hire AI not buy software' and the prediction that mid-managers will be hit as hard as coders - with an 80% reduction figure and a 2028 timestamp - are specific and mildly contrarian. Most other ideas (RAG+knowledge graph, vibe coding risks, AI replacing headcount) are circulating widely and are not argued from first principles here.

“managerial levels, especially mid manager, is hit probably just as hard as coders, which means that maybe 80% of mid managers like director levels, maybe even senior manager levels, may be removed for eight out of every 10. If not today, maybe as soon as 2028”

“we at subatomic evaluate CRM systems. We decided why do we need it? We have AI co workers who can build one for ourselves and we do.”

Karl Simon is a genuine practitioner - CTO and co-founder with 30 years of experience - operating in a specific vertical (wealth management, manufacturing) with paying customers, one of whom became a seed investor. He is not a career podcast guest, though the company is small (13 people) and the scale of deployments is modest, limiting the depth of battle-tested insight.

“we have a client who originally was a pessimistic prospect, turned excited, heavily satisfied customer, uh, and then became our lead investor in our seed round that closed last October”

“we basically eliminated 8,000 hours of labor across, you know, the entire advisor team there and now they can actually grow their business”

The episode offers concrete figures - 8,000 hours of labor eliminated, 13 humans to 100+ AI coworkers, 80% AI-generated code, named internal products (Nexus, Nucleus), and a specific stack (LangChain, LangGraph, LangSmith, Postgres, Python) - but client names are absent, the 8,000-hour claim lacks denominator context (how many advisors, over what period), and the manufacturing example carries no numbers at all.

“we basically eliminated 8,000 hours of labor across, you know, the entire advisor team”

“we're only a 13 human being team, but we have over 100 AI co workers, predominantly in engineering as we have two different internal products. Nexus for data engineering, Nucleus for the workloads”

The host earns credit for genuinely challenging the guest on vibe coding and the continued relevance of CS fundamentals, but squanders significant airtime on stock-pick hypotheticals, a three-course meal question, and a sprawling ten-minute self-monologue that the guest had to sit through. Follow-up drilling on technical claims (e.g. how the knowledge graph is actually built, what accuracy numbers look like in production) is largely absent.

“I'm going to challenge you on that simply because I want to one, play the devil's advocate and two, I've actually had people tell me quite the opposite many times.”

“What would be your order of appetizer, entree and dessert? Be it all does not have to be part of the same cuisine. Just one dish from each of these three categories.”