Ninety percent token savings, IBM hanging on because nobody can leave, and a Google Ventures partner saying the $100M seed round is basically a trap — Wednesday. This is Tech Podcast Podcast — we’re in the podcasts so you don’t have to be. Today we’re stress-testing a few specific claims: a token-reduction number with an actual mechanism behind it, an enterprise-gravity thesis that cuts against the “incumbents are cooked” take we’ve heard all week, and a VC framework on speed versus velocity that’s either genuinely useful or just conference wallpaper. The bar after this week is pretty high. Let’s see which one actually does the work. Here's Learning Podcasts:
The user-facing promise is context hygiene. Bigger context windows give capacity. RTK tries to stop low-value terminal noise from entering the conversation in the first place. Right. That matters even if your model has a huge window, because context is not just storage. It is attention, latency, cost, compaction pressure, and future confusion.
RTK — Rust Token Killer — is treating shell output like a designed input boundary, and that’s a pretty specific architectural claim. The idea is that when Claude Code or Codex shells out to Git, Docker, Kubernetes, whatever CLI your team built, those tools were built for humans — progress bars, ceremonial headers, five useful lines buried in two hundred boring ones — so RTK sits between the shell and the model and strips that down before it turns into context. The 90% number is the one I want sourced. They’re building from a public GitHub repo and a local snapshot, which at least means there’s something you can inspect — but 90% of what? If your denominator is a Docker pull with the full progress stream, that’s a very different claim than 90% across a real mixed agentic workflow. What I’ll give them is this: they’re asking the right question. What should a command result look like when an LLM is the reader? That’s not prompt engineering dressed up in systems language — it’s a real interface question nobody was asking when these CLI tools were built. Sure, but I need the episode to show me before-and-after token counts from a real agentic session, not just the headline. If they do that — actual numbers from an actual Claude Code run — it’s worth your time. If it’s vibes and a GitHub star count, skip it. From Mark Vigoroso at The Enterprise Edge:
So, so these announcements from these four heavyweights reveal something bigger than just say product launches. They they expose the emerging architecture of the AI era enterprise. And the winners may very well be the companies that control operational context, trusted business data, execution workflows, governance layers, and decision velocity all simultaneously.
EdgeBytes this week is making a structural claim worth taking seriously: IBM, SAP, Workday, Intuit aren’t winning because they’re innovating faster — they’re winning because switching costs are basically load-bearing infrastructure now. Mark Figaroso calls it “operational gravity,” and after a week of hearing that incumbents are getting disrupted, that framing goes the other direction. The “half of operational decisions by 2030” line is doing a lot of work, and it’s coming from Figaroso himself — a founder and CEO — not from an operator with scar tissue from a bad implementation. No methodology, no denominator, no definition of what counts as an “operational decision.” That number is decorative. Fair — but the SAP piece underneath it is more grounded. Christian Klein at Sapphire literally said “almost right just isn’t good enough” and framed the whole push around deterministic execution, not chatbot novelty. That’s a specific positioning call, and it’s aimed straight at the vibe-layer AI that’s been getting funded all year. SAP saying “we’re about precision, not chatbots” is SAP doing what SAP always does — reframing whatever’s happening as something they were already built for. I’d want to see a customer who actually unplugged something because of it before I call that architecture instead of marketing. The Digital Executive Podcast, with Brian Thomas:
Zult regularly shares insights on the real world applications of AI, cutting through the hype to focus on what actually works in production environments. His talks cover topics ranging from agentic support systems to operationalizing machine learning at enterprise scale, making him a trusted voice at the intersection of AI security and customer experience.
Zsolt Balogh, VP of technology operations at Liferay, on agentic AI and enterprise security — and this one’s directly relevant to something we’ve been tracking all week: Dimitri Sirota’s claim from Monday that AI risk lives in the data layer, not the model layer. The setup is promising — actual operator, actual scar tissue, Budapest to Kansas City — but the excerpt we have is almost entirely intro copy. We don’t get to what Balogh actually said about how agentic systems interact with enterprise security posture. Which is the problem, because that’s the exact data point that would tell us whether Sirota’s framing holds up in production or whether this is still a vendor winning an argument. Here's John Wilson at ServiceTitan:
We took the business from 10 to 30 million organically, no acquisitions, and we've gone and acquired these three businesses this year, and we've doubled two of them in the first month. Like the playbook works. So we're just like, "Hey, that's the easy button."
ServiceTitan’s running a segment on acquisition strategy — operator went from 10 to 30 million organically, then bought three businesses this year and says he doubled two of them in the first month by handing them the same playbook. Doubled in the first month is a big claim, and the whole pitch is “it’s an easy button” — which, fine, but that’s a decade of playbook-building compressed into a soundbite. Does the episode get into what’s actually in the playbook, or just the wins? That’s the gap. He’s clear on what he won’t do, which is buy anything that forces him to learn a different service model. The discipline is real. The mechanics of the playbook, though, aren’t in this clip. Rho writes:
People are mistaking speed for velocity. And I think the difference between the two is that velocity invokes direction. Founders who kind of know where the puck is going just around the bend. They see things that other people don't.
KJ Sidberry, GV partner, on Rho — and the hook is speed versus velocity, velocity meaning speed with direction. That’s the kind of distinction that either points to something real or gets printed on a conference slide and forgotten by lunch. The Rho host did not ask for a concrete example — a founder with speed but not velocity, something you could actually test the frame against. It just floated. Which is the tell. If the frame is real, there’s a name attached to it. A company that burned through a big round moving fast in the wrong direction. Without that, it’s a bumper sticker with GV’s logo on it. Got thoughts on today’s briefing, a story we should be tracking, or a correction we need to hear? Send us a note anytime at techpodcastpodcast at lantern podcasts dot com.
You’ll find links to every story we covered today in the show notes, so if something caught your ear, they’re there for a closer read.
That’s Tech Podcast Podcast for this Wednesday. Thanks for listening. This is a Lantern Podcast.