The Circuit Breaker

Why would NVIDIA pay 86x revenue for Hugging Face? It's not a scale play. It's a move to keep the AI market open and stop inference revenue from piling up inside a handful of hyperscalers.In this episode of The Circuit Breaker, Nati Shalom sits down with Boris Renski, co-founder of Mirantis and now founder of Hypershell, to unpack NVIDIA's ecosystem strategy. They compare it to the OpenStack era, when the industry tried (and failed) to stop the cloud oligopoly, and ask why this time might be different.We cover:→ What NVIDIA actually bought with Hugging Face, and why neo clouds are key to the plan→ Why OpenStack fell short, and what Kubernetes and Android got right→ Open source vs. lock-in: is NVIDIA's "open" play really neutral?→ Who is the "Samsung" of the inference world?→ OpenShell: why enterprises need an MDM-style runtime for AI agents→ Will AI kill us all? Boris's surprisingly optimistic take

Join The Circuit Breaker Community- https://www.linkedin.com/groups/32760101/

The answer is uncomfortable: the better the AI gets at reading your logs, the more logs your developers will send it.

Bolt an agent onto an observability platform and you haven't changed the cost structure — you've given everyone a reason to feed it more. The storage bill was always the bill.

Anthony Critelli is a career firefighter and an independent consultant who spent over a decade in tech, so he's seen both sides of the fire drill. Nati sits down with him to ask what's actually happening inside customer environments, underneath the AI pitch decks.

The conversation gets to a question most observability tools quietly avoid: when you're investigating an incident, would you rather have all the evidence, or a witness?

Today we ship every log to a remote lake and reconstruct what happened in retrospect. Anthony's point is that you don't need the witness to carry pieces of the car back to the crime lab. You ask them what they saw.

Also in this one: why DevOps is an ideology and not a job title, why "you build it, you run it" is really "until it breaks," why an agent still needs a deterministic substrate underneath it, and why a 90% accurate agent walking a ten-step rollback is not a 90% accurate rollback.

Everything on the dashboard is green - and your customers are still leaving. Why?

In this episode of The Circuit Breaker Podcast, Nati Shalom sits down with Joe Crawford, Head of Solutions Engineering at Glassbox, who spent 17 years running enterprise monitoring and observability at Morgan Stanley and Citi before moving to the other side of the stack: watching real users instead of servers.

We dig into the blind spots that traditional observability can't see, why MTTR is the metric that quietly decides your revenue, the organizational silos between SREs, developers, frontend and backend — and what actually has to be true before you hand production over to an AI agent.

Joe's take on autonomous SRE is refreshingly unromantic: ~80% of SRE work is repeatable and automatable, but 90% accuracy in a mission-critical system is closer to a liability than a feature. And remediating fast without handing over context isn't a fix — it'sa rollback button that destroys the evidence.

🎧 What we cover:

• Why synthetic monitoring and RUM solve different problems
• Dead clicks, form validation errors, and accessibility issues as

• The real cost of high MTTR: churn, revenue, and the customers who

• Assisted vs. autonomous: the verification tax of 90%-accurate agents
• Remediation vs. resolution — and why context handoff is the whole game
• Why topology and dependency awareness have to come before the

• The converging worlds of system observability and user observability

👤 Guest: Joe Crawford, Head of Solutions Engineering, Glassbox

🎙Host: Nati Shalom

Join The Circuit Breaker Community-https://www.linkedin.com/groups/32760101/