What's Up with Tech?

Interested in being a guest? Email us at admin@evankirstel.com
Your AI budget isn’t blowing up because someone asked too many questions. It’s blowing up because AI agents turn one request into a chain reaction of inference calls, and the infrastructure underneath can’t deliver fast tokens efficiently at scale. We sit down with Vasanth Mohan from SambaNova to get specific about what changes when AI moves from “chat” to “do,” and why enterprises suddenly feel latency, throughput, and cost as operational constraints.

We break down the difference between training and AI inference, then zoom into the agentic reality: swarms of agents, model tiering, and the constant balancing act between tokens per second per request and tokens per second per chip. You’ll hear how model size drives memory pressure, why batching affects your bill, and how to decide when “fast enough” is actually costing you productivity, especially for coding workflows and other tight feedback loops.

Then we go hardware-deep in plain language. Vasant explains SambaNova’s reconfigurable dataflow unit (RDU), why mapping dataflow graphs to silicon can change performance, and how disaggregated inference splits prefill from decode. We also talk about GPUs working alongside other chips, extending the life of existing deployments, and why energy efficiency and air-cooled data centers matter for sovereign AI and global rollouts. We close with a reality check on benchmarks, “benchmaxing,” and why private evals and testing on your own workloads are the only trustworthy path.

If you care about enterprise AI, AI agents, inference optimization, and the real economics of tokens, hit subscribe, share this with a builder on your team, and leave a review with the toughest inference problem you’re trying to solve.
Support the show
More at https://linktr.ee/EvanKirstel

Interested in being a guest? Email us at admin@evankirstel.com
Enterprise tech is having a weird moment: innovation is real, but so is the noise. AI shows up in every pitch, “leading” is used in every sentence, and buyers who actually have budgets and deadlines are stuck asking the same question: what’s genuinely worth my time? I’m joined by Rose Ross, founder and chief Trailblazer at the Tech Trailblazers Awards, to talk about how credible startup awards can act as a practical filter for enterprise IT leaders and a meaningful growth lever for founders.

We dig into what makes an enterprise tech awards program valuable instead of performative: clear criteria, a respected panel of judges, and an emphasis on proof over polish. Rose shares how the Tech Trailblazers Awards grew into its 15th edition across categories like cybersecurity, networking, telecoms, containers, sustainability, AI in a B2B context, plus newer frontiers like robotics and quantum computing. We also talk about why customer impact matters more than “sexiness,” and how buyers can use awards as a shortlist without outsourcing their judgment.

If you’re a founder or operator, we get very specific on the practical side: who can enter (nine years or younger and not beyond Series C), the Firestarter subset (three years or younger with no Series A), how to think about the entry like an investor pitch deck, and what judges hate most. The simplest advice wins: answer the question, drop the vague buzzwords, and back every claim with evidence.

Subscribe for more founder and enterprise tech conversations, share this with a builder or buyer who needs a better filter, and leave a review with the best buzzword you never want to hear again.
Support the show
More at https://linktr.ee/EvanKirstel

Interested in being a guest? Email us at admin@evankirstel.com
Frontier AI is changing the rules of security, and enterprises feel it first. From the floor at VMware Explore, we sit down with Purnima Padmanabhan
Vice President and General Manager, Tanzu Division, to unpack what customers are asking for as AI agents move into real systems and real data. The headline: private AI is getting a serious bump, not as a trend, but as a control strategy for regulated industries and anyone worried about risk, governance, and cost. 

We walk through the “threat side” of AI, where advanced models can amplify attacks and push open source security to a breaking point. Purima explains why so many enterprise applications depend on open source, especially Spring and Java, and what happens when vulnerability reports spike. We also dig into TrueSource by Broadcom, the idea of a secure software supply chain for curated open source packages, libraries, and images that enterprises can trust and operationalize. 

Then we shift to the “opportunity side”: moving beyond chat and RAG into workflow automation with enterprise AI agents. That raises hard questions about sandboxing, limiting autonomy, and keeping agents secure. Just as important, we talk about why data projects slow everything down, and why governed access to the right data is what makes agents useful at scale. 

Finally, we get practical about developer experience and SDLC automation, including Purima’s example of running a swarm of 21+ agents across planning, Jira creation, coding, and testing, plus the caution that not all AI-generated code is equal. If you’re building on VMware Tanzu, thinking about private cloud AI, or trying to balance security, sovereignty, and TCO, this is for you. Subscribe, share with a builder on your team, and leave a review. What would make you trust an AI agent in production?
Support the show
More at https://linktr.ee/EvanKirstel

Interested in being a guest? Email us at admin@evankirstel.com
Agentic AI is showing up everywhere, and it is not waiting for your security program to catch up. From the VMware Explore floor, we talk with Umesh Mahajan, General Manager for Broadcom’s Application Networking and Security Division, about what changes when AI agents become real workloads that talk constantly, call tools through APIs, and evolve by the week. The headline is uncomfortable but useful: AI helps defenders ship faster, but Frontier AI also helps attackers find vulnerabilities and drive more zero-day behavior than many teams are ready to handle.

We dig into the security model that holds up when agents are dynamic and mistakes travel fast. Umesh explains why you have to solve a dual problem: protecting agents from attacks while also preventing a compromised or misdirected agent from attacking others. That leads to concrete priorities like lateral segmentation, distributed firewalling, IDS/IPS, and NDR-style behavioral detection, plus network-level controls to reduce the risk of sensitive data exfiltration. We also talk about discovery, because shadow AI is already a reality. If you cannot identify which agents, MCP tools, servers, and models are running, you cannot confidently separate authorized usage from risky sprawl.

Finally, we tackle the fear every operator has: security that slows the business down. We cover why performance and latency matter more for AI workloads, how enforcement close to the hypervisor can reduce overhead, and what it takes to deploy protection quickly so “bought but not deployed” does not become your weak spot. If you’re building enterprise AI on VMware infrastructure, this is a practical map for getting safer without stalling momentum.

Subscribe for more VMware Explore conversations, share this with the person owning AI rollout, and leave a review with your biggest question about securing agentic AI.
Support the show
More at https://linktr.ee/EvanKirstel

Interested in being a guest? Email us at admin@evankirstel.com
AI is making decisions that shape real lives, yet most people cannot see how those decisions get made. We sit down with Scott Zoldi, Chief Analytics Officer at FICO, to unpack what “trustworthy AI” actually requires when the stakes include fraud, credit risk, and customer outcomes in heavily regulated financial services. If you have ever wondered why black box models create so much fear and backlash, this conversation puts clear language around the real issues: data provenance, explainable AI, ethical testing, robustness, and the ability to audit a decision after the fact. 

We go beyond buzzwords and get specific about AI governance. Scott explains why responsible AI starts with a shared model development standard, so a large organization is not running a hundred different approaches that no one can consistently defend. We talk about why monitoring is often the weakest link in real world machine learning, when to retire models that drift, and why enterprises need to stay in control instead of outsourcing critical decisions to models they did not build. 

Then we dig into a practical enforcement mechanism: coupling AI governance with blockchain to create an immutable record of requirements, testing, verification, and release decisions. Think of it as an operating manual that travels with the model and can be inspected years later by regulators, customers, or internal teams. We also look ahead to what changing regulation could mean, including the push toward interpretable models, trust scoring for generative AI, and focused language models or small language models built for narrow tasks with auditable data. 

If you care about responsible AI, AI transparency, and building systems people can actually trust, hit play, then subscribe, share this with a friend who works in AI or compliance, and leave a review with the one governance rule you think every model should follow.
Support the show
More at https://linktr.ee/EvanKirstel