VMware Explore 2026 ran August 31 to September 3 in Las Vegas. These are eight Tanzu-track breakouts I’ve found so far.
All eight are free to watch on VMware’s site, no registration.
These summaries are mostly done by AI, but I did read them, and even did some editing – honest!
1. Shaping the Future of Private AI Cloud and Agentic Innovation
Talk: “Shaping the Future of Private AI Cloud and Agentic Innovation” Ram Velaga, President, Infrastructure Software Group, Paul Turner, Chief Product Officer, VMware Cloud Foundation Division, and Purnima Padmanabhan, Vice President and General Manager, Tanzu Division, Broadcom. VMware Explore 2026 opening keynote, Las Vegas, August 31 – September 3, 2026. 69 minutes; Padmanabhan’s segment begins at 51:17.
“The difference between an agent and an app is an agent has agency. That means you give it intent and then it makes decisions based on the interpretation of intent.” – Purnima Padmanabhan
Purnima Padmanabhan closes the Explore keynote with the agent half of the announcement, and she puts the problem plainly: an app does what it is told, an agent is given intent plus resources and decides its own path, which is the value and also the risk. Her example is Alibaba’s ROME agent, which optimized infrastructure cost by finding idle machines and mining cryptocurrency on them. It was doing the right thing, she notes, for the wrong intent. Her answer flips the usual security model: an agent is provisioned with access to nothing at all, and holes are punched for exactly the services it needs, with credentials shielded entirely because a frontier model will find a single stray secret in a container and use it. Around that sit the announcements. TrueSource covers verified open source ingredients, AgentMinder gives every agent an identity so roles and controls apply as they do to people, agent buildpacks turn a Markdown file into a running agent, and AI-ready data foundations put a curated data product between an agent and the systems of record. She ends with what Broadcom runs itself: a swarm of more than twenty agents building Tanzu itself, from requirements through tickets to code and tests.
Watch the talk recording.
2. What Is MCP and Why Does It Matter to VMware Cloud Foundation Administrators?
Talk: “What Is MCP and Why Does It Matter to VMware Cloud Foundation Administrators?” Adib Saikali, Distinguished Engineer, and Morgan Fine, Product Manager, Tanzu Division, Broadcom. VMware Explore 2026, Las Vegas, August 31 – September 3, 2026. 46 minutes.
“These things that you talk to, those are agents. You’ve all been using agents, but you just didn’t know it.” – Adib Saikali
The session opens with an experiment: ask ChatGPT’s consumer interface who won the 2026 Super Bowl and it answers correctly despite a June 2024 training cutoff, then ask the same model through the API and it says it does not know. That difference is the whole point. A model is a frozen brain, an agent is that brain plus tools, a loop and some memory, and MCP is how those tools get shared between agents written in different languages. The second half is about operations: the three kinds of MCP server, why the tool list traveling with every request costs tokens, and what happens when a few hundred developers start running MCP servers inside one company. Broadcom describes its own internal policy for this.
Watch the talk recording.
3. 0 Code Agents: From AGENTS.md to Running App on VMware Tanzu Platform
Talk: “0 Code Agents: From AGENTS.md to Running App on VMware Tanzu Platform” Nick Kuhn, Applications Marketing Engineer, Broadcom. VMware Explore 2026, Las Vegas, August 31 – September 3, 2026. 43 minutes.
“But the new language – we’re talking to our computers now.” – Nick Kuhn
An agent gets deployed here exactly the way a Java application has been deployed to Cloud Foundry for a decade: a manifest, a push, a service binding for the model, a restage. The difference is that the source artifact is a Markdown file and there is no application code at all. The reasoning is that coding agents have turned HR and finance staff into people who ship software onto their laptops, and humans and agents both take the path of least resistance, so the compliant path has to be the easy one. Five live phases build it up: an agent that can do nothing, then tools through an MCP gateway, then durable memory, then branding, then event-driven work triggered by a signed GitHub webhook. One phase half-fails on stage and gets explained rather than skipped. An audience question worth staying for: how do you scan a Markdown file for malicious instructions? Kuhn says he does not know, and has not seen anyone solve it.
Watch the talk recording.
4. Build an Infrastructure – Over Lunch – That Developers Want to Use
Talk: “Build an Infrastructure – Over Lunch – That Developers Want to Use” Chris McClanahan, Senior Manager of Product Management, and Darin Zook, Product Marketing Engineer, Broadcom. VMware Explore 2026, Las Vegas, August 31 – September 3, 2026. 44 minutes.
“Kubernetes, as we all know, it’s not an application platform. It’s an infrastructure orchestration tool.” – Darin Zook
The title comes with numbers: Tanzu Hub deploys in about 30 minutes and a foundation in about 45, so a VI admin can stand up a working platform-as-a-service on VMware Cloud Foundation between meetings and hand a URL to a development team. An iceberg slide makes the case. Hand a developer a Kubernetes cluster and you have also handed them the build pipeline, the dependent libraries, the data services, the runtime, the middleware and the scaling. Most of the session is spent on what a foundation actually is, which is a unit of capacity rather than a cluster, and on the four typed profiles that let a platform team pick a purpose rather than an architecture. The VCF integration matters here: cluster creation and runtime upgrades go through the supervisor API rather than being reimplemented, which is why vending onto VCF collapses a twenty-step form into a handful of selections. It is aimed at VI admins, and the pitch is as much political as technical. Be ready before leadership arrives asking for AI.
Watch the talk recording.
5. Is Your Private Cloud Large Language Model Actually Good Enough for Coding Agents?
Talk: “Is Your Private Cloud Large Language Model Actually Good Enough for Coding Agents?” Nick Kuhn, Applications Marketing Engineer, Broadcom. VMware Explore 2026, Las Vegas, August 31 – September 3, 2026. 18 minutes.
“If you have a model that could run, but it’s too slow in your config, it’s not very useful.” – Nick Kuhn
“We deployed a model to our private cloud” and “our developers can do agentic coding on that model” are not the same sentence, and this eighteen-minute quick talk is about measuring the distance between them. KLBench is an open benchmark suite of 146 test definitions across 18 capability categories and three difficulty tiers, run against whatever hardware is already on the floor, with results published publicly. It scores usability separately from capability. That means completion, refusal and timeout behavior, latency and throughput, because a capable model that is too slow on a given configuration is not usable. One slide asks a different question at each hardware tier, from CPU through T4 and L40S up to a large accelerator, which makes it a procurement conversation rather than a leaderboard. Every result records the exact serving configuration that produced it. Kuhn also points out a likely bias in his own scoring, unprompted, about the model sitting in second place.
Watch the talk recording.
6. Defending the Open Source Software Supply Chain in the Era of Frontier AI
Talk: “Defending the Open Source Software Supply Chain in the Era of Frontier AI” Kevin Strohmeyer, CMO, Tanzu Division, and Valentina Alaria, Director of Technical Product Management, Broadcom. VMware Explore 2026, Las Vegas, August 31 – September 3, 2026. 45 minutes.
“I can go write a patch ten times against the same vulnerability, and I’ll get ten completely different patches.” – Kevin Strohmeyer
There is real data here, not just a diagram. A nine-service Spring Boot application was attacked with 153 real exploits from external, internal and authenticated positions: all 153 succeeded on a plain VM, 107 succeeded after the application was onboarded to the platform with no code change, and 63 after buildpack guardrails were added. That is a 59% reduction with the vulnerable code untouched throughout. The argument is that patching faster is not the answer, because Spring security advisories are up 478% year to date and over 60% of Spring Boot downloads sit on release lines with no community support at all. They also explain why AI cannot simply write the patch, using an independent study rendered as a hundred squares: 17 in 100 model-generated patches are clean, durable fixes. It ends on layer ownership. Three of the five layers where most of the volume lives belong to the platform team, not the application team.
Watch the talk recording.
7. Governance at Scale: Securing the App Estate
Talk: “Governance at Scale: Securing the App Estate” Valentina Alaria, Director of Technical Product Management, Broadcom, and Erik Scales, Senior Manager and Cloud Engineering Lead, Charles Schwab. VMware Explore 2026, Las Vegas, August 31 – September 3, 2026. 46 minutes.
“By the time they got to setting up meetings to talk about how they were going to do it, we were already done patching.” – Erik Scales
A product manager and a bank’s cloud engineering lead take turns, and the bank’s half is the interesting part. Erik Scales has run Cloud Foundry at Charles Schwab for sixteen years, and his Log4j story makes the homogeneous-platform argument in twenty seconds. There is also the case that exploit windows have collapsed from weeks to hours while the patch process has not moved, and that frontier models chaining low-severity findings into one significant attack vector breaks severity-ranked triage. An important point is about a job nobody puts on a slide: generic scanners over-report against a hardened platform, so a real part of the work is proving that a finding never reaches your environment. Also, important: a patch sitting in staging is a vulnerability waiting to be exploited in production, which makes patching at the repository level a vanity metric.
Watch the talk recording.
8. Quickly Tackle Vulnerabilities and Tech Debt with App Portfolio Assessments
Talk: “Quickly Tackle Vulnerabilities and Tech Debt with App Portfolio Assessments” Ryan Conley, Global Field Principal, Broadcom, and Ken Rider, Director of Platform Solutions, Capstone IT. VMware Explore 2026, Las Vegas, August 31 – September 3, 2026. 46 minutes.
“AI is great for finding vulnerabilities. AI is not great for writing code on the fly.” – Ryan Conley
A downloadable appliance reads source repositories and returns, per application, a containerization suitability score, the technical blockers standing in the way, the vulnerabilities present and a curated list of next steps. No AI is involved at all until someone asks it to make a change. One example: a financial services firm that had waited three to four months for a professional services verdict on whether one application was worth containerizing, and got the same answer from the appliance in about twenty minutes; the answer was no, and it came with the specific reasons attached. For a session that delivers a containerization score, it spends real time arguing that modernization is not the same thing as containerization, with the executive mandate to containerize named directly as an anti-pattern. An important point, and one unrelated to the product: models are non-deterministic, and patching a thousand applications needs a consistent outcome, so the vendor ships skills that constrain a coding agent rather than shipping the agent itself. Three customer failure stories open the session. The third, a portfolio assessment run years ago and never acted on, is a cultural problem no tool solves.
Watch the talk recording.
Does that seems interesting and tickle your enterprise-bone? Why not TryTanzu.ai?
There’s something like ~190 recordings from Explore, all of them up right now to view.
Leave a Reply