cloud2030

In this episode, we continue our discussion of governed execution for agentic AI and how to prevent agents from taking actions they should not take. We use a maintenance example to examine access, approvals, revocation, rollback, and human checkpoints across systems such as ERP, inventory, HRIS, planning, procurement, and maintenance tools. We also talk about the difference between the agent and the harness around it, and why deterministic workflows are often preferable to direct model action.

Transcript: https://otter.ai/u/QNu4NYGzk7xe1Knv3bzVdXBo7ZE?utm_source=copy_url

This week we discuss the role of a harness in AI systems and why it matters for infrastructure, tool use, rules, and operational constraints when an AI acts on behalf of a person or organization. We also distinguish agency, delegation, authorization, authentication, and accountability, and explain why keeping them separate matters for policy and responsibility.

Transcript here: https://otter.ai/u/vSVbzrW3Eih3mLdT6OsCGRJ2WPY?utm_source=copy_url

In this episode, we are joined by guest Byron Dill from Summit for a great deep dive into virtualization. We talk about virtualization from an enterprise and infrastructure perspective, with a focus on VMware, its recent announcements, and how the market is responding to changes in virtualization strategy. We also discuss VMware’s expanded VCF platform, including Kubernetes and AI-related features, and the contrast between AI marketing and the practical realities of GPU-based workloads. Really great episode!

Transcript here: https://otter.ai/u/wYYQZyzPxrzNF_3MSU4NnPf8zS4?utm_source=copy_url

In this episode, we focus on token cost and token optimization in generative AI, and we discuss why token use needs budgets and controls rather than being treated as unlimited. We also talk about how repeated prompting, testing, and downstream work can quickly increase token consumption. We also talk about where AI agents make sense and where cheaper models or existing data should be used instead, including a bug triage example where systems should gather history and reproduce issues before escalating. We close by discussing responsibility, accuracy, and governance, including marking unverified claims clearly and using strict prompts and review passes. We also note the need for better expectations around safety and liability in AI systems.

Transcript here: https://otter.ai/u/0Y-EBKoKsN7UWxC6YzMkb44ZN6w?utm_source=copy_url

In this episode, we get into current market news, including pending IPOs, hardware investment, and IBM’s AI-related announcement with Red Hat, and what this may mean for open weights models and hybrid AI. We also talk about hyper-compressed and ultra-quantized models, specialized hardware, and how memory, compute, and power limits are shaping what can run on device and at the edge. We examine the enterprise software implications and the economics of AI pipelines, including token budgeting, guardrails, human supervision, and the need for validation, governance, and operational control.

Transcript: https://otter.ai/u/z-PbDS6v8w_4n2EeeyDPndqkJD8?utm_source=copy_url

This week we have an exciting conversation about harnesses, neoclouds, open weights models, and the current AI infrastructure landscape. We discuss bare metal automation, agent-friendly APIs, CPU compute, GPU economics, and why some providers are moving up the stack toward models and APIs.

We also look at the harness as an execution, optimization, and governance layer, including tools, sub-agents, routing, and safety controls. The conversation covers smaller and fine-tuned models, continuous validation, model drift, governance, security, compliance, and AI sovereignty, as well as sandboxing and air-gapped execution for agentic systems.

In this episode, we discuss Red Hat’s enterprise AI work, including OpenShift AI, safety guardrails, and router-based model selection. We also talk about smaller AI systems, long-term memory, and the limits of frontier models in enterprise workflows.

We cover governance, forensic traceability, and human review for regulated AI systems.

Transcript: https://otter.ai/u/aJ7q0qOyhhhiDRkdaJcLrOY6R7Y?utm_source=copy_url

We discuss the acquisition of Mirantis by the neocloud company IREN, and use it to examine the current market for infrastructure and AI deployment. We look at Mirantis’ consulting business, Kubernetes distribution, and its earlier acquisition of parts of Docker, including the registry and enterprise components.

We also consider whether trusted registries could play a role in distributing agents and tools for AI systems, and we compare this idea to package management. We get into an interesting discussion on neoclouds and managed agent providers as part of the operational layer for running AI workloads, including GPU capacity, model hosting, and repeatable deployment.

Transcript: https://otter.ai/u/ZBtX6hwwHIqqOBbLDiJwM66wFuM?utm_source=copy_url

In this episode we discuss whether AI and LLMs are changing the role of open source, including the pressure on maintainer review queues and the growing appeal of private forks and internal maintenance. We also talk about the security tradeoffs of those forks, especially around visibility, dependency tracking, and audits. We also get into how domain expertise, documentation, interfaces, and operational knowledge may become more important than the code itself, and why documenting archival knowledge and provenance matters.

Transcript: https://otter.ai/u/h61DNW_fV-5QrNBepX7NZvq8dCU?utm_source=copy_url

In this episode we work through a Vibe Coding Operations session using Claude AI inside Digital Rebar as an analysis tool. We describe a three-phase flow of plan, execute, and analyze for inspecting a machine, gathering command output, and reviewing the results. We also talk about troubleshooting an installer problem, differences in user account flag behavior, and improvements in Claude’s debugging across sessions.

Transcript: https://otter.ai/u/-wk5QfPFPgYOxm_dOvTSIezDwTs?utm_source=copy_url