“In Kubernetes, ‘pod’ refers to both a configuration stored in etcd AND the actual running container.”

In this episode of 🌩️Thunder, Abby Bangser (Principal Engineer at Syntasso) and I untangle CRDs, custom resources, controllers, and operators. These terms can get used interchangeably, but they mean very different things.

A CRD is a schema. A custom resource is data in etcd. A controller is what does the work.

Watch and learn! youtu.be/xlBMpLNaP…

Whitney Lee and Abby Bangser standing at a lightboard covered with colorful handwritten text about Kubernetes custom resources, CRDs, controllers, operators, and admission controllers. The 🌩️Thunder show branding is visible.

This is the finished lightboard from the 🌩️Thunder episode on Keycloak. Ever wonder why it’s so easy to plug an app into Keycloak? This board explains the OpenID Connect flow, step by step.

Read the full board notes here: gist.github.com/wiggitywh…

A lightboard covered in handwritten neon notes about Keycloak. Sections include "Before Keycloak," "Identity & Access Management (IAM)," "Log-in Flow," "Installing Keycloak," "Administrator Experience," and "Open ID Connect Flow," the last with a numbered sequence diagram showing arrows between a stick figure labeled User, an Application box, and Keycloak. A boxed callout in the lower left reads "Keycloak is the open source Identity + access management solution."

“Observability is solved.”

In this episode of 🌩️Thunder, Duffie Cooley, Field CTO at Isovalent and CNCF TOC member, explains why Cilium does far more than a traditional CNI.

Cilium attaches an eBPF program to every pod’s network namespace at creation time. That single hook is how it delivers networking, security, and observability at native kernel speed.

Have a look-see! youtu.be/slJP_CS5X…

The Thunder show logo appears in the top left. In the center, bold yellow text reads "Cilium Explained" over a dark chalkboard covered in handwritten notes about eBPF and Kubernetes networking. Below, yellow text reads "EBPF-POWERED KUBERNETES NETWORKING." On the left, a woman with long dark hair with reddish highlights smiles broadly, wearing a black denim jacket, pointing both index fingers toward the center text. On the right, a circular headshot of a bearded man with a white-gray beard wearing a black wide-brim hat. Bottom right text reads "With DUFFIE COOLEY" in yellow.

[Keynote] Cloud Native Days Romania: Ozzie & Nova: A Kubernetes Security Play

Datadog Illuminated: How to Evaluate AI Code against Your Company’s Standards, with Ville Hellman

Datadog Illuminated: How Datadog Detects Malicious Code at Scale, with Christoph Hamsen

Lin Sun, co-creator of kagent, joined me at the lightboard to walk through how it works.

kagent is a framework to help you make AI agents and run them in Kubernetes declaratively, with YAML.

Read the full board notes here: gist.github.com/wiggitywh…

A hand-drawn lightboard diagram on a black background, filled with colorful marker text and boxes mapping out how kagent works. Sections include "Before kagent," "kagent config" with a Kubernetes CRD circle, "kagent controller watches," "Agent Sandbox," "Agent Runtimes," "Why run your agent on Kubernetes," "AUTH," and "How do people use kagent," connected by arrows and underlines in pink, yellow, and green.

“AI’s role in software development is primarily that of an amplifier.”

In this episode of Software Defined Interviews, Nathen Harvey, who leads the DORA research program at Google Cloud, explains what 15 years of research data reveals about AI’s impact on software delivery.

AI amplifies what’s already there, good or bad.

Give it a listen! www.softwaredefinedinterviews.com/118

Three people appear in the thumbnail against the Software Defined Interviews logo. On the left, a woman with long reddish-brown hair laughs and claps her hands, wearing a red top. Upper right, a bearded man with short dark hair wears a plaid shirt. At the bottom center, a man with short gray hair and glasses smiles, wearing a blue floral shirt. An arrow and the text "with guest Nathan Harvey" point to him. The center displays the show logo in bold black and red lettering: "Software Defined Interviews." The sides feature a tiled doodle-style background with clouds, lightning bolts, and checkered patterns.

During my sunrise hike today, I spotted six sneaky Underwing Moths on a single tree.

Can you find them all? ♫

On 🌩️Thunder, Kubescape maintainer Ben Hirschberg shows how the tool catches misconfigurations, then watches runtime behavior with eBPF & turns that into a least-privileged NetworkPolicy proposal.

Read the full board notes here: gist.github.com/wiggitywh…

A completed lightboard covered in handwritten colored notes about Kubescape. Sections include Before Kubescape, Threats in Kubernetes, Kubescape Can Do a Lot (infrastructure scanning, workload config scanning, RBAC configuration, vulnerability scanning with GRYPE, hardening and remediation proposals), a Glossary of security terms, and the Kubescape User Experience covering CLI, Operator, and GitHub Action.

Your GenAI feature returned a 200. The model did indeed return text to the user. But was the answer any good?

In this 🌩️ Thunder episode, Austin Parker (Director of Open Source at Honeycomb) explains why traditional observability falls short for GenAI. With deterministic software, success is binary: it worked or it didn’t. With GenAI, a successful HTTP response tells you nothing about whether users are actually happy.

Austin covers how OpenTelemetry’s semantic conventions let you combine user feedback (thumbs up, thumbs down) with performance data (token counts, model, cost) in the same telemetry. He also talks about “architectural blindness” in coding agents and how feeding trace data back into your agent helps it understand your system’s execution paths.

Watch now → youtu.be/RNaa_48LW…

Thunder episode thumbnail featuring Whitney Lee and Austin Parker. Large stylized text reads 'Making GenAI Observable with OpenTelemetry' with 'with Austin Parker' below. Whitney appears on the right smiling and pointing upward, wearing a plaid shirt. Austin appears on the left wearing a black hat and glasses with a full beard. A lightboard with GenAI observability diagrams is visible in the background. The Thunder logo appears in the top left corner.

Datadog Illuminated: How to Evaluate AI Code against Your Company’s Standards, with Ville Hellman

“I quit my job in tech because I just couldn’t do it anymore.”

In this episode of Software Defined Interviews, Whitney and Coté talk with Lian Li, Cloud Native Expert at Vorwerk Group. Lian Li spent 16 years in tech, quit to pursue stand-up comedy and improv theater, and came back with a completely different take on what makes teams work.

We cover her transition from platform engineering to stand-up comedy, musical theater, and improv theater, how performance art changed the way she thinks about tech teams and conference talks, her improv workshops for engineers at conferences, and the KubeCon karaoke parties she organizes.

www.softwaredefinedinterviews.com/117

Software Defined Interviews podcast cover with hosts Whitney Lee (top left, laughing and clapping, wearing a black outfit, long brown hair) and Michael Coté (top right, smiling, bearded, in a dark plaid shirt) above the show logo with a stylized cloud graphic. Guest Lian Li appears in the bottom right corner, smiling and wearing a blue blazer with teal drop earrings. An arrow points to text reading 'with guest Lian Li.'

Here’s the final board from the Datadog Illuminated episode with AJ Stuyvenberg, who explains how his team rewrote Datadog’s serverless agent in Rust, and the benefits of doing so!

Read the full board notes here: gist.github.com/wiggitywh…

A lightboard titled "WHY WE REWROTE OUR SERVERLESS AGENT." Three sections: left "The Problem Space" notes that AWS Lambda freezes the CPU between requests and runs code in tiny, isolated sandboxes, while the Datadog Agent was built to be long-running with a 55mb binary and 400ms+ startup, shown with a before/after timeline where startup drops from 400ms to 50ms and the agent continuously flushes data to Datadog during I/O waits. Middle "Why Rewrite?" lists making the agent lighter, eliminating upstream bugs, and switching to Rust for no garbage collection, small binaries, fast startup, and memory safety, with a playful "Suspense" note about more reasons to come. Right "Challenges" notes the team had no Rust knowledge, next to a "Results" box showing startup time improving from 400ms to 50ms, binary size shrinking from 55mb to 7mb, memory reduced by 50%, a 2x reduction in tail latency, and a safe rollout shipping Rust and Go binaries together with failover to the old Go version. Chalk-style text on a dark lightboard.

The client doesn’t know or care where the Network Service is running. Read the full board notes here: gist.github.com/wiggitywh…

A lightboard covered in dense handwritten-style text in white and yellow, organized into labeled sections including "Before Network Service Mesh...", "A Network Service is", "Network Service Mesh is", "Use Cases for Network Service Mesh (NSM)", "Network Service Registry", and "Application Service Meshes." Black background with multiple columns of technical notes about K8s networking, Network Service Mesh concepts, and comparisons between application service meshes and NSM.

My Datadog Illuminated episode with Reilly Wood, whose team built the Datadog MCP server from scratch, distilled to a single lightboard. Read the full board notes here: gist.github.com/wiggitywh…

A lightboard titled "BUILDING AN MCP SERVER." A flowchart shows Datadog Backend to Datadog-hosted MCP Server via HTTP to Agent (labeled "WILDCARD") to User to "SUCCESS = HAPPY USER." Below: "TAMING THE AGENT (+ LLM) = Correctness + Efficiency." Three sections: left "Context Efficiency" (data format: JSON crossed out, replaced by YAML nested / CSV flat / TOON; agent control over context: let agents write SQL queries; good tool descriptions: smart yet succinct), center "no tool sprawl" (default core toolset: logs, metrics, traces; opt-in specialized toolset: synthetics), right "guide the agent" (specialized search docs tool with RAG; good error messages: Actionable; good server-level instructions: ex "use search docs RAG tool"). Chalk-style text on a dark lightboard.

If a dev leaves, it is a one-click experience to remove permissions. Read the full board notes here: gist.github.com/wiggitywh…

A lightboard covered in neon handwritten notes on a dark background. The board is divided into sections with RBAC problems, a center architecture diagram showing PARALUS connecting to Kubernetes clusters, and PARALUS MAIN POINTS. Text and arrows are written in neon marker throughout.

WHOMEVER OWNS THE KERNEL SEES ALL. Read the full board notes here: gist.github.com/wiggitywh…

A dark lightboard covered in handwritten colored notes about Confidential Containers. Multiple sections visible including "Confidential Computing Trust Model," "Components of a Confidential VM," and in large letters: "WHOMEVER OWNS THE KERNEL SEES ALL." Diagrams show pod-in-VM architecture and attestation flows.

Datadog Illuminated: How Datadog Detects Malicious Code at Scale, with Christoph Hamsen

Here’s everything we covered on the lightboard when Nick Isaacs walked me through how Datadog’s LLM-powered monolith migration agent works. Fun!

Read the full board notes here: gist.github.com/wiggitywh…

A color-corrected lightboard covered in a handwritten workflow diagram showing the 13-step LLM-powered migration process. Sections include "Gather + Plan," "DO THE THING! Execute the Plan," a Verification Process loop, "Code review by GPT5," "Address discrepancies," and the final steps through "HUMAN REVIEW" to "MERGED & DEPLOYED!!! HUZZAH!!!" The board also shows "Room for Improvement" and "What's Good" columns on the right.