Joe Beda has been on the frontlines of a changing industry for many years. From shipping Internet Explorer on physical CDs at Microsoft in the mid-'90s, to helping stand up Google Compute Engine and Google Cloud Storage, co-creating Kubernetes, founding Heptio and selling it to VMware, then walking away from the industry entirely, he's seen it all (or at least, seen a lot). 

In this episode, Joe joins Matt Klein to unpack what it actually felt like inside Microsoft's hyper-aggressive culture versus Google's passive-aggressive one, why Google, despite superior engineering, lost ground to AWS's relentless customer focus, and the battle to create an internal engineering bubble in the face of bureaucracy. Joe also opens up about burnout, stepping away after VMware, and what brought him back as CTO of Stacklok, where he's now building on-prem MCP governance and LLM gateway infrastructure for regulated enterprises. They close on where MCP, CLIs, and token economics are headed, and whether AI progress looks more like Moore's law or a leveling S-curve.

Charity Majors, co-founder and CTO of Honeycomb and co-author of Observability Engineering and Database Reliability Engineering, sits down to talk about how AI is forcing a rethink of observability, engineering practice, and the "three pillars" model she's spent years pushing back on.

Charity traces her path from an off-grid childhood in rural Idaho to a piano scholarship, San Francisco startups, Parse, Facebook, and eventually starting Honeycomb - by her own admission, partly to get out of burnout. She gets into why "work is family" is the wrong frame, why she took Intercom's 2X productivity challenge seriously enough to issue a rare mandate at Honeycomb, and why she's tired of the triumphant vibe coder discourse.

On AI: the three pillars made sense when infra teams operated code they didn't write and couldn't change. Now agents are writing, shipping, and debugging that code, and Charity argues it's forcing a rare reset, a chance for the median engineering team to finally get feedback loops only the top few percent have ever had.

Dillon TenBrink, distinguished engineer at The Home Depot, joins Matt to trace an unusual path into the industry: farm kid from central California, no college, straight into Bay Area startups at 19, then dot-com bust, seven or eight years inside a regulated power utility, and now 15 years at Home Depot. Along the way he picked up networking as a hands-on apprentice at Nortel, six months of a senior engineer teaching him BGP and LACP on the whiteboard.

The back half gets into what technology actually looks like at retail scale: 2,300 stores across North America, an edge platform Dillon's team built that runs in the back of every store, and the integration work that pulled disparate data sources into a single view of customer, catalog, and SKU. His observability principle is the sharpest through-line: bandwidth is finite, so summarize at the edge and ship only what matters, and apply the same discipline to cloud. "The metric has no value if it's just noise." He and Matt close on AI opening prototyping to people who don't write code, with Dillon's line "today's vibe code is tomorrow's legacy" landing as the hot take.

Cedric Beust, creator of TestNG and JCommander, one of the early members of the Android team, and the engineer behind the original Android Gmail app, joins Matt to trace 20+ years of mobile from the inside. He walks through building Gmail for J2ME feature phones before Android existed, what Google actually acquired when it bought Android (concepts and Sidekick experience, not a working OS), why Java won as the platform's language, and the Eric Schmidt/Steve Jobs dynamic playing out as the iPhone loomed.

The back half moves through LinkedIn, a startup that LinkedIn later acquired, Yahoo, and his current role as a mobile architect at JPMorgan Chase. The throughline: crash rates, performance, and observability are the same problems he was solving 20 years ago, now at a different scale and with real regulatory stakes when a banking app breaks. He and Matt close on AI, where Cédric uses it for the parts of the job he finds boring while worrying out loud about what happens to engineers who never learn to write code by hand.

Aaron He, Director of Engineering for Tinder's Client Platform team, joins Matt Klein to trace a decade of mobile engineering at scale, from writing Xamarin apps in C# to native Android, and into today's debates over React Native and Kotlin Multiplatform.

They talk through a database downgrade that crashed the app during Aaron's first week at TrueCar, and a more recent bug where a hung flag lookup silently dropped analytics events: no crash, no performance hit, and no way to backfill afterward. From there, they turn to performance at Tinder today: crash rates are strong (99.5 to 99.7% crash-free), but app hangs and ANRs are the bigger challenge. Aaron breaks down how his team attributes regressions to the right team and experiment.

Aaron also shares how he's built leadership trust on long-term infrastructure bets, and what he learned managing up during a CI consolidation debate with the backend team. He passes along a rule of thumb for using AI without losing the learning: write something once with AI to move fast, once yourself to actually understand it, and a third time to get the architecture right.

In this episode of Beyond the Noise, Matt talks with Nicki Stone, a Principal Engineer who oversees development of the Amazon Shopping App. Nicki's path there runs through a finance desk where she taught herself to code out of boredom, one of the first coding boot camps in existence, a quarter-million-dollar hackathon win, and a startup that raised $1.8 million before an abrupt close.

Nicki built a framework at AWS that unioned on-device and cloud ML models years before anyone was talking about edge inference, and she and Matt dig into why that architecture is only going to get more relevant as models keep shifting to the device. She also pulls back the curtain on what engineering at Amazon's scale looks like: 50 separate networking stacks that nobody shares, no monorepo, and why adding a single request header becomes a company-wide problem. Nicki explains the bet she's making to fix it.

Plus: what she found under the hood at Instagram, Meta's code culture, and why AI makes her feel like "a kid in a candy shop" after 12 years in the industry.

In this episode of Beyond the Noise, Matt talks with Romain Guy, who spent 18 years as an engineer and engineering leader at Google on Android, joining when the team was around 40 people and the product was still a secret internal project. 

Romain walks through the technical reality of those early days: a VM with no JIT compiler, stop-the-world garbage collection, and a UI toolkit with no GPU acceleration. The conversation covers the bandwidth ceiling that forced Android's UI toolkit onto the GPU, the surprising complexity of text rendering, and the tradeoff behind Jetpack Compose sharing a rendering layer with the old View system.

From the pain of Java Native Interface (JNI) to building Filament, a 3D rendering engine now powering Android XR headsets, the episode shows that performance is less about benchmarking and more about the mindset you bring to every decision.

In this episode of Beyond the Noise, Matt speaks with Ryan Cooke, Director of Engineering at Pinterest, where his teams own the infrastructure that keeps Pinterest's Android, iOS, and web apps running. Nine years, several promotions, and eight teams later, he joined as an Android engineer and never left.

The conversation covers two formative projects: an experiment-driven overhaul of Android image loading that produced outsized engagement gains, and a pre-IPO push to get Pinterest's core metrics accurate enough for public markets: standardizing definitions, building data checkers, and catching silent inflation from sources like Facebook's background auth refresh.

Ryan then gets into managing infrastructure at scale: why Pinterest adopted Bazel for iOS builds and walked it back, how AI agents are unlocking migrations that were never worth funding before, and why the same tool produced completely different results in two different engineers' hands. They close on native versus cross-platform: native by default, except where identical behavior across platforms matters more than anything else.

In this episode of Beyond the Noise, Matt sits down with Miguel de Icaza, co-founder of GNOME, Mono, and Xamarin, and one of the most influential figures in the history of free and open-source software. Miguel takes us back to his early days in Mexico, where a crashing Windows machine and a printout of FTP sites set him on a path toward GNU, Linux, and eventually building one of the first serious open-source desktop environments. He traces the arc from true-believer GNU fanatic to pragmatic entrepreneur: what it felt like to ship a proprietary Exchange plugin and get kicked off the FSF board for it, why pure open-source business models kept hitting the same wall, and how .NET offered a clean slate that stalled Java efforts never provided.

The conversation then shifts to mobile, fragmentation, and what comes after a long career building for other developers. Miguel breaks down the operational nightmare of Android fragmentation versus iOS's relative stability, and why the sheer volume of crashes was enough to shift his platform allegiances for good. He covers Xamarin's acquisition by Microsoft, what a cultural reset inside a famously combative company taught him about management, and his current chapter: a post-Microsoft bet on Godot, a Swift binding hacked together on spare afternoons, and a game engine IDE that half his users are running on their phones.

In this episode, Jay Gengelbach, longtime Engineer at Alphabet and current Software Engineer at Vercel breaks down the story of a hard-to-track-down bug he encountered: a mysterious spike in server errors that hit exactly at the top of every hour (and other “round” time boundaries) across Google Now/Discover’s massive mobile footprint. Jay walks through how the team traced it to battery-saver behavior on certain devices/apps that effectively rationed background connectivity, causing synchronized bursts of requests the infrastructure couldn’t react to in time. The fix wasn’t just “handle more load,” but reshape it: using server-controlled scheduling to create intentional “load divots” so misbehaving clients could spike into empty space. The conversation wraps with Jay’s current work at Vercel scaling CI/CD reliability and speed, plus a measured take on AI: real value, real hype, and a looming reckoning when “free” stops being free.