Name the Workload
Everybody's K5
The Destination Is The Origin
Nobody Asked
In That Room
Stop Being Surprised
Distribution Unlocked
A satellite-shaped pitch arrives at the desk of an industry that already has more compute than it knows what to do with. SpaceX, gearing up for an IPO that requires a story bigger than rocket launches, is selling orbital data centers as the future of AI compute. Google is reportedly listening. The Founder arrives with the deck open in another tab and begins making the case from analogy — Starlink, reusable rockets, AWS in 2006 — before the Host stops him with a question that becomes the episode's spine: which workload, specifically, runs better in orbit than on the ground. The Founder names training. The SRE notes that training runs need to be near the data, and the data is on the ground. The Founder names batch inference. The DBA asks what data, stored where, served to whom. Nobody answers. Across the runtime the panel surfaces the actual constraints — radiative cooling in vacuum being harder than the pitch claims, latency floors that physics enforces, single-event upsets from cosmic rays, the impossibility of swapping a failed disk in low Earth orbit, the launch-schedule economics that replace traditional capital expenditure. The Founder eventually names a real constraint: terrestrial data centers are running into power, water, land, and permitting walls that orbit doesn't have. The SRE acknowledges this as the first real answer and explains why the solution isn't orbit but grid capacity. The DBA confirms what the episode has been demonstrating: the orbital pitch is what you do if you've decided in advance that orbit is the answer and you're working backward from there. The episode lands on the question of how a decision like this gets made in the first place — somebody walked into a room with a slide that said "orbital compute" and the room said yes, without anyone in the room asking what workload it was for. The DBA's closing structural verdict ties the episode together: the architectures that worked, somebody could always name the workload in the first meeting. Nobody in this article has named the workload. Everything downstream of that — the launch costs, the cooling problem, the radiation, the maintenance impossibility — is consequence. Source article Report: Google and SpaceX in talks to put data centers into orbit — Rebecca Bellan, TechCrunch · May 12, 2026 Panel The Startup Founder The Burnt-Out SRE The Database Administrator The Goat Farmer's Counsel
The Linux kernel retired support for the AMD K5 this week — a chip that failed commercially in 1996 and hasn't existed in production since the Clinton administration. The panel sits with the question of what 'we still support this' was actually doing for thirty years, and who it was doing it for. This episode takes the Linux kernel's removal of non-TSC code paths — and with them, support for the AMD K5 and several Cyrix parts — as an entry point into something much larger: the way long-lived systems accumulate maintenance debt that nobody is tracking, on behalf of users who stopped existing decades ago. The Legacy Sysadmin opens with firsthand context. He bought K5s in 1997 because they were cheap, regretted all of them, and replaced them with K6s by 1998. The chip wasn't beloved. It was a commercial failure AMD shipped because they had to ship something against the Pentium. The kernel carrying support for it until 2026 wasn't loyalty to a meaningful platform — it was inertia compounding quietly for thirty years. The DBA reframes the actual technical decision: the kernel didn't retire the K5 specifically. It decided to require the TSC — the timestamp counter — as a baseline assumption, and the K5 happened to be on the wrong side of that line. The headline says 'K5 retired.' The reality is 'we finally allowed ourselves to require a 1995 instruction.' The K5 is collateral. The cost of carrying the non-TSC fallback code for thirty years was real but distributed — a small tax on scheduling code, on timer refactors, on anyone who had to remember the K5 case — none of which appeared on a dashboard or triggered a postmortem. The Startup Founder argues the kernel team made a hard call and that engineering cultures can be built around deprecation discipline. The DBA's response is direct: Stripe is fourteen years old. Talk to me about their deprecation hygiene in twenty years. Everyone gets their K5. Source Article "AMD K5 CPUs The Latest To Be Retired With Linux's Aging & Stagnate Hardware Support" By Michael Larabel, Phoronix, May 8, 2026. https://www.phoronix.com/news/AMD-K5-CPUs Panel The Legacy Sysadmin The DBA The Startup Founder The Goat Farmer's Counsel
Gartner published a report this week arguing that for some enterprise workloads, migrating to a mainframe is now cheaper than continuing to license VMware. Simon Sharwood covered it at The Register. The math is real — Broadcom's post-acquisition pricing has pushed VMware total cost of ownership past IBM Z for a non-trivial set of workloads, and Gartner is on the record telling clients to do the comparison. The Legacy Sysadmin opens with a thirty-year arc: helped a bank get off MVS in 1999, helped them standardize on VMware in 2011, and is bracing for the call asking about a Z migration in 2027. The Burnt-Out SRE sketches the eighteen-month post-mortem in advance — half-migrated workloads paying for both platforms, the CFO who approved the project moved to a different company, the new CFO commissioning a study that will recommend migrating back to cloud. The DBA, in his third panel appearance, is contemptuous of TCO models that don't include the cost of senior database engineers quitting because they got tired of being the only person in the room who could read an explain plan. The episode is about whether the loop completing — mainframe to Unix to VMware to mainframe — is a strategy or just the average tenure of a CIO playing out across infrastructure. The panel concludes, with notable unanimity, that institutional memory does not survive long enough to prevent its own repetition. Source Article "Mainframes are now cheaper than VMware, says Gartner" By Simon Sharwood at The Register, May 4, 2026. https://www.theregister.com/on-prem/2026/05/04/moving-to-a-mainframe-can-be-cheaper-than-vmware-gartner/5229237 The Panel The Legacy Sysadmin The Burnt-Out SRE The DBA The Goat Farmer's Counsel.
Google Chrome, in a recent update, silently installed a four gigabyte local large language model on every machine running the browser. The model ships as part of a feature called Gemini Nano. There was no prompt at install time, no notification afterward, and an opt-out that exists if you know where to look and what to uncheck. The Register covered it. The Legacy Sysadmin places the move in a thirty-year arc of installer-bundled software, from late-90s browser toolbars through to today — same pattern, larger payload. The Paranoid CISO declines to reach for nation-state attribution and instead names the actual issue: Google has built a new high-bandwidth software-delivery channel into a billion endpoints, and the public conversation about how that channel is secured should have happened before the model shipped, not after a journalist noticed his disk was full. The DBA, in his second appearance on the panel, identifies the toggle itself as the confession — opt-outs are what you build when you've already made the decision and want to deflect the consent question. The episode is about what happens when a vendor with root-level access to a billion machines decides, in a meeting the public was not in, that the right thing to do is allocate four gigabytes of unprompted storage on devices it does not own. And about whether the existence of an opt-out is a feature or an admission. Source Article "Google quietly slips four-gigabyte AI model into Chrome" By Liam Prove, The Register, May 7, 2026. https://www.theregister.com/ai-and-ml/2026/05/07/chrome-silently-installs-a-4-gb-local-llm-on-your-computer/5230893 Panel The Legacy Sysadmin The Paranoid CISO The DBA The Goat Farmer's Counsel
IBM updated its Db2 Genius Hub on May 5th with new support for Google Vertex AI and Intel Gaudi, on top of existing integrations with Amazon Bedrock and watsonx, plus a newly added Microsoft Azure AI Foundry. The pitch is what IBM calls "Level 3 automation" — AI agents that can propose and execute database operations on a DBA's behalf, with user approval. 43% of Db2 customers are banks. The Register covered it. The pitch comes with a quote from Sanjeev Mohan, a former Gartner analyst now working as an independent advisor, given directly to IBM: DBAs should be "upskilled," sitting in rooms with "business decision makers," while Genius Hub handles the "nitty gritty, nuanced, heavy lifting of the database." The DBA, in his first appearance on the panel, takes the analyst quote apart sentence by sentence. The Legacy Sysadmin places the IBM pitch in a thirty-year arc of identical pitches aimed at mainframe operators, L1 sysadmins, and L1 SREs in their respective decades. The Startup Founder defends the framing on the grounds that AI handling routine work elevates senior practitioners — and is asked to name a single DBA he's worked with who became a CDO. The Goat Farmer's Counsel interjects from the back pasture with reasons. The episode is about the recurring industry move of selling automation to a profession by promising to elevate it, and the predictable gap between the pitch and the outcome. Source article: "IBM asks DBAs to trust AI to act on their behalf" By Lindsay Clark at The Register, May 5, 2026. https://www.theregister.com/2026/05/05/ibm_db2_automation_google_intel/ The panel: The DBA, The Legacy Sysadmin, The Startup Founder, The Goat Farmer's Counsel.
Brian Krebs published a piece on April 30th about Huge Networks, a Brazilian DDoS protection firm whose own infrastructure was used to launch a sustained botnet campaign against small Brazilian ISPs. The botnet was a Mirai variant running on compromised TP-Link routers. The exposed file archive included private SSH keys belonging to the firm's CEO, Erick Nascimento. Nascimento told Krebs the activity was the work of a competitor trying to tarnish his company's image, and that he has strong evidence stored on the blockchain of who did it, but he won't share the evidence because it would lose the surprise factor. The Legacy Sysadmin opens by pointing out this is the fourth time this exact story has surfaced in nine years. The original Mirai authors turned out to co-own a DDoS mitigation firm in 2017. A May 2025 incident traced to the same Brazilian operator hit Krebs's own site with a 6.3 terabit attack. The FBI seized several DDoS-for-hire services run by that operator. And now Huge Networks. Same country in three of the four. Same business model in all four. The Paranoid CISO declines to speculate about attribution and instead works through what the artifacts in the archive actually show. The Startup Founder steel-mans the CEO's defense as far as it can go, and discovers that's not very far. The Goat Farmer's Counsel notices that he doesn't miss this part. Source Article: "Anti-DDoS Firm Heaped Attacks on Brazilian ISPs" by Brian Krebs, KrebsOnSecurity, April 30, 2026. https://krebsonsecurity.com/2026/04/anti-ddos-firm-heaped-attacks-on-brazilian-isps/ The panel: The Legacy Sysadmin, The Startup Founder, The Paranoid CISO, The Goat Farmer's Counsel.
App intelligence firm Appfigures published a report finding that image-model launches drive 6.5 times more app downloads than chatbot upgrades. Google's Nano Banana added 22 million downloads to Gemini in 28 days. ChatGPT's image generator added 12 million. The findings were widely reported as evidence that image AI is the new top of the funnel for consumer AI products. The revenue numbers, in the same report, told a different story. Nano Banana generated $181,000 in consumer spending over the same window. ChatGPT generated $70 million. Meta AI generated effectively nothing. The Startup Founder opens by celebrating the download numbers as proof that distribution is unlocked. The panel works through what that framing leaves out. The Burnt-Out SRE walks through what 22 million installs in a month actually looks like at the infrastructure layer — the inference cost, the queue depth, the pager schedule, the unit economics of serving a free product to a hundred thousand new users a day. The Legacy Sysadmin places the chart in a forty-year arc of identical-shaped growth curves attached to products that turned out to be free promotional services rather than businesses. Source Article "Image AI models now drive app growth, beating chatbot upgrades" By Sarah Perez at TechCrunch, May 4, 2026. https://techcrunch.com/2026/05/04/image-ai-models-now-drive-app-growth-beating-chatbot-upgrades/ Panel The Startup Founder The Burnt-Out SRE The Legacy Sysadmin The Goat Farmer's Counsel.
Microsoft Defender pushed a signature update on April 30th that flagged DigiCert's root certificates as malware. For about ninety minutes, enterprise endpoints across the internet started removing the certificates that validate roughly half of public web traffic. The signature update was, in Microsoft's framing, working as designed. The panel works out what "working as designed" actually means in this context. The Legacy Sysadmin places it in a thirty-year arc of antivirus software taking down the systems it was supposed to protect. The Burnt-Out SRE walks through what the incident looked like at 3 AM from inside a NOC. The Paranoid CISO declines to reach for a nation-state attribution and instead names the actual problem: the threat detection vendor and the certificate authority are both root-trust holders, and there is no audit layer underneath either of them. Source Article "Microsoft Defender wrongly flags DigiCert certs as Trojan:Win32/Cerdigent.A!dha" by Sergiu Gatlan, BleepingComputer, April 30, 2026 https://www.bleepingcomputer.com/news/security/microsoft-defender-wrongly-flags-digicert-certs-as-trojan-win32-cerdigentadha/ Panel The Legacy Sysadmin The Burnt-Out SRE The Paranoid CISO The Goat Farmer's Counsel