SysAdmin Weekly #24: Doom, Discounted
When the Headline Says Layoff, the Press Release Says Capex
TL;DR
The AI doom narrative has hit saturation. Some of the doom is warranted. A lot of it is marketing.
“AI” is doing PR work for layoffs that were really overhiring corrections. The investor signal is the point.
Episode 046 of the podcast is out: the honest case for Claude Code in SysAdmin workflows, plus three fresh Linux LPE CVEs.
Tool of the Week: Ollama. Run open-weight models locally, no data leaves your network.
A full data breakdown of the doom narrative is coming on SysAdmin Weekly Podcast Episode 047 with Eric Siron in a day or two.
From the Console
The AI doom narrative has hit saturation. You feel it on LinkedIn. You see it in every other headline. Microsoft’s AI chief telling lawyers, accountants, and marketers they have eighteen months. Big tech CEOs floating that AI will replace half of white-collar work by 2030. Investors clapping. The rest of us reading the room.
Some of the doom is real. The transition is going to be painful for real people. Data center water usage today is a problem. Real jobs have been cut and real households are absorbing that. I am not minimizing any of it.
The rest of the doom is marketing dressed up as inevitability. And it is loud enough that even people who already do this work for a living are starting to second-guess themselves. That is the part that bothers me.
So I went looking at the actual data. Layoff numbers. The WEF Future of Jobs Report. Energy and water projections. What the hyperscalers are actually doing about it versus what the headlines say they are doing. There is a full breakdown coming on the podcast with co-host Eric Siron in the next day or two. Until then, the short version: the data does not say what the headlines say it says.
The rest of this issue is about the part of the doom narrative that does not get enough scrutiny in our circles: how it is being used.
And now, back to our regularly scheduled programming.
The Latest on the SysAdmin Weekly Podcast
Episode: 046 - Can Claude Code Help SysAdmins? Scripting, Log Analysis, and the CLAUDE.md Workflow
Topic: Andy makes the honest case for Claude Code in SysAdmin workflows, with the caveats and the parts that still fall short. Plus three recent Linux kernel local-privilege escalation vulnerabilities and what they mean for your patch cycle.
Reasons to tune in:
Most AI demos are built for developers. This is a long-time SysAdmin walking through what actually helps and what does not
The CLAUDE.md workflow is the single biggest thing most folks are missing when they try AI in their day job
Three Linux LPE vulnerabilities (Fragnesia, DirtyFrag, CopyFail) dropped in quick succession. Local-only, but still on the patch list
If you prefer to read about my experiences with Claude Code instead of the podcast episode, I did a write-up over on my blog at andyontech.com
All Links For the Show: https://www.sysadminweekly.com or watch / listen below!
The Take
The piece of the AI layoff story that does not get enough airtime is the part where AI is the excuse and not the reason. Again, not downplaying those jobs that were legit replaced by AI, but the enterprise space seems to love making AI the scapegoat, warranted or not.
The early wave of 2023 tech layoffs had a different reason on the slide deck. IBM and a few others actually said it: we over-hired during COVID by something close to thirty percent, the demand correction hit, and we cut. That is not a comfortable thing for an executive to say in public. It also is not an investor-friendly thing to say at an earnings call. So the language quietly changed.
By 2024 and into 2025, the same kinds of layoffs were getting a different label. “Restructuring around AI.” “Reallocating capital toward AI initiatives.” “Aligning our workforce with the AI-first future.” Conveniently, that wording does two things at once: it makes layoffs feel like strategy instead of correction, and it signals to investors that the company is an AI play, which moves the stock.
We are not just speculating here. Cisco cut roughly 4,000 people on the back of strong earnings. Oracle is reportedly cutting up to 30,000 jobs (about 18% of its headcount) while standing up a massive data center buildout. Meta announced 8,000 more cuts in May while telegraphing $115B to $135B in 2026 capital expenditure, almost all of it AI infrastructure. Read those press releases against each other and the pattern is consistent: trim the headcount line, route the savings into AI capex, take the investor bump for being “AI-first.”
Again, that doesn’t mean AI is NOT displacing some jobs. It is, in specific roles, at specific companies, on a real timeline. It does mean, though, that “we cut X people because of AI” is doing a lot more PR work in 2026 than it is doing operational work. And if you are a SysAdmin, an MSP operator, or an IT manager trying to read the temperature in your own organization, that distinction matters. The internal memo and the external press release are not telling the same story for the same reason.
The piece I keep coming back to is something Eric Siron mentioned in a recent recording of the podcast….. when an executive who has never been an accountant tells accountants they have eighteen months, ask that executive what THIER numbers actually need to look like by then. Eighteen months is often less about when the technology will be ready and more about when the latest AI bet has to start showing returns.
We will get into the data side, the WEF Future of Jobs numbers, the historical tech wave precedents, the water and power question, in detail when Episode 047 drops with Eric Siron. For now: if your CIO walks in citing the eighteen-month deadline, ask them to source it. Real conversations get easier from there.
Community Signal
High-signal community work worth your attention this week.
Brandon Lee (VMware vExpert) - “I Built a Local AI Coding Agent Home Lab Setup With OpenCode and Ollama” - The practical antidote to the doom narrative… practical use of AI systems! In this post Brandon Lee walks through standing up OpenCode against a local Ollama instance, with the model running on his own hardware and no data leaving the lab. The part worth the read is the honest scoping: where the local setup actually competes with the cloud offerings, where it does not yet, and what hardware you need to make it usable. Practitioner work, no vendor influence.
Tool of the Week
Ollama - Run open-weight LLMs locally. Single binary, simple model pull syntax, OpenAI-compatible API surface for any tooling that already speaks that protocol.
The pitch in 2026: if you are tired of either feeding sensitive data to a hosted model or being told you are not allowed to use AI at all, Ollama is the third option. Pull a model (Llama, Qwen, Mistral, DeepSeek, Codestral, others), point your client at the local endpoint, and run inference on your own hardware. Pairs cleanly with OpenCode, Continue, and any IDE plugin or CLI tool that supports a custom OpenAI-compatible base URL.
To be straight: the local model quality at consumer hardware tiers is not the frontier-model quality you get from Anthropic or OpenAI. For privacy-sensitive workflows, internal-only use cases, log triage, and scripting against data you cannot ship outside the network, the gap closes considerably. For greenfield engineering work, you will still reach for the hosted option more often than not based on my experience. Either way, nice to have the options.
Quick Win of the Week
Pick one “AI is replacing X” headline you have seen this month. Trace it back to the primary source. Is it a CEO quote, a press release, a paper, or a survey? Whose survey? How were the numbers gathered? You will be surprised how often the headline number does not survive the trip back to where it came from. Sourcing your own claims is one of the most useful skills a SysAdmin can keep sharp in 2026.
Fun Retro SysAdmin Fact
The first “AI is coming for your job” panic in IT was not in 2023. It was in the late 1980s, during the expert systems boom, when symbolic-reasoning platforms running on Lisp machines were supposed to replace whole categories of decision-making work. By the early 1990s the entire commercial expert systems market had collapsed into what historians of computing now call the second AI Winter. The Lisp machines went to museums, the panic dissipated, and the next wave of IT growth (the consumer internet) created more jobs than the AI Winter ever displaced.
From My Other Corner of the Internet
AndyOnTech
Can Claude Code Help SysAdmins? Real Use Cases for IT Pros - The companion write-up to Episode 046. Covers the CLAUDE.md workflow, scope-based settings, and where to stay skeptical with sensitive data and production-adjacent prompts. If the episode landed, the post is the version you can hand to a teammate who prefers reading to listening.
Thing I’m Re-Thinking
The assumption: When a major tech CEO makes a specific timeline claim about AI displacement, I should weigh it as informed analysis from someone close to the technology.
Why it no longer holds up: After a few iterations, it REALLY seems like a lot of those claims are not predictions, they are deadlines. Eighteen months is a number that lines up with when AI infrastructure spend has to start producing returns. The person saying it has a financial reason for that number, and that reason is not “this is the date the technology will be ready.”
What I am testing instead: Treating those statements as investor communications, not as forecasts. If the same claim comes from someone who has actually done the job being replaced, who is working at the keyboard with the tools, and who has no AI-adjacent stock price to defend, that is the version I weigh.
Until Next Week
If the doom is loud enough that you are starting to repeat it without checking, that is the moment to go check.
Stay Frosty,
Andy
SysAdmin Weekly