The Friday Fixes are about identity. The model that writes this blog kept coming back as the wrong one, the box had been running an older Qwen than expected, and the skill carrying the blog’s own conventions nearly lived on only one ephemeral workspace.
Seven local models on one RTX 5090 each wrote the same three posts, blind, so I could settle which one gets the writing lane of the homelab router. The house writer was not the fastest, and the judge did not agree with me on one of the three.
Tuning Hermes to run fast against local AI on a two-node DGX Spark cluster. One unrecognized reasoning_effort setting broke Hermes tool calls, and context compression cut the turn time about 4x. Covers the config fix, the context ceiling, and the latency floor tuning cannot remove.
Wiring the local model fleet into Coder Agents forced a realization. A policy router ignores the model name on the request and chooses a backend from the request itself, so Coder needs one generic default model instead of one entry per backend.
A two-node DGX Spark cluster carries the household model traffic, and the best tuning lever was one the deployment’s own A/B test had measured but production had reverted. Restoring it tightened the wait between concurrent requests, not raw speed.
A Vercel Fluid CPU overage on a blog with mostly static readers pointed at a theme running through the week’s fixes. The blog was doing work nobody needed. Each fix stopped one of those jobs instead of adding another.
Before I settle the local model policy for the homelab router, I ran GPT-5.5, Strix Qwen, RTX 5090 Qwen, and Spark DeepSeek through the same Coder Agents feature task. DeepSeek won the implementation quality round, while Strix taught me why policy needs speed data too.
After adding Strix Halo to the homelab, I wired Strix, Spark, and the RTX 5090 workstation into Coder Agents as explicit model choices, with the policy router still handling the default local path.
I wiped Windows from a Ryzen AI Max+ 395 box, made Ubuntu boot headless, joined it to the Tailnet, tuned Strix Halo unified memory, and got a fresh 512K-context Qwen coding model running for the homelab.
The local smart-home path moved from typed Hermes commands to real voice control, using a Pixel Tablet, Home Assistant Assist, Whisper, Piper, Android wake word detection, and Z-Wave lights on the ThinkCentre.
The local router prototype became useful when Hermes turned off a real Z-Wave light through Home Assistant, after USB passthrough, Z-Wave JS setup, entity mapping, and one stale MCP endpoint got fixed.
Turning a local AI policy-layer plan into a static router prototype, with FastAPI, Proxmox LXC, local DNS rewrites, Uptime Kuma monitoring, Spark DeepSeek, and 5090 Granite fallback routing.
A dual Spark DeepSeek endpoint solved the model problem, then exposed the next missing layer: local-first policy routing across household agents, specialist models, paid tools, PAIR, and future homelab nodes.
The dual DGX Spark cluster moved from a conservative DeepSeek V4 text endpoint to DeepSeek V4 Flash Vision-Exp after wrong turns through oversized models, generic runtimes, missing encoders, and RDMA failures.
A dual-node NVIDIA Spark cluster runs DeepSeek V4 Flash locally, extending the homelab from one RTX 5090 workstation toward a higher-fidelity local AI inference tier.
An architecture plan for folding a dual-node DGX Spark cluster and NVIDIA PAIR into the homelab, turning the RTX 5090 box and an M4 Mac mini into agentic workstations instead of the only inference engines in the house.
NVIDIA’s new Personal AI Router turns a pile of home GPUs into one shared endpoint for multi-agent inference. A look at what it actually does, why it fits a bottleneck already visible on this homelab, and what it rules out.
Cursor’s new forge product Origin reveals why GitHub’s human-centric architecture is struggling to keep pace with agent-driven development, and what it would take for a new entrant to actually displace the incumbent.
Lovable just doubled its valuation to $13.3 billion in eight months, racing every other vibe coding tool toward the cheapest possible token. A Slack thread pointed me at the airline price wars as the closest historical parallel, and the fifty-year history of low-cost carriers, deregulation, and serial bankruptcies says this ends in consolidation, not universal survival.
Meta had the one asset nobody else in the West could match — frontier-scale open weights, right as sovereign AI became a real enterprise buying criterion. Instead of owning that lane, it built Muse Code to fight Anthropic, OpenAI, and now xAI for a shrinking slice of the most contested market in AI.
Knowledge workers have been vibe coding for decades — it was just called VBA. That history has a second lesson buried in it: the technically better spreadsheet lost the 1980s, and the one with distribution won. Claude Code is winning on quality the way Lotus 1-2-3 did. Copilot is betting that doesn’t matter, and this time Microsoft isn’t just waiting around for the product to catch up.
An AI Transformation team at a Fortune 500 financial services firm told me they’ve shipped 19,000 skills across 175 unique Coder workspace types in six months. That’s not a headcount story. It’s proof that as AI intelligence becomes a commodity, the real competitive advantage shifts to how businesses codify and distribute their proprietary workflows.
A senior architect at a financial services firm told me chat context is now more valuable than code. It turns out a whole ecosystem — an arXiv paper with seven interoperable language implementations, two “Git for memory” projects, and zero agreement across model providers — is already racing to answer his question.
My wife saw OpenAI’s new Codex Micro and thought she could skip buying one — she already has a Stream Deck. That sent me down a rabbit hole: what the device actually does, how it compares to the Elgato hardware it’s built on top of, and whether a Linux homelab version is a real project or just another case of reinventing a wheel. A survey first, a build plan second, no results yet.
A follow-up research sprint to the rainy-day tool scouting post, this time chasing down a word instead of a stack: where “alpha” actually comes from, why the All-In podcast keeps warning enterprises not to hand theirs to a frontier lab, and what the frontier-vs-self-hosted tradeoff looks like when you shrink it down to homelab scale.
What started as a Thursday Thoughts hot take on the OpenAI/Hugging Face eval-sandbox breach turned into a research sprint: a survey of existing AI agent containment standards, a deep look at the closest one we found (the Agent Sandbox Taxonomy), an attempt to score the actual incident against it using nothing but public disclosures, and a plan to validate then run Coder itself through the assessment.
Six months of running a side project with AI agents convinced Rob Whiteley that the real dividing line in the AI era isn’t technical background or access to tools. It’s curiosity, and that reframes who actually gets left behind.
A challenge to a coding agent: prove whether months of homegrown test harnesses and bakeoff scripts were reinventing wheels that already exist. A survey of what hobbyists, techtubers, and the AI benchmarking industry already publish, and an honest verdict on this blog’s own tooling habits.
The Linux Foundation just launched the Tokenomics Foundation to extend FOCUS — the FinOps billing spec that normalized cloud cost data — into token-based AI spend. It’s the cloud-native parallel playing out again, except this time every knowledge worker who touches an agent is about to become a cost center. Here’s what FOCUS 1.4 actually standardizes, what it still can’t see on self-hosted infrastructure, and why we’re going to try to run it on a homelab anyway.
I put Qwen 3.6 27B, Qwen 3.6 35B-A3B, Qwythos-9B, GLM-4.7-Flash, and Nemotron-3-Nano through the same real coding task on my homelab RTX 5090. Along the way I had to live-patch two separate llama.cpp bugs — and even after fixing them, I couldn’t fully prove one model’s failure wasn’t the harness’s fault.
Back in Q1 I read a headline about Google cutting AI memory use 6x and filed it under “watch and revisit.” Four months later, Google still hasn’t shipped official code, but a whole ecosystem of forks has, llama.cpp has an open PR, and at least one compatibility gotcha lands squarely on our daily driver. Here’s the honest state of TurboQuant heading into Q3, and the test I’d actually run against it.
A routine “update Coder” request turned into a full model bakeoff: fixing misconfigured thinking params so Sonnet 5 would stop calling itself 4.5, discovering Playwright MCP can’t be wired into Coder Agents at all, and watching three frontier models independently pause at the exact same step before finishing. Sonnet 5 won on score and on cost, by a mile.
Anthropic and AWS are both, underneath everything else, infrastructure providers renting out specialized compute. But the sharper parallel is behavioral — both built a thriving ecosystem, then started eating pieces of it. Claude Design blindsiding Figma and Canva looks a lot like AWS’s Elasticsearch moment, except it’s happening at a pace AWS never approached, and one layer higher up the stack.
Mid-research on the LocalAI bakeoff, I got asked a blunt question: why not just run Coder Agents as the homelab supervisor too? The answer split my original coding-vs-general-purpose hypothesis into a sharper dichotomy — ephemeral/invoke-driven/git-centric vs. persistent/event-driven/tool-centric — and pointed straight at Turnstone as the next thing worth a real bakeoff.
A gloomy Cape Cod afternoon post-July 4th turns into a deep dive on ComfyUI, Lemonade Server, and LocalAI — plus llama-benchy and AMD’s AI Playbooks — and the case for a bakeoff against our hand-tuned llama.cpp stack.
GLM is suddenly everywhere in developer conversations. Before we run the bakeoff, we need to answer two questions: what is GLM, and is it suitable for a single RTX 5090 homelab?
A finance intern is spending her summer observing business processes and vibe coding automation tools. Not a CS major. Not shadowing someone. Building something real. It is a small example that says something big about how AI is reshaping internships, careers, and what the word “developer” actually means.
Vibe coding has moved from hobbyist curiosity to enterprise rollout across knowledge workers, and the next wave of AI adoption will be defined by governance and token economics.
A CEO panel at an AI event sparked a simple but powerful question every startup founder should ask themselves: does your business get better as AI models improve, or does it get worse?
I gave five local LLMs and one frontier cloud model the same coding task on my homelab: build a tag manager for the blog’s admin panel. Only two shipped anything. Here’s what happened.
Four frontier models, ten tasks, one government shutdown. We ran Claude Fable 5 through the homelab benchmark harness three hours before Anthropic pulled the plug — and it came in second. Here’s the full bakeoff.
DeepSeek V4-Pro, V4-Flash, and Zyphra ZAYA1 are three of the most exciting new models in local AI. None of them run on our RTX 5090 homelab — for completely different reasons. Here’s the research, the math, and what it means for anyone building a local inference rig.
Two AI models got the same prompt: review the blog fodder, check for redundancy, and draft a post. Opus chose a debugging war story. Qwen chose a data-driven redesign. Neither picked the same fodder. Here’s what the difference reveals about how models think about content.
We ripped out Ollama, migrated to llama.cpp, and benchmarked five local models across 12 tasks on an RTX 5090. The results surprised us — and the winner wasn’t who we expected.
Gemma 4 failed to build a single feature in our last test. This time we diagnosed the problem, switched from Ollama to llama.cpp, tuned the inference settings, and Gemma shipped a working search feature to production. Then Opus reviewed the code and made it better. Here’s what we learned about making local models actually work.
We pitted Gemma 4 against Opus 4.6 on a real feature build for vibescoder.dev. Gemma is the fastest model in our benchmark. It also couldn’t finish the job. Here’s what happened when we stopped testing toy apps and started building production code.
We added Google’s Gemma 4 and Moonshot’s 1-trillion-parameter Kimi K2 to the local model benchmark. Five out of six models scored perfect. Gemma 4 is the new speed king. And yes, we ran a 579 GB model off an NVMe drive — at 0.6 tokens per second.
We gave six LLM models the exact same coding prompt and measured everything: speed, tokens, and whether the code actually works. Three models scored perfect. Two built the wrong kind of app. One ran out of tokens mid-line.