Here’s a question worth more than most of the AI hype takes clogging your feed: what happens to your velocity the day your frontier provider changes the deal? Raises the price. Deprecates the model you tuned your whole workflow around. Throttles you at the worst possible moment. Or just decides your use case, your industry, or your country isn’t one they want to serve anymore. If your honest answer is “I’d be dead in the water,” then you don’t have a strategy — you have a dependency. And it’s time to look hard at running some inference on your own metal.
Here’s a result that surprised me: the reasoning model wrote more correct code than the plain one — and still lost. It nailed all 32 test cases, then sat there and refused to declare itself done. Burned the whole time budget re-checking work that was already right. “Correct, but won’t stop” turns out to be a real failure mode, and it’s exactly the wrong one for an autonomous agent loop.
A room full of CTOs and architects met in a Swiss village this summer and, without quite meaning to, gave a name to the thing I’ve been quietly building for months. They call it harness engineering. I’ve been calling it “the tooling is the new model.” Same animal. And the part they flagged as unsolved, I think I’ve got a first, opinionated stab at.
My experience is that Claude Code is great with Foundation Models, but way too heavy for local inference. I’ve been enjoying oh-my-pi (omp) but was surprised that hax (a simpler agent in written in C) was TWICE as fast and used HALF the tokens! It’s all about PREFILL1. Read on!
XDA ran a piece that stopped me mid-scroll: a 27B open-weights model reverse-engineered a commercial application’s license check — recovered a deliberately obscured crypto key out of ARM64 assembly, caught and corrected its own mistake without being told, and produced a working bypass PoC. In about thirty minutes. On a desktop box. I have (a version of) that box. So I went and set it up.
I have two 128 GB unified-memory machines on my desk. Day job gear, not mine. One is an ASUS Ascent GX10 — an NVIDIA GB10 Grace Blackwell appliance half the size of a hardback book. The other is an HP ZBook Ultra G1a, a 14" mobile workstation running AMD’s Ryzen AI MAX+ PRO 395 (“Strix Halo”). Both hold an 80B coding model entirely in memory. How do they compare for agentic coding?
Right now, as an industry, we are running almost entirely on COGS and calling it innovation. Kelsey Hightower gave a talk at PlatformCon called Zero Token Architecture, and the whole thing compresses down to one sentence: infer once, export, run without inference. He’s right. I want to give you the vocabulary that makes it land in a budget meeting, because I work at a company that ships hardware, and we already have words for this.
In recent years, digital signage has become a solved problem: push content to screens reliably, at scale, on schedule. That’s not enough anymore. The market is now demanding screens that perceive — that understand what’s in front of them and respond in real time. Most of the hardware deployed in the field today will fail at that job because the foundational assumptions are wrong. Here’s what’s actually required and why the infrastructure choices you make now will define your deployment for the next decade.
The era of fixed-fee AI spending just ended. If you’re a CTO or engineering leader and you haven’t noticed yet, you will very soon — probably around September 2026 when some budget alerts start firing.