Sam Sabey

Airlines price for fuel. Trucking passes it through. SaaS products ran 80 to 90% gross margins because the software was cheap to operate. What happens when inference eats another 40%? otagelabs.com/blog/the-shake…

otagelabs.com/blog/finding-t… 2/2

Seeing the label on a 10mm socket is not the same as knowing when to use it. MCP gave AI agents a garage full of labelled drawers. The knack of knowing which drawer to open first? That comes from somewhere else entirely. 1/2

otagelabs.com/blog/beware-th… 2/2

Knowing where the dragons are was always bundled with the person doing the work. Two production databases got wiped in a fortnight because nobody in the room carried the map. 1/2

Some of my sharpest technical insights happen on singletrack with my heart rate at 125. Not at a desk staring at a screen. The AI bottleneck is biochemistry. otagelabs.com/blog/hacking-t…

otagelabs.com/blog/the-elect… 2/2

Factories took a hundred years to reorganise around electricity. AI is compressing the same transformation into months. Most businesses are still bolting a bigger motor onto the old lineshaft. 1/2

otagelabs.com/blog/the-ai-al… 2/2

Opus 4.7 ignored every branch constraint I wrote. The fix for the first alignment problem in production wasn't interpretability. It was typing /model claude-opus-4-6 into a terminal. 1/2

otagelabs.com/blog/you-are-m… 2/2

I built agentic workflows on Claude's connectors. They broke. The only fix: disconnect and reconnect, like rebooting a Windows machine. This morning the management plane was down too. My agents, my reputation, someone else's outage. 1/2

4/ I keep thinking about how much of that was already in the data. Full breakdown: otagelabs.com/blog/an-advent…

3/ A participant asked to see a different time window. The analysis ran live, carrying forward every decision from earlier. It surfaced something nobody had seen. The founder looked at one number and said he'd never been able to put a figure on it before.

1/ Most data workshops are fishing trips. You talk about what data you might have, what questions you might ask. I walked into that boardroom with five years of production data already prepared. Different category.

otagelabs.com/blog/an-advent… 2/2

Five years of production data in backend systems. Nobody had ever asked what it meant beyond the operational view. One day surfaced 91 missing devices, an invisible market position, and a 99% improvement the founder hadn't computed. 1/2

The M1 Monash Freeway has been under construction for thirty years. Every new lane made congestion worse. Software is about to do the same thing. otagelabs.com/blog/the-freew…

4/ Four months from shipping faster every week to burning tokens on workarounds. Terminal velocity; the extra effort goes into drag. otagelabs.com/blog/are-we-ap…

3/ The inference ceiling is physical. Chips, RAM, cooling, network infrastructure, people to build and operate data centres. Every input constrained. Investment only works if revenue follows.

1/ Every week there's a new tool. A new model. A new benchmark. The narrative is the same: we're early, build faster. Four months ago that felt true.