About — The Frontier Lab

Most AI writing tracks what happens at the frontier — the labs, the funding rounds, the benchmark wars. This is about something else: what actually runs on hardware you control, in 2026, when you stop reading about AI and start operating it.

I'm Erik, an engineer from Germany, and The Frontier Lab is my notebook on running frontier AI in practice. Most of what gets covered in AI media focuses on the model labs at the top. That matters, but it's not where most of us work. Most of us are downstream: figuring out which open-weight model to run for which task, how to set up hardware that doesn't melt, when local inference earns its keep versus when to just call the API, and how to wire it all together into something that actually works.

That's what this newsletter is about.

What you'll find here

Each week I publish a piece on something I've actually tested: model benchmarks run on real consumer hardware, working configurations including the frustrating edge cases, honest assessments of open-weight models against specific tasks, hybrid local-plus-cloud setups that make practical sense, multi-agent architectures that hold up under load, undocumented kernel and BIOS settings that affect inference performance, and the trade-offs that only become visible when you're paying the electricity bill yourself.

I write from my own lab — running a mix of recent open-weight models across consumer-grade hardware — but the goal isn't to document my setup. The goal is to give you the kind of information I wish existed when I was figuring this out: practical, tested, and free of marketing varnish.

Who this is for

Engineers, AI enthusiasts, and technically curious people. That covers a wide range: solo developers running their first 7B model on a laptop, ML engineers operating multi-node inference clusters, hobbyists who bought a used workstation GPU and don't know quite what to do with it yet. Some do this for work, some because they genuinely cannot stop tinkering. The common thread is that they want to understand the engineering reality — not the marketing narrative.

What this isn't

This isn't an AI news roundup — good ones already exist and I'm not trying to compete with them. It's not a tutorial site for absolute beginners, though an occasional "where do I start" piece will appear when it genuinely fits. It's not written for executives shopping for tools. And it's not a forum for AGI debates or existential risk discussions. Just engineering: what runs, how, and what it costs.

How it works

The newsletter publishes weekly, every Thursday. Most content is free. When a piece requires deeper testing, exclusive data, or extended analysis, it may sit behind a paid tier — but the core newsletter stays open.

Subscribe to get every Thursday's piece in your inbox. If something here is useful, forward it to one person who'd also find it useful. That's how this grows.

Questions, corrections, suggestions: erik@thefrontierlab.ai