When I first heard the term “context engineering,” I thought it was BS.
I’ve changed my mind.
Since September 2025, I’m spending 95% of my time at Animalz on AI systems building and vibe coding. What I’ve come to realize is that context matters. Not how you phrase your prompt, but the information you provide the AI with and when (in your process).
Do this well, and it can feel like AI pulls a rabbit out of a hat. Case in point: the article I would eventually write about context engineering.
Back in November, I jotted down “write article on context engineering” as a task with a few bullet points. Three months later, a background AI agent picked it up, went off on its own — pulling from my Notion workspaces, reading highlights, and existing article portfolio — and came back with a draft that became ~65% of the final article. (See my LinkedIn post about that.)
The article about context engineering was itself a product of context engineering.
Check it out on the Animalz blog:
Your Prompts Are Fine: Context Engineering Is Your Next AI Problem
The piece explores the topic in more depth, including the skills you need for it, the seven elements of great context, and what makes it hard to pull off.
Reads & Listens
Here’s what stood out in my reading and listening this week.
Claude Opus 4.6 Now Has a 1M Context Window
Anthropic made 1M context the default in Claude Code and available via the API. That’s a 5x jump from the previous 200K limit.
In theory, more context is better, but there’s a catch called context rot: even though a model can handle a million tokens, the quality of its responses degrades as the context grows. The model loses focus, misses details, and gets confused.
These side-effects still happen with Opus 4.6. But the drop-off is much gentler than previous and competing models: meaningful degradation doesn’t really kick in until past 500K tokens. So you can go double as long as before without a quality hit.
Europe Wants Its Own AI but Can't Escape the US Stack
This Sacra research piece on the $100B+ sovereign AI boom shows just how screwed Europe is when it comes to tech independence.
Governments want to build their own AI models, but then they realize those models need to run somewhere, and there’s no real European cloud provider. So their “sovereign AI” still ends up running on AWS, Azure, or Google Cloud and NVIDIA GPUs.
What’s funny is that in the same week I read this, I listened to the Dwarkesh podcast with Elon Musk, where he makes a similar point about the US itself: “Nvidia’s output is FTPing files to Taiwan. It’s digital... a bitstream going to Taiwan.” Apple sends design files to China.
Even the companies you’d think of as self-sufficient hardware giants have the same kind of dependency one layer down, sending digital design files to the real manufacturers in Asia (TSMC and Foxconn).
Why the Helsinki Bus Station Theory Tells You to Keep Going
The Helsinki Bus Station Theory, from photographer Arno Minkkinen,1 goes like this: all buses leaving Helsinki share the same route for the first few stops. Each stop is a year of creative work.
After three stops, you show your work to someone and they say: “This looks like what [established person] already did.” The temptation is to hop off, grab a cab back to the station, and pick a different bus with a different creative direction, in search of originality.
Many people do this over and over.
If you stay on the bus, the routes eventually diverge. Your work becomes distinctly yours. But every time you switch buses, you reset to zero.
“The Helsinki theory suggests that if you pursue originality too vigorously, you’ll never reach it. Sometimes it takes more guts to keep trudging down a pre-trodden path, to the originality beyond. ‘Stay on the fucking bus.’”
This month is my five-year anniversary at Animalz, which is why the theory resonated.
Five years seems like an eternity, and it made me wonder if I’ve been going in circles around the same bus stops, or whether the route is starting to diverge.
Working at the intersection of AI systems building, content crafting, and productivity/creativity feels like it’s heading somewhere uniquely mine. And I’ve never enjoyed my role at the agency as much as now, so maybe that’s what divergence feels like?
Thanks for reading!
Tim
Someone sent me this a long time ago; if it was you, thanks! I finally got around to reading it.