Blog
Field notes from the Scholarus team
Essays on craft, product decisions, and what we're learning from the community.
The quiet skill that separates great AI builders
It isn't a deeper model. It isn't a better tool. It's the willingness to read the trace before writing the fix.
A quiet manifesto for Scholarus AI
Why we think a community of serious builders matters more than another pile of tutorials.
Stop benchmarking. Start shipping.
Benchmarks feel like progress. Often they're a way of delaying the harder question: does your thing actually work?
What 'agentic' actually means to us
The word has been hollowed out by marketing. We use it for exactly one thing, and that clarity helps the team.
A week with Devin: honest notes
Ran Devin against a real backlog for a week. Some of it felt like the future. Most of it felt like an intern who doesn't know when to stop.
Tool use is underrated, and everyone's overrating it
Tool use gives you the most leverage of any feature in the LLM toolbox, if you limit it. Let a model call anything and you've built a slot machine.
Interface matters more than the model now
Two teams can use the same model and one will have a product people love. The difference is almost always the interface.
Reading a paper that matters, in 20 minutes
You will not read every AI paper. You can still get 80% of the insight from the ones that matter. Here's the cheap version.
The first 30 days with a new model
A protocol for evaluating a new model against your actual workload, not the press release.
What year-end looks like inside an AI team
A quiet, honest look at how teams that have been shipping AI for a year are ending 2025. Not many demos. A lot of re-architecture.
The case for small, specific prompts
The best prompt is usually the one you could fit on a sticky note. Everything past that is decorating around the idea instead of stating it.
Why your eval set is probably the problem
If the model keeps getting better but your metric doesn't move, the metric is lying to one of you. Usually it's lying to you.
On showing up for AI
The builders who end up good at this aren't the ones who read the most papers. They're the ones who kept showing up after the novelty wore off.