Category
Work we do in the open, with the community that does it with us.
5 articles
Research
An agent that performs well in English and adequately elsewhere is not a multilingual agent, it is an English agent with a translation layer and an uneven failure rate. The difference matters most in the places it is hardest to see.
Yaju Team · 12 August 2026
Research
A research programme run from one city sees the problems of that city. Most of what we have learned about operating agents in difficult conditions came from members working somewhere with different constraints.
Yaju Team · 4 August 2026
Research
Education is the sector where the gap between what an agent can plausibly do and what it should be allowed to do is widest. That makes it interesting to study and a poor place to move quickly.
Yaju Team · 17 February 2026
Programmes
The Scholars Program gets more applications than places, and the ones that stand out are rarely the ones with the longest list of credentials. Here is what we are actually reading for.
Yaju Team · 18 August 2026
Engineering
The hardest part of improving retrieval is not technique, it is having something to measure against. If you cannot yet build an evaluation set from your own documents, building one from a public corpus first is a genuinely useful rehearsal.
Yaju Team · 13 April 2026
