Category
Questions we are still working on, and what the evidence says so far.
6 articles
Research
Before you can say anything useful about what agents do to work, you need a map of what agents are actually connected to. We built one, and the shape of it was not what we expected.
Yaju Team · 16 February 2026
Future of work
We looked at what happened in the first ninety days after teams switched on their first automations. The pattern was consistent enough to be worth writing down, and unflattering enough that most vendors would not.
Yaju Team · 12 January 2026
Models
The reflex when an agent underperforms is to reach for more capability. Sometimes that is right. More often the failure is somewhere the model cannot reach, and upgrading simply makes the same mistake more expensively.
Yaju Team · 2 April 2026
Models
We built each version of Oran against an assumption about what teams would do with it. Some of those assumptions held. Three did not, and the gaps between what we expected and what happened are the more useful thing to write down.
Yaju Team · 5 September 2026
Future of work
Almost every claim about agents replacing jobs rests on estimates of what a task is, produced before anyone had watched an agent attempt one. The estimates are not dishonest. They are just measuring something other than what is happening.
Yaju Team · 21 April 2026
Research
Retrieval quality is usually treated as a ranking problem. A large share of the failures we investigate turn out to be absence problems: the answer was never written down, or was written down somewhere the index never reached.
Yaju Team · 24 June 2026
