Blog
Accuracy is the wrong number to pick a model on
Two models that score the same on a benchmark can cost ten times apart per correct answer. Here is the number a finance team should look at instead.
Rayyan DarugarWhy we started Dynamo, and why we started with finance
Every company we talked to had built an agent that worked in the demo. None of them had one in production.
Why most enterprise AI stalls before production
The proof of concept is the easy part. The last two months, where security, permissions, and the systems of record get involved, are where projects go quiet.
From a shared inbox to a matching agent in five working days
What an AP team built on Dynamo in its first week, which steps it kept for a person, and what the first month of cases showed.
An agent should never be able to do more than the person who ran it
Service accounts are why security teams say no. Here is the permission model that gets them to yes.
Which models write a usable campaign brief? We tested nine
We gave nine models the same forty briefs and had the marketing team grade them blind. The ranking was not the leaderboard's.
The CRM hygiene agent: what it changes, and what it leaves alone
It fixes stages, owners, and close dates overnight, and asks the rep before touching anything with a number on it.
The five questions a controller asks before an agent touches the ledger
None of them are about the model. All of them are about the record.
Bring us a process your team still does by hand.
We’ll give you automated workflow agents.