FilmConcept piece
Evergreen Softwash: the launch film
A five-second hero spot cut for a regional exterior-cleaning company. One shot, one idea, and the dirty half of the house still in frame.
· 4 min read
What we have learned, written down
Notes from the work itself. Two running series — loop engineering and graph engineering — plus the occasional piece on what we made and why it was cut that way.
Series 01 · 2 pieces
Loop engineering
The while-loop around the model: context budgets, retry discipline, phase design, and stopping conditions that actually stop. The part that decides whether an agent survives production.
Start here
Loop engineering: the loop is the product
Everyone tunes the prompt. Almost nobody designs the loop around it — which is where agents actually succeed, stall, or quietly burn a thousand dollars overnight.
· 9 min read
Loops that know when to stop
"Keep going until the task is complete" is not a stopping condition. It is a wish. What to write instead, and how to make a loop fail in a way you can act on.
· 7 min read
Series 02 · 2 pieces
Graph engineering
Modelling the relationships in a business as first-class data, and what becomes answerable once you do — including the grounding problems retrieval alone cannot touch.
Start here
Graph engineering: model the edges first
Most business questions that seem hard are relationship questions being asked of a schema that only stores things. Modelling the connections first is what makes them ordinary.
· 8 min read
Traversal beats retrieval when the answer is two hops away
Vector search finds passages that sound like the question. Plenty of real questions are not answered by any single passage — and no amount of better embeddings fixes that.
· 6 min read
Faster than reading all of it
If one of these describes a problem you have, describe it to us
The quickest route to an answer is to tell us the actual problem. We read every enquiry ourselves, reply within one business day, and the first answer costs nothing.