The studio keeps a small set of working notes — positions we have argued for in commissions, ideas we wanted to test in writing, and the occasional polite quarrel with the prevailing temperature of the trade. The collection is short on purpose. We write when the writing matters; we publish when the writing is honest.
Studios live or die by their positions. A practice that cannot tell you, in three sentences, what it believes about its craft will, in time, lose the work that requires conviction and keep only the work that requires capacity. We are not a capacity practice. The notes below are the three positions we needed to hold publicly in order to do the work we have done in 2024 and 2025; the next three, when they arrive, will appear here too.
Every note has been argued internally for at least a month, drafted on paper, debated against a real commission, and edited down until it survives a polite hostile read. We send them to peers in advance of publication. We make the corrections they earn. We are still wrong about some of it.
Why we train small models for one studio at a time rather than sell prompt scaffolding for the general one. The economic question, the craft question, and the awkward truth about why a per-studio model usually wins on the only metric that matters: whether the senior designer signs the work.
An argument for designing the constraint set before the training data. Why a small, hand-drafted constraint document outperforms a large, automatically-curated corpus when the goal is voice rather than coverage, and how we have learned to write constraints that survive the architects who will eventually read them.
The hardest commission we took on in our first year: a type foundry asked us to build a model that knew their house grammar but did not produce work that could be diffed against their existing library. The note describes what we learned about the difference between style and tells, and why most identity-AI work fails at exactly the boundary we were asked to defend.
A note begins as a single sentence on the wall of the drafting room. The sentence has to be either a position the studio holds or a question the studio has been asked too many times to keep answering by hand. The wall is a real wall; we have a photograph of it from the first month of the studio, and the sentence "we don't sell prompt libraries" is in the same handwriting as the one above it, "we don't take work we wouldn't sign".
The sentence sits there until someone is willing to draft against it. Drafting against a sentence means writing the argument the sentence implies and then attempting to break the argument. We have an unwritten rule that a note must survive a serious attempt to disprove it before it is allowed to leave the drafting room; in practice this means each note is killed at least once and rebuilt before it is published.
We do not publish lukewarm notes. A note that survived the drafting room but is no longer believed by the studio is moved to the archive with the date it was withdrawn and the reason. We have withdrawn two notes; both withdrawals are public.
The notes are the studio's public position; the commissions are where the position is tested. If one of the three pieces above describes a problem your studio is currently facing, the most useful next step is usually not a meeting but a paragraph — a single paragraph describing the situation you are in and the position you are weighing. We read those carefully. Replies arrive within a working week; they are written, not templated, and they are honest about whether we are the right hand for the work.
Open the brief sheet for the prompts we tend to write back to, or send the paragraph directly via the contact page. We will write back even when we decline.