Graaft, the AI studio I co-run, sells AI people, so its site has to make one feel human in copy nobody would mistake for generated. Mine to design, build and write.
Colour and type
We sell AI people, so the site selling them could not read like an AI wrote it
AI marketing has a house style now: fluent, agreeable, empty. A buyer decides in about a second whether our work reads like that, so the site cannot use a word of it.
Which makes the site the product rather than an advert for it. Beneath it sits one template, where workers, people and projects all answer an identical short set of questions from one source. That is why thirteen pages share a voice.
The awkward fact goes in plain sight too: two humans behind sixteen AI specialists. Publishing it was my call. Hide the small thing and nobody believes the big one.
I did not design a system and then fill it
No library and no specification up front. I worked page by page, and promoted whatever the design asked for twice. Fifty components came of it, every one made for this site.
Discards were broken up for parts, so a failure usually lived on inside whatever replaced it. Systems planned in advance tend to sit half idle. This one only grew when the work insisted.
Thirteen pages and seventy-six entries
One content model, fifty components and 1,216 changes, by one person. Live, and the first thing a prospective client sees.
The voice holds at that scale and no page coins its own terms, which is what separates design from assembly. A new entry is a form to fill in.
Judge it on restraint, not volume: nothing on it promises more than it can deliver. Reach is unmeasured, which is exactly what the studio's own console was built to settle.