"AI-first" has the problem most marketing labels eventually get: everybody is. Every consultancy from Big Four to a freelancer with a ChatGPT subscription claims it. So this note is what the label actually means inside this studio: the specific things that change vs. a 2020 studio operating model.
1. The intake doesn't bounce to a human
The form on the homepage doesn't email the studio. It posts to an endpoint that classifies the brief (paid / earned / wrong-fit / spam), writes a structured log, then sends a summary digest. The studio reads the digest, not raw mail. The classifier costs about a hundredth of a cent per intake; it replaces what would otherwise be a real-time triage tax.
Every handoff we remove is time back. If nobody has automated the sorting, somebody is doing it by hand every morning, and that somebody is usually the owner.
2. The site is legible to language models
Schema.org markup on every page. llms.txt at the root. Citation-shaped content: the most quotable sentence is paragraph one, not paragraph five. Inbound through ChatGPT, Claude, and Perplexity is treated as a real channel, not an afterthought.
This isn't future-tense. Search referrals to small business sites started dropping in 2024 and they haven't recovered. The bypass is in. The question isn't whether to be ready. It's whether you're ready already.
3. Closed loops, not open ones
Every important process at Xala is a loop, not a fire-and-forget. Day 0 / 30 / 60 / 90 AI-citation snapshots. Weekly digest of new intake. Project-status pings. Each loop captures what happened and either acts or surfaces. No item sits in someone's head waiting for them to remember.
4. Case studies are spec-driven, not hand-written
Every case study on this site started as a spec (a one-page CASE_STUDY_SPEC.md describing what's required) and a generation pass against each project's existing notes. Then a hand-edit pass for tone. Hand-writing six case studies from scratch is two weeks of work; spec → generate → edit is two afternoons.
5. We don't fake it
What's not AI-first here: the design judgment. The case studies are spec-generated but tone-edited line by line. The intake classifier helps triage but every brief eventually goes through us. The citation tools are real working code, not "powered by AI" badges on stock services.
That's the whole shape. AI-first is operational, not decorative. Most of the work is figuring out which decisions to keep human and which to let the machines handle. The studios that do this well will outpace the studios that don't. The difference will be invisible to the buyer, but it will compound for the studio.
Questions are welcome: hola@xala.studio