Yeita · Ops & AI

SKU PROJ-007 · Release 2025 to today

Yeita
Ops & AI

AI Build AI circle

Yeita treated as a client, on the operations side. First, upskill the team on AI, then put products in its hands: an art direction assistant and an online design system, with a tool that made marketing self-sufficient.

~15 people coached one-to-one, in workshops and through a Notion base
2 products made available to the team: the Brand Assistant and the Yeitibook
100+ conversations on Prompt Docteur, the most used assistant

Spec sheet

Role
Launched and runs the AI circle. Designed the internal products with the team (the design system with Eva and Romain) and built them through vibe coding.
Goal
Make AI fix real pain points at Yeita first: a team that grows its skills, and tools that make it more self-sufficient.
Duration
Since 2025, ongoing.
Team
Around fifteen people coached, volunteer champions, Eva and Romain on the design system.
Tools
Claude, Claude Code, Claude Cowork, Notion, Figma, GitHub, Lovable, Make, Figma Make. Tried, then dropped: Dust, Relume.
Tracked indicators
Real usage of the assistants and tools, licences saved, and eventually the share of the team using AI every week.
Industry
Consulting: Yeita, product & design studio, internal.

The problem

AI usually arrives through the tool: you buy a licence, then look for what it's for. The result: lots of tests, little usage, and a team split between those who experiment alone and those who don't dare to touch it. On the design side, the same bottleneck: the brand lived in Figma, and every visual or art direction question went back through the designers.

Upskilling the team

01 Start from the pain point, not the tool

Four rules, written from day one. We start from a use case and a user, never from a tool. We stay light: no-code and integrations, no project that needs a dev team. No project without an impact indicator set upfront. And AI suggests, humans decide. Four rituals keep the circle alive: 1:1 coaching on a real business problem, live build sessions, a weekly watch with one thing to test, and a squad with two hours to fix one Yeita pain point.

The AI circle's whiteboard: the AI stack and the battle plan

02 Centralise instead of piling up

One tool per need, none of them talking to each other: that was the debt we were building. Everything the circle learns now lives in a single Notion base, "IA qu'à apprendre": fundamentals, tutorials on real cases, workshops. Claude plugs into it: a person or an agent finds the same answer in the same place.

Yeita's internal AI assistants: Yeit'approche, Vibe Coding, Prompt Doctor

03 Build, then decide

We launched seven internal assistants. Prompt Docteur passed 100 conversations, Vibe Coding 30. Most of them evolved or merged, others were killed. Dust too: no internal licence, no usage, we cut it. What holds is an agent harness that starts from a UX repository (personas, insights, journeys) and proposes mockups in Figma. It was used in a client demo and runs in our internal demo tool. Not yet in production on a project: that's the next step.

The knowledge base of Yeita's AI circle

Putting products in the team's hands

04 The Brand Assistant: art direction open to everyone

Design questions come in faster than designers can answer them. The assistant answers, audits, creates and edits screens from a base of about 200 brand rules, without ever inventing, and every output is traced for a designer's review. It is built white label: switching brands means switching rule bases.

See the details →

Brand Assistant · White label

05 The Yeitibook and Y.Cover: the design system as self-service

The design system taken out of Figma and published as code: 40 documentation pages, 1,513 icons, eight rules for newcomers. With Y.Cover, marketing produces its own visuals (podcast, LinkedIn lives), on brand, and some Figma licences could be cancelled.

Open the Yeitibook ↗ See the details →

Yeitibook · Yeita design system

The outcome

Around fifteen people coached, seven assistants tested and a stack narrowed down to what's used. On the product side, marketing produces its visuals on its own with Y.Cover, and some Figma licences could be cancelled; the Brand Assistant is demo-ready, white label. Not measured yet: the share of the team using AI every week. What didn't work: Relume led nowhere, Dust found no usage, some workshops drew few people.

Other drops

SKU PROJ-001 Product, data & AI · Citeo → SKU PROJ-002 Espace Territoires · Citeo → SKU PROJ-003 Clearing · Gireve →

Want to talk about it?