The team
AIP is R/GA AI Products, the global group that builds R/GA's own products rather than client campaigns. Its first product, BRDNA Sequencer™, launched September 2026 and teaches coding agents to build on-brand.
BRDNA Sequencer makes a brand machine-readable. It takes the design system, components, patterns, voice and rules a brand keeps in Figma files and PDFs, and encodes them into structured context a coding agent can read and act on. Point an agent at a brand, hand it a creative brief, and it works in Figma and Claude using the right components, without a human checking every output against the style guide.
It is not a chatbot or a template generator, it's a collaboration tool. It's a way to encode a design system and create new designs from it while working alongside the designer or marketer.
The role
You turn design systems into something an AI can build with.
The tool takes Figma designs, translates them into markdown, manipulates them, and translates them back. That round trip is the whole product. Making it faithful in both directions, across any brand, is the hard part.
The job needs two skill sets:
- Design systems. You have built, shipped and maintained one. You know what makes interface components and token architecture valuable, and scalable for product team.
- AI fluency. Prompt engineering is a craft you practice rather than read about. You have a feel for what context a model needs, why it fails, and how to restructure input until output is dependable.
You also own the front end of the customer experience: the onboarding and self-serve flow that gets a new team from "we have a Figma library" to "our agents are building on-brand." You'll build alongside the product team.
Why this seat
We are still building product-market fit and the maturity to get there: onboarding that lands without a services engagement wrapped around it, and encoding craft repeatable enough to scale past bespoke work. You are not inheriting a finished thing to maintain. You are helping decide what it becomes.
What you'll do
Three things, in order of weight.
1. Polish the encoding pipeline
The core of the job, and a round trip in both directions.
Figma → markdown. A brand's design system, meaning components, patterns, tokens, rules and voice, encoded into structured markdown and tokenized into a vector space agents retrieve from.
Markdown → Figma. The same understanding translated back out, so what an agent produces lands as real designs rather than something approximate.
The round trip has to hold up across brands of all kinds, and the structure you choose decides whether a model reaches for the right component or an adjacent one. The same brand and the same brief should not produce meaningfully different output run to run. You will be in the loop constantly: encode, point an agent at it, look hard at what comes back, work out whether the variance came from the source system, the structure, the retrieval or the prompt, then go again. No settled practice exists for any of this yet.
2. Own onboarding and self-serve
You own the experience that carries a new team from "we have a Figma library" to "our agents are building on-brand assets." Audit what exists, work out where it loses people, rebuild it.
This is the difference between a service we deliver and a product that scales, and it is why leadership and research are close partners here. You'll set the self-serve KPIs together and let the numbers settle the arguments.
3. Design and build the product surface
You own the front end of the customer experience, designing it and shipping it. No handoff, no spec thrown between two people. You'll build production alongside full-stack product engineers.
What we're looking for
5+ years in design engineering, product design, or front-end product work.
Non-negotiable
You have built a design system and lived with it.
Owned one, not consumed one. You have made the call on component, watched a token architecture meet a real product team, and still been there two years later when it had to change without breaking everyone downstream.
You have designed a full product.
You have reasoned about what to expose and what to hold back, about extensibility, and about the subsequent teams to adopt it.
You build layered prompt systems that produce production-grade output.
The specific bar: multi-hierarchy prompt inputs and skills that yield high-fidelity, production-level assets. Work that shipped and was held to a real quality standard, not demos. You understand how structure at one level changes behavior three levels down, and you have a method for locating which layer is at fault when output degrades.
You think on a product's timeline, not a project's.
You own what you ship. You argue about long-term product health and against technical debt, you care what the metrics say months and years out, and you have opinions about decisions whose cost doesn't land until next year.
On agency backgrounds:
R/GA is an agency. AIP is a product team that happens to live inside one. Agency work is built to ship and hand off; this work you live with. We are looking for people whose first instinct is to ask what a decision does to the product in a year, a reflex that startup and early-stage product careers build and campaign work generally does not.
Strong signal, not required
- Figma as a platform rather than a canvas: Variables, the Plugin or REST API, Code
Connect, library architecture across multiple consuming teams - Retrieval and embeddings in practice: chunking strategy, RAG, fine tuning, LoRAs, what
makes encoding and decoding work and why - Evals: you measure model output rather than eyeballing it, and you have built the
harness to do it - Self-serve and product-led growth: you have shipped an onboarding flow and watched
activation numbers move because of a change you made - 0→1→10: you have been early somewhere and know what that costs, and how to maintain it once customers are using it.
The stack
- React / Next.js for the product surface
- Supabase for data and backend
- Figma MCP to reach into design systems
- Model-agnostic by design. We do not build against one LLM. Interoperability is a product requirement: BRDNA Sequencer has to land in Figma, Claude, ChatGPT and Frontify alike, so nothing may assume a single provider's behavior. In practice this raises the bar on your prompt work, which has to hold up across models rather than be tuned until it works on one.
Cursor & Claude Code
Your first 90 days
Onboarding is robust and resilient. A team can get themselves running, and the path holds up when they do something we did not anticipate.
Workflows are meaningfully less non-deterministic. Same brand, same brief, dependable output, with variance reduced to where people trust it without checking every result.