AI agents
Defining what the agent may do and what it must ask first
Identity, interfaces, motion, 3D, research and emerging technology, brought together as one system. Known, and is my lab art collective. Innovation from a unique perspective. Follow below for my active work and the numbers as I go through design and development sprint cycles.
New tech, new products, what shipped. Weekly. Unsubscribe by replying.
Programs: Claude, Figma, Photoshop, Illustrator, After Effects, Procreate, GitHub, Solana, AutoCAD, Godot, Shopify, Blender, ChatGPT, Mistral, Aleph Alpha, Cohere, Qwen, Kimi K3, DeepSeek, Ethereum, Zcash, Monero.
A browser for the web, Web3 and AI agents with one shared security core. Zero-dependency prototype for Colosseum. Product vision, UI direction and community on a team of three.
Building and researching personal harnesses for AI agents: the rules, tools, memory and review steps an agent works inside, tested on my own projects first.
Writing article 1 of a plain-language series on what a wallet, a signature and a chain actually do, drawn from building RobotFac3.
A block puzzle made of cats I drew, built in Godot with AI help. Design, art and build by me. Signed Android test build.
Brand and design from Known, and; development and security from Veruthia. For home-service contractors.
A custom Shopify theme built around one flagship product, a daily food topper for dogs.
A brand-led 2026 website for a growth and exit strategist.
Android app for tradespeople. A spoken job description becomes a priced draft quote. Designed the flows, the AI chat behaviour, 50 screens and a starter component set. The AI requests a rate when none is on file.
Agency and creator portals for a TikTok Marketing Partner: casting, campaigns, scripts and payments in one system. Led design for both portals, the shared style guide and the v2 login.
UX and AI research for a retail stock-and-pricing platform. Friction workshops and prompt research, then a redesign around three ranked actions with the reason for each.
UX research across three learning platforms for an education publisher, name withheld. Interviews, usability tests, accessibility audits and the scorecard leadership tracked. Opening an eBook went from eight clicks to two.
Page-by-page audit of a 130-year-old manufacturer's lead-generation site. 31 findings, 12 high priority, ranked for direct hand-off to developers. Reusable audit kit built in Figma.
Lead designer at Kenvue: creative direction, e-commerce and the design system, built to ADA and North American brand standards.
Motion, icons and templates for Bath & Body Works email, social and landing pages. Brand guides for Conversant and AdParlor.
Art direction, email and motion for SalonCentric (L'Oréal USA) across email, web, social and decks, tuned with analytics.
Email, landing pages and social for Estée Lauder, Clinique and FoodKick redesigns and seasonal campaigns.
I share what I am learning and building, including the tools I am upgrading each day.
Defining what the agent may do and what it must ask first
One security core for web, Web3 and agents
Shipping a game I designed and drew
Brand systems that move
Themes built around one flagship product
Where the models enter my work, where they stop, and the rules I hold to
Most of what I build now touches money, wallets, agents or code an AI helped write. Each fails in its own way, and the design has to account for that before the first screen is drawn.
A person approves every signature. The app shows what a transaction does in plain words before it is signed, holds no keys and sends nothing on its own. In RobotFac3 the wallet sends; the browser only prepares.
An agent that can pay gets a strict grammar for payment sentences, a leak check and an injection detector. Anything outside the grammar stops and asks a person.
Quote On: with no rate on file, the AI requests one. Rules like that are designed into the flow, written down and tested like any other feature.
AI-written code ships after an adversarial review and regression tests. RobotFac3 v0.2: 35 findings from one review, all fixed, each covered by a test.
Hand-written Solana layer, standard-library relay, no npm, no CDN, no web fonts. The relay allow-lists files and RPC methods, refuses wrong Host headers and cross-origin posts, and never reuses a connection.
Every client build starts the same way: MFA and domain lock, SPF, DKIM and DMARC, per-client isolation, backups and a written incident plan. The client owns the domain, the data and the code.
Card and identity data go straight to the processor or the KYC provider and never pass through code we wrote. A certified security engineer is on every build. The proposal names the standards that apply to you, and the build is reviewed against them before launch.
A brand is built to surprise, and what a model has seen it can reproduce. So unreleased work, client files and data never go into one. Ideation draws on every model worth using, on prompts that name nothing, because each thinks differently and that range is part of the craft. Client material reaches only models designed for data sovereignty. Where each model sits.
I am Josie Rigali, the brand incubator and systems design engineer behind Known, and. I design brands that behave like products: identity, interfaces, motion, 3D, research and emerging technology as one system.
Before Known, and I spent 10+ years designing inside beauty, retail, education and AI companies, from email and motion to UX research and design systems. Off the clock I build games and hackathon projects.
Tap-to-open NFC business cards first. The long game is a walk-in brand booth. Both are on the bench for 2026–27.
Stay tuned. Make it known.