# Rogue Agents > Boutique AI production partner for ambitious agencies and brands. Rogue Agents is a three-person senior team that builds AI-powered campaign experiences, interactive activations, and branded AI systems. We work as a production partner — integrated into agency and brand workflows, not alongside them. We are not a consulting firm, a dev shop, or a staffing solution. We build things that ship. ## Who We Are **Sarah Collins — Client Strategy** Twenty years of agency and CMO experience. VP at Hanson Dodge for six years. Understands what audiences want and builds the campaigns to reach them. **Pete Kaminski — Engineering** Has been building software since 1981. Designed one of the first graphical interfaces for the internet. Co-founded multiple successful technology companies. **Nate St. Pierre — Creative Technology** Product builder and storyteller. Built AI screenwriting tools acquired by a Hollywood entertainment production company in 2025. Twenty years across product, brand, and creative technology. Credentials across the team: VP-level leadership at creative firms and tech startups, Fortune 500 campaign production, AI platform architecture, Silicon Valley exits, Hollywood AI acquisition. ## What We Build - **AI-powered microsites** — interactive branded web experiences where AI is the engine, not a feature - **Interactive campaign activations** — experiential systems built for participation, shareability, and campaign amplification - **Branded AI experiences** — AI integrated into storytelling, campaign content, and audience participation in ways that feel native to the brand ## How We Work We keep our engagement roster small on purpose. Every project gets the same three senior operators. The person you brief is the person who builds. We think in briefs, not tickets — we understand how campaigns are built, budgeted, and sold. --- ## Selected Work ### Fashion Insights — Competitive Intelligence System URL: /case-studies/fashion-insights/index.html Category: Market Intelligence / Fashion & Apparel / 2025 A fully spec'd competitive intelligence system for athletic and lifestyle apparel. One product photo in, structured market data out — designed to land inside the PLM (product lifecycle management) workflow rather than next to it as a separate tool. **The problem:** Design and merchandising teams spend weeks assembling competitive sets by hand — combing competitor sites, screen-scraping product pages, eyeballing fabric blends, tracking colorways across price tiers. By the time the deck lands, the line has already moved. **What we built:** A competitive intelligence API (Garment Insights) that takes a single garment image and returns structured market data. Key capabilities: - Visual product matching across 7 major athletic and lifestyle catalogs (Lululemon, Ten Thousand, Athleta, Fabletics, Gymshark, TLF, and mass-market) - Multi-piece detection — tops and bottoms analyzed independently from a single image - Pricing benchmarks — current price, promotional status, price tier across the competitive set - Fabric composition and construction details lifted from live product pages - Color variant analysis — same-color matches first, alternates surfaced to reveal trend gaps - Token-based authentication for direct PLM system integration **The principle:** Three constraints shaped the build. One input (a product photo is the only thing a designer should have to provide). Real catalog data (from live competitor catalogs with style names, fabric composition, current pricing). PLM-native (output lands as structured data inside the existing PLM workflow — not a PDF or separate dashboard). **Five extension domains spec'd:** Customer Intelligence, Market Intelligence, Product Intelligence, Operational Intelligence, Workflow Enhancements — 30+ modules across all five. **Speed:** Traditional competitive set assembly takes 1–3 weeks per category and is stale on delivery. The Garment Insights prototype was hand-built in under a week. Full V1 specification and five-domain extension roadmap ready inside three weeks. **Outcome:** Rogue Agents spec work. Production-grade thinking applied to a brief in a category we'd never worked in before — from standing start to fully documented, deployable system design in under three weeks. --- ### YouBots — B2B AI Platform URL: /case-studies/youbots/index.html Category: Contextual AI / B2B Platform / 2024–2025 A production-ready B2B AI platform centered around the Organizational Profile system — a persistent contextual intelligence layer shared across an ecosystem of specialized operational bots. Built to solve the problem of generic AI that knew nothing about the organizations using it. **The problem:** In 2024, AI tools produced generic outputs because they had no persistent understanding of the organizations using them. Every session started from zero. Companies wanted AI that understood their brand voice, competitive positioning, product knowledge, workflows, and institutional context. Custom AI builds were out of reach. ChatGPT and Copilot weren't designed for this. There was no middle option. **The architecture — Organizational Profile (OP):** Before any bot produced a single word, it worked from a structured, hierarchical context system built around the company. Before "context engineering" became standard terminology, YouBots was permanently reserving 20+ pages of organizational intelligence inside every interaction. This isn't prompt engineering hacked together by individual users — it's institutional memory, standardized and deployed across an entire team. When one person adds context to the OP, everyone benefits immediately. **What we built:** - **Retail Platform** — Core SaaS for SMBs. A dozen role-based bots (CMO, content strategist, operations lead, researcher, board advisor, educator, analyst, and more) all grounded in the company's OP from day one. Concurrency-based pricing, not per-seat. - **Agency Suite** — Multi-client management from a single interface. Per-client OP isolation and configuration. Produce on-brand work for each client without cross-contamination. - **White Label** — License the entire platform, rebrand it, deploy it under your own name. - **Vertical Solutions** — Bespoke bot configurations for specific industries (legal, education, nonprofit). **Retrieval system:** Organizations could upload internal materials (SOPs, strategy decks, campaign archives, research). Deep Search activated retrieval against those materials when needed. Most AI systems overused retrieval and sacrificed conversational responsiveness. YouBots treated retrieval as an escalation layer — not the default. **Validation:** A $1B+, 1,000-person company ran a paid three-month pilot for their marketing department. YouBots was ranked in the top tier alongside ChatGPT Teams and Microsoft Copilot for a possible enterprise-wide rollout. **Client testimonials:** - "I let go of my $5,000/month marketing agency and now use YouBots to do more, better, and faster." — CMO, $2MM small business - "The most comprehensive, most helpful, most intuitive, and far and away the most in-depth AI tool that I have experienced." — Executive Director, nonprofit - "It's like having a $10k/month marketing and business development agency working on your behalf." — Founder, PR agency **Outcome:** Profitable from the first month with zero investor money and no debt. All customer growth through word of mouth. By early 2026, persistent context and organizational memory had become table stakes across the industry — the architecture YouBots had been running in production for a year. The platform was sunsetted as the team shifted focus toward Rogue Agents. --- ### Talemaiker — AI Screenwriting Toolsuite URL: /case-studies/talemaiker/index.html Category: Creative AI / Entertainment / 2024 A professional-grade AI toolsuite for screenwriters — built to compress weeks of development work into hours, without once writing a word of their script. Acquired by a Hollywood entertainment production company, January 2025. **The problem:** The WGA strike put AI at the center of a cultural moment. Writers weren't opposed to technology — they were opposed to being replaced by it. The industry needed a responsible alternative. **The principle:** Writers should write. AI should not. Every tool was evaluated against one question: does this help the writer work, or does it try to do the work for them? AI that writes scenes or dialogue removes the writer from the craft. AI that accelerates ideation, sharpens structure, and builds out worlds faster — that's a force multiplier. Same way Photoshop didn't replace photographers, Talemaiker doesn't replace screenwriters. **What we built:** - **Screenwriting Assistant** — Full script analysis and structural editing, conflict and resolution mapping across acts, character arc and motivation analysis, pacing and rhythm diagnostics. A full-time developmental editor available at any hour. - **World Builder** — Scene concept development from raw ideas, character suggestion and relationship mapping, plot and subplot ideation, act structure and story architecture, rapid world and setting research. Start with a logline, a synopsis, or notes scribbled on a cocktail napkin — what used to be weeks of work becomes hours. - **Ideation Engine** — Double-twist framework that generates creative alternatives on any concept. - **Character Art** — Character sketches and visual references from script descriptions, natively via DALL-E or exported as Midjourney prompts. - **Storyboard Generator** — Scene-level image generation in four visual options per panel. - **Research Mode** — Deep contextual research integrated directly into the writing workflow. **Input flexibility:** Full screenplays, loglines, synopses, character concepts, scene fragments, rough notes, research prompts, world concepts. The system adapts dynamically to whatever the writer brings. **Workflow acceleration:** Traditional development takes 3–4 weeks. With Talemaiker: 8–12 hours. Approximately 30 pages of development output per session. **Industry validation:** Tested with 12+ working Hollywood professionals — writers, producers, and directors with $2B+ in combined box office earnings. One total piece of negative feedback. It was fixed. **Testimonials:** - "As a tool for creativity, Talemaiker is impressive and already doing more than other platforms I've tried. This could be what writers need in this burgeoning wild west of AI." — Arne Schmidt, Writer (Hunter Killer, Chain Reaction), Producer (Robocop, Big Fish) - "I am so impressed by Talemaiker as a tool writers can use to maintain their competitive edge." — Jonathon Komack Martin, Producer (Deadpool Franchise) - "As the industry emerges, Talemaiker is the ally we need for development." — Walt Becker, Director (Van Wilder, Wild Hogs, Alvin and the Chipmunks) **Outcome:** Acquired by a Hollywood entertainment production company with active development operations, January 2025. First AI toolsuite purpose-built for responsible integration with WGA workflows. Turnkey acquisition including codebase, brand rights, and interim CEO with active industry deals. --- ### Locus + Atrium — Agent Management Infrastructure URL: /case-studies/locus-atrium/index.html Category: Agent Infrastructure / Platform Suite / 2026 Agent management infrastructure for individuals, small teams, and enterprise. Two surfaces, one underlying model — from a solo operator's desk to an org chart, without rebuilding anything. **The problem:** Agentic AI works in demos. In production, it falls apart at the coordination layer — which agent did what, when, with whose data, on whose authority, with what cost. Teams that try to scale past a single agent hit the same wall: no visibility, no audit, no control. Impressive demos that can't survive contact with a real organization. **The principle — three commitments:** 1. Legibility over magic. Everything an agent does is auditable in plain text. No opaque state, no decisions you can't review later. 2. Authority is explicit. Every agent has a defined scope. Widening it requires a human decision, recorded. 3. The right tool for the size of the team. A solo operator needs a different surface than an enterprise — so two surfaces, sharing the same model underneath. **What we built:** - **Atrium** — Desktop-based agent management for individuals and small teams. Cloud-backed for persistence. Local-first for speed. Manages a full team of agents from one desktop application with full audit trail and plain-text state across every agent. Ships summer 2026. - **Locus** — Enterprise agent infrastructure. Multi-team, multi-tenant operations. Org-level authority and audit. Built for the organization where agents need to be managed the way employees are — with identity, accountability, and control at scale. **Key architectural feature:** Atrium and Locus share the same underlying architecture. A team starts with Atrium and graduates to Locus without rebuilding their agents, their data, or their workflows. **Four architectural layers:** 1. Identity & Authority — Every agent has persistent identity (name, history, relationships, defined authority scope). Widening scope requires recorded human decision. Audit trail spans the agent's full lifetime. 2. Operations & Audit — Agent state lives in markdown. Readable, diffable, version-controllable. Every action timestamped and attributed. Replay and inspection are first-class operations. 3. Coordination — Handoffs agent-to-agent when work changes character, agent-to-human when decision requires authority or judgment beyond scope. Shared memory persists across the full agent team. 4. Surface — Atrium (desktop, cloud-backed, solo/small team) vs Locus (multi-team, multi-tenant, enterprise RBAC). Same model underneath. **Status:** Atrium shipping to individual operators and small teams, summer 2026. Locus in deployment for enterprise environments. Available for evaluation engagements, design partnerships, and enterprise rollouts. --- ### Agent Workshops — AI Workshop Series URL: /case-studies/agent-workshops/index.html Category: AI Workforce Training / Workshop Series / 2026 Focused AI workshops for non-technical teams — one topic, one to two weeks, every engagement ends with something deployed in your actual job. No engineering background required. Field-tested across twenty-three non-engineers before productizing. **The problem:** Most AI training produces vocabulary, not capability. Teams leave knowing what an agent is. Few leave with one running on their actual job. **The principle:** The workshop changes what you can do. Two formats were run in parallel during the validation cohort — weekly plenaries and hands-on topic-based workshops. Both were valuable, but one was transformative. The workshop format — narrow topic, fixed outcome, real builds against the participant's own work — was where the leverage lived. Participants who built left with capabilities. Participants who only learned left with vocabulary. **The offering:** - **Single-topic workshops** — one- to two-week engagements, one business goal, hands-on construction against the team's actual work. Every workshop ends with the participants having shipped something they couldn't have shipped going in. - **Full six-week cohort** — for organizations committed to building agentic AI fluency across a whole team at once. Two parallel tracks, 24 sessions, 19 workshop topics. **19 validated workshop topics include:** - Agentic AI fundamentals — build the first working agent - Software development by non-software developers - Comparing AI and PKM stacks — when to use what - Building a live knowledge base from your AI conversation history - Website design and deployment with Claude Code - Philosophy of agents and non-human intelligence **Four workshop stages (each available as a stand-alone engagement):** 1. Foundations — build the mental model and first working agent 2. Single-Agent Work — point one agent at one real job, real outputs by end of week 3. Multi-Agent Patterns — handoffs, review loops, escalation gates 4. Production — trust boundaries, reliability, handoff documentation for colleagues **Field validation:** A six-week intensive with twenty-three non-engineers produced production agentic systems — including a 3-agent autonomous pipeline (crawler, enrichment, quality agents running on cron) built and operated by a participant with no software development background, and a live language-learning web app built and shipped during the cohort. **Outcome:** Workshop format available now under the Rogue Agents banner — one engagement at a time, built to a single business goal, built around the team's actual work. --- ## Contact Pitch a project: /contact.html