About
AI strategy without the theatre. Implementation that actually ships.
AI Strategist · Builder
Founder, Agency in a Box
Say Hello
hello@clivemoore.me
I run Agency in a Box, an AI implementation studio for small and mid-sized businesses that need real automation — not a chatbot bolted onto a landing page. The work covers AI strategy, custom GPT and agent builds, n8n workflow automation, and the unglamorous integration plumbing that makes any of it actually useful.
Alongside that, I'm building two products in public. Workilo is a productivity platform where specialized AI assistants hand work off to each other like a real team. GPT Studio turns static documents into working knowledge with GPT, Claude, and Cohere integrated. Both are pre-revenue. I post what's shipping, what's broken, and what I learned — instead of the polished founder narrative.
The 25 years of design and brand leadership before this aren't decoration. They're why I can sit down with a hospital, a mortgage broker, or a fintech, translate what they actually need, and ship a system that fits the business. The projects below are that body of work.
Projects
GPT Studio
Enterprise-grade document collaboration platform with integrated AI assistants that transform static documents into strategic business assets.
Project Highlights
- Designed multi-model AI orchestration layer routing tasks across GPT, Claude, and Cohere based on task characteristics — invisible to users
- Built layered context management system (document, workspace, conversation) enabling persistent AI memory across sessions
- Prototyped and iterated in production code (Express/TypeScript/PostgreSQL) — design decisions validated against real latency and token constraints
- Delivered 70% time savings in automated workflows through agent-driven document processing pipelines
- Designed extensible tool architecture supporting 100+ integrations, driven by observed user workflow patterns
Multi-Model AIContext ArchitectureEnterprise UXWorkflow OrchestrationSaaS Design
visit gptstud.ioDesign Process
The Core Problem
Most AI document tools treat every conversation as a blank slate. Users upload a file, ask a question, get an answer, and lose all context when they start the next task. For enterprise teams working across dozens of documents over weeks, this makes AI a novelty rather than infrastructure. The design challenge: how do you build a system where AI assistants maintain persistent context across documents, conversations, and team members — without overwhelming users with the complexity of what's happening underneath?
Multi-Model Orchestration as a Design Decision
Early prototypes used a single LLM for everything. Testing revealed that different tasks needed fundamentally different model characteristics — GPT-4 for nuanced analysis, Claude for long-document synthesis, Cohere for fast retrieval. Rather than exposing model selection to users (a technical decision they shouldn't need to make), I designed a routing layer that matches task type to model strength automatically. The user sees one assistant. Behind it, three models collaborate. The design principle: the complexity of orchestration should be invisible; the benefit should be obvious.
Context Architecture
The hardest design problem wasn't the chat interface — it was context windows. Enterprise documents are long. Conversations reference multiple files. Users expect the AI to "remember" what was discussed three sessions ago. I designed a layered context system: document-level context (what's in this file), workspace-level context (what files relate to each other), and conversation-level context (what we've discussed). Each layer feeds the AI differently. The UX challenge was making this feel seamless — users don't think in "context layers," they think "the AI should just know this." Getting that right required iterating through 4 major architecture changes over 6 months, each time simplifying what the user sees while adding complexity to what the system manages.
From Prototype to System
I built every prototype as working code (Express, TypeScript, PostgreSQL) rather than static mockups. This wasn't a philosophical choice — it was practical. You can't evaluate whether a multi-model AI system feels coherent from a Figma file. You need real latency, real token limits, real failure modes. The design emerged from building, testing with real documents, and redesigning based on what broke. The 100+ tool integrations weren't planned from the start — they grew from watching how enterprise teams actually tried to use the platform and designing extensibility into the architecture early enough to support it.
Workilo
AI-powered productivity platform featuring specialized AI assistants (Workalongs) that collaborate to deliver 10-15× faster completion of professional tasks.
Project Highlights
- Designed "visible autonomy" interaction pattern for 12 specialized AI agents — users see orchestration decisions as they happen
- Created Brand Brain: 10-question onboarding that auto-configures all 12 agents' voice, tone, and reasoning constraints from minimal user input
- Designed agent-to-agent handoff protocol with full decision transparency — users can trace any output back through the agent chain
- Built content pipeline orchestrating agents to produce 14 deliverables from a single brief, with visible relay points at each transition
- Improved output quality from 3.4/10 to 7.2/10 across 9 evaluation cycles by redesigning agent constraint systems rather than prompt engineering
Agent OrchestrationAgentic UXWorkflow AutomationSystems DesignMulti-Agent Systems
visit workilo.ioDesign Process
Designing Agent Roles and Autonomy
The first design decision was the hardest: how much autonomy should each AI agent have? Too little, and you've just built a chatbot with tabs. Too much, and users lose trust because they can't predict what the system will do. I started with 3 agents and a fully autonomous model — agents would hand work off to each other without user approval. Users hated it. They couldn't tell what was happening or why. The redesign introduced a visible orchestration layer: users see which agent is active, what it's working on, and what it plans to hand off. Autonomy happens, but it's narrated. This "visible autonomy" pattern was the breakthrough — the agents became 12 specialized roles (Workalongs) once I had a UX model that could scale trust alongside capability.
The Brand Brain Onboarding Problem
12 agents need to know your brand, your voice, your audience, your strategic positioning — before they produce anything useful. The naive approach: a 40-field configuration form. Nobody fills that out. I designed Brand Brain as a 10-question conversational onboarding flow where each answer cascades into configuration decisions across all 12 agents simultaneously. Question 3 (about audience sophistication) adjusts vocabulary constraints for the content writer, formality levels for the email agent, and technical depth for the research agent — all from one answer. The design insight: onboarding for a multi-agent system isn't about configuring agents individually. It's about capturing the minimum context that lets you configure all of them coherently.
Agent-to-Agent Handoff as an Interaction Pattern
The content pipeline — one brief in, 14 deliverables out — required designing a handoff protocol between agents. The research agent passes findings to the strategist. The strategist produces a brief for the writer. The writer's draft goes to the editor, then to the social media agent for platform-specific adaptation. Each handoff is a design surface: what context transfers? What gets summarized vs. passed verbatim? What does the user see at each transition? I designed the handoff as a visible "relay" — users can inspect what each agent received, what it produced, and why it made the decisions it did. This transparency wasn't in v1. It came from watching users get frustrated when agent #8 produced something unexpected and they couldn't trace back to which upstream decision caused it.
Iterating on Quality Through Evaluation Cycles
The security quality improvement (3.4/10 to 7.2/10 across 9 cycles) was a design problem, not just an engineering one. Each evaluation cycle required designing what "quality" meant for that agent's output, building the evaluation criteria into the agent's reasoning chain, and then redesigning the agent's tool access and constraints based on where it failed. The pattern that emerged: agents don't improve from better prompts alone. They improve from better-designed constraints — what tools they can access, what outputs they must validate before passing work downstream, and what failure modes trigger a retry vs. an escalation to the user. Designing those constraint systems was more impactful than any prompt engineering.
Element 6 Branding
Led a comprehensive rebranding initiative for Element 6, positioning it as a leading digital media agency within the ION Group of Companies.
Project Highlights
- Developed external client-facing brand identity
- Established brand guidelines and design systems
- Secured new clients including Hitachi, Marathon Mortgage, and LifeLabs
- Brought in $540K annual revenue from internal and external customers
BrandingNamingBrand MarketingMulti-channel CampaignsStrategic Planning
visit element6.io
SMART FS Product Launch
Development and branding of SMART FS product from conception to client acquisition in 6 months.
Project Highlights
- Shaped UX design for the new product
- Ran multichannel product campaigns
- Achieved product launch within a tight 6-month timeline
UX DesignProduct DevelopmentMarketing Strategy
visit smartfs.app
Marathon Mortgage Rebrand
Comprehensive rebranding and website redesign for Marathon Mortgage to enhance broker experience.
Project Highlights
- Created new logo and brand identity with vibrant colors and clean typography
- Developed custom website with CMS and embedded mortgage calculator
- Transformed brand perception in a conservative industry
- Increased website traffic and broker engagement
BrandingWeb DesignDigital Innovation
visit marathonmortgage.ca
PACE Cardiology Digital Transformation
Patient-first website redesign for PACE Cardiology that simplifies how Ontarians find, book, and prepare for specialized cardiac care across three regional clinics in Newmarket, Barrie, and Orillia.
Project Highlights
- Designed a dual-audience experience serving both patients and referring clinicians from a unified site architecture
- Integrated OceanMD e-referrals and AFib self-referral forms directly into the booking flow
- Built location-aware navigation across three Ontario clinics: Newmarket, Barrie, and Orillia
- Structured 14+ cardiac services and diagnostic tests into a plain-language, patient-friendly taxonomy
- Established a content system supporting patient education videos, provider FAQs, and clinical updates
UX DesignHealthcare Web DesignMarketing StrategyIntegrated MarketingDigital Innovation
visit pace-cardiology.com