# Gramercy Studios > Boutique consultancy specialising in AI-native and agentic applications. ## About Gramercy Studios designs and builds AI-native and agentic applications — software where AI is the interface, not a feature bolted on. ## Work - [Mosaic](/work/mosaic.md): AI-native interactive persona platform. Instead of reading a document about someone, visitors have a conversation with their AI persona — in first person, in their voice. Same data serves four surfaces: conversation, visual display, external AI tools via MCP, and agent-first web discoverability. - [Neo](/work/neo.md): Autonomous AI agent with a web interface for exploring and explaining any public GitHub codebase through natural language conversation. A hybrid architecture where Next.js handles the frontend while a Cloudflare Worker orchestrates agent execution in ephemeral Sandbox containers — solving the fundamental incompatibility between autonomous agents and serverless. - [Revision3 — SBX Crypto iGaming Platform](/work/revision3.md): Senior Software Engineer at Revision3, a software development agency specialising in blockchain gaming solutions. Full-stack engineering on SBX — a crypto iGaming casino and sportsbook startup. Multi-faceted role spanning admin dashboard development, consumer frontend, and backend microservices. Pioneered AI-assisted development workflows that codified repetitive patterns into documented, autonomous processes. Web3 technical advisor to the wider team. Full-time role, Jun 2024 – Dec 2025. - [ANZ Digital Asset Services](/work/anz.md): Embedded as lead frontend engineer on ANZ's digital asset services pilot — real-world asset tokenisation, portfolio management, and marketplace services for transacting tokenised assets with tokenised currencies across blockchain networks. Led a team of four frontend engineers, focused on distilling Web3 concepts for non-blockchain-native institutional users. Contract engagement with one of Australia's big four banks. - [BricktOrigins](/work/bricktorigins.md): dApp frontend for Abnormal Studios' product-driven Ethereum NFT collection — five individual NFTs with phased whitelist and public mints, a dashboard with reveal mechanics, and a Forge ceremony where holders burn all five to mint The Monument. Built with React, Next.js, and RainbowKit. - [Web3 Jobs](/work/web3-jobs.md): Web3 job board aggregating 1,000+ listings across blockchain, smart contracts, and cryptocurrency roles. Algolia instant search with filtering by location, category, type, and experience level. OpenAI for unifying job data from multiple sources into a consistent format, Firebase for storage and database. - [Dwight Club](/work/dwight-club.md): Proof of concept Ethereum turn-based play-to-earn NFT game. Mint a hero character, battle Dwight Schrute, and watch your NFT's on-chain metadata evolve with each fight. Chainlink VRF for verifiable randomness, IPFS for image storage, deployed to Rinkeby testnet. ## Thinking - [Trust, Autonomy, and Principles](/thinking/trust-autonomy-principles.md): Give agents principles, not scripts. Trust comes from authentic data, not prompt constraints. The evolution from explicit guardrails to principle-based prompting — and why the data layer is where trust actually lives. - [Content Engineering for the Agent Era](/thinking/content-engineering.md): How you structure data for AI consumption determines how agents present it. WAAG over RAG — reasoning at write time, not query time. Freeform content over rigid schemas. And it all starts with the human providing quality data, not the engineer or the agent. Content engineering sits alongside prompt engineering and context engineering as a distinct discipline. - [Designing for Multiple AI Consumers](/thinking/designing-for-multiple-consumers.md): MCP lets any AI tool access your data — Bring Your Own LLM. The value: your product reaches every model and every client without rebuilding for each. The trade-off: when you give up control of the LLM, data quality and tool design become your only control surfaces. What MCP enables, what changes, and what compensates. - [The Dual-Voice Architecture](/thinking/dual-voice-architecture.md): Intent drives the agent's voice, capabilities, and purpose. One interface, one conversation — the agent reads what you want to do and adapts. Visitors hear the persona speak as itself; owners get a collaborative helper. No mode switch. No dashboard. The agent replaces the CMS entirely. - [AI-Native Interfaces — Conversation Meets Visual](/thinking/ai-native-interfaces.md): Conversational alone isn't enough. The future of AI interfaces weaves conversation with a visual surface — the agent doesn't just talk, it shows. The product spectrum from traditional to AI-native to agentic. The portal concept: conversation adds dimensionality to web content. How the agent-UI relationship varies — from controlling the surface to providing depth behind it. - [Multi-Surface Discoverability — SEO for the Agent Era](/thinking/multi-surface-discoverability.md): Making web content discoverable by AI agents is the next SEO. Four surfaces, three implementation layers, and a practical framework any website can follow today. The most immediately applicable of these articles — the framework is product-agnostic. - [Evaluating Conversational AI Agents](/thinking/evaluating-conversational-agents.md): Traditional software has deterministic tests. AI doesn't — the same prompt produces different responses every time. LLM-as-judge uses one model to evaluate another against rubrics that encode what "good" means for each behaviour. The hardest part isn't building the framework. It's deciding what to measure — and discovering that agent quality and content quality are different problems. ## Accessing Content Each piece of content is available as markdown via two methods: **URL suffix** (recommended — discoverable and linkable): ``` https://www.gramercystudios.com/work/{slug}.md https://www.gramercystudios.com/thinking/{slug}.md ``` **Content negotiation** (standards-compliant): ``` curl -H "Accept: text/markdown" https://www.gramercystudios.com/work/{slug} ``` ## Contact - Email: andrew@gramercystudios.com - Web: https://www.gramercystudios.com - LinkedIn: https://www.linkedin.com/company/gramercystudios