AI SDK
Vercel's TypeScript AI SDK for streaming conversations, tool calling, and model integrations in LegalAgent, Bitlauncher, and Masterbots.
Implementation
Wired LegalAgent's chat UI to the Vercel AI SDK with streaming responses and tool calls for legal workflows.
Built Bitlauncher's site chatbot with AI SDK tools for news, video, and structured product answers.
Automated Masterbots and Bitlauncher locale copy generation with AI SDK prompt templates.
LegalAgent
Software Engineer
A voice and chat AI assistant for legal professionals, combining case retrieval, document intelligence, and bilingual interactions.
Bitlauncher
Product Engineer
An AI and crypto launchpad combining token auctions, cross-chain wallets, market data, and AI-assisted project research.
Masterbots
A platform for domain-specific AI assistants with specialized chat interfaces, social features, and multilingual support.
Related posts
Context Engineering
An agent harness orchestrates memory, tools, skills, and retrieved information so models receive the right context at the right moment.
AI Agents are Infrastructure
Running agents in production means managing tools, state, permissions, and recovery—not just choosing a capable language model.
Agentic Commerce
Google's Universal Commerce Protocol, with A2A, AP2, and x402 around it, moves checkout into AI surfaces and makes shopping a workflow agents can execute.
Evolution of AI UX
AI interfaces are moving from prompts and answers to agents that hold intent, run workflows, and deliver outcomes, which makes visibility and control the core of the UX.
Why I'm Bullish on 2026
Why I expect AI agents, stablecoins, wallet signing, and tokenized markets to converge in products over the coming year.
Intersection of AI and Web3 (Part 2)
How agentic finance, stablecoin payment rails, and decentralized AI infrastructure are starting to connect in real products.
Agentic Finance
How AP2, x402, and agentic commerce protocols could let AI agents authorize payments and settle transactions with stablecoins.
Prompt Injection
How direct, indirect, and tool-response prompt injection hijack LLM behavior, and the layered defenses I use in AI applications that call tools.
See the work
Explore the project portfolio for implementation details, systems, and technology.