Notes

My current AI stack

2 min read

Current AI tech stack

As a (constantly evolving) AI-native builder, I use Zapier’s AI fluency rubric to measure how I stack up in my AI usage and then use it against to measure how I can level up.

Imagine what Coinbase aims to be: a company with humans centred around intelligence. Think of what role you would occupy in such a future organization, and work towards becoming that person.

My current AI usage stands at:

  • Ship prototypes using multi-agent coding (Claude Code for code generation + Codex for code review) with adversarial review
  • Using ChatPRD for writing PRDs - but also use Claude Code to optimize prompts that can double as PRDs
  • Using Refero for design is an inspiration
  • Build design systems using Paper/Figma Make
  • Integrate coding agents like Amp and + Factory AI droids into workflows
  • Test UI/UX flows using Playwright
  • Capture client meetings and user interviews with AI (Granola/Circleback) and auto-extract decisions and action items > email them to the client right after. Once I hit a certain threshold of knowledge, the natural next step would be to build a knowledge base/RAG or maybe even train an LLM so they can provide me with greater context on client relationships going forward.
  • Use voice-first workflows (Wispr Flow) for rapid and more effective prompting and drafting texts - this is one product I can not live without
  • Research augmentation using deep research tools (GPT Deep Research) for structured insight synthesis
  • Run cron jobs for recurring workflows (summaries, syncing, cleanup)
  • Generate brand-aligned PowerPoints and Excel models using Claude Code (However, these are never truly client-ready and require a lot of work to polish. I would expect the frontier models to get there in a few more updates!)
  • Use Do browser agents to automate repetitive web tasks
  • Using Remotion to make videos for new product launches
  • Get Claude Code-generated email summary over Google Drive, emails, and past notes before the start of the day
  • Automated daily briefings via Claude Skills across Slack, email, calendar, and Drive
  • Design structured Claude skills to improve LLM output quality and consistency

Core Claude.MD skills

  • Design skill on GitHub
  • Coding workflow on GitHub

AI tools I am still learning to use:

  • Sauna.AI
  • Sub-agents in Claude Code
  • Palantir Ontology
  • Conductor for multi-agent orchestration
  • Devin by Cognition Labs
  • Antigravity/Augmented IDE - for now, I am somewhat indifferent to IDEs, especially given that Codex and Claude Code UIs are tolerable for now
  • multi-agent orchestration looks like without agent drifting
  • Braintrust for agent observability