David Liew

Most people use AI tools.I build AI teams.

Software now has agency and personality.

The paradigm shift: from tools to colleagues

Claude Code orchestrationMCP-powered agentsTrust engineering

Building AI systems that work like capable colleaguesβ€”
with context, initiative, and boundaries.

Context Engineering

Architecting what AI knows before it reasons

πŸ“‹

System Prompts

CLAUDE.md files
The AI's persistent memory

πŸ”§

Tools & MCP

External capabilities
What the AI can do

πŸ’¬

Conversation Design

Information flow
How understanding emerges

Prompts are just one part of context. The real skill is designing the entire informational environment.

I don't write prompts. I architect the context where AI insight emerges.

See How I Work

Rather than tell you what I can do, here's what I'm actually building. You're looking at one of them right now.

This Portfolio Site (Meta Example)

Live Now

You're looking at it. Built to embody the Fitzgerald Principle - holding opposing ideas (simple surface + rich depth) in productive tension.

What's Interesting:

  • β€’ Next.js 15 + React 19 (upgraded today!)
  • β€’ Philosophy validation scripts that score design decisions
  • β€’ Zero hardcoded content (all JSON-driven)
  • β€’ Automated quality gates before deploy

Still Figuring Out:

  • β€’ How to improve philosophy score (currently 7%)
  • β€’ Better ways to showcase technical depth
  • β€’ Balance between showing code and keeping it clean
Next.js 15React 19TypeScriptFramer MotionPhilosophy Automation

WhatsApp MCP Server

Active

Created an MCP server that turns WhatsApp exports into insights. Turns out I was missing patterns in thousands of client messages.

Discovery: Most valuable insights came from messages I thought weren't important at the time.

PythonMCP ProtocolClaude API

3 Client Projects

Learning Mode

Working with 3 businesses on AI-powered automation. Can't share details yet, but here's what I'm discovering:

  • β€’ Most "AI projects" fail because they automate the wrong things
  • β€’ Time-to-first-success < 2 minutes is make-or-break
  • β€’ Simpler systems win, even when they feel too simple

Each project teaches me something for the next one. Still figuring out the patterns.

Building and learning in public. If something I'm exploring helps you, that's enough for me.

Got a task that takes too long?

I'm exploring how AI can help delete it from your day. If something I'm discovering might help, let's compare notes.