Panel at AusFinTech 2026 - the legal, technical, and ethical scaffolding for AI that talks to customers.
Building AI agents that call humans requires more than good prompts. Here's the legal, technical, and ethical scaffolding.
Regulations like ACCC in Australia, FTC rules in the US, and GDPR in Europe all apply. They weren't written for AI agents, which makes them ambiguous. But the spirit is clear: disclose, get consent, provide an out.
State machines again. Recording and retention windows. Human escalation as a feature. All of this is in the code, not in the prompts.
The agent should never deceive. The customer should know they're talking to an AI. Manipulation is technically possible but ethically indefensible.
A walkthrough of the state machine, audio pipeline, and fallback design I use for Chasyr.
When RAG actually beats fine-tuning, when it doesn't, and how to tell which one you need.
The exact prompts and CI workflows my team runs on every PR. Copy-paste, MIT licensed.
The boilerplate I clone for every new SaaS bet. Auth, billing, RLS, AI hooks pre-wired.
How I restructured a 7-person team around AI tooling. Velocity numbers, cultural pitfalls, what worked.
A guest lecture at COMSATS on how mid-career engineers can move into architecture roles.
A handful of titles - short list, opinionated commentary, no affiliate nonsense.
My exact dev stack - IDE, terminal, AI agents, productivity hacks. Updated quarterly.
Papers, blog posts, and talks I send every engineer I mentor on getting up to speed with AI.
Remote-friendly companies, visa sponsors, OSS scholarships - things I wish I had a decade ago.