Now

A living page, in the spirit of nownownow.com. Updated when my attention shifts, not on a schedule.

Last updated: July 2026.

Work

Sustaining the True Platform after launch. The public window opened in April; now it's the less glamorous part: holding the release rhythm week over week, and closing the gap between what the product claims and what it actually does on real customer work (test generator, AI runner, requirement analysis).
Running the agentic pod and the Kai agents. Kai is the orchestrator that holds product intent and routes work to the right capability. Day to day that's steering the pod, keeping the roadmap honest, and prepping for Q1 FY27 planning in Vung Tau, mid-August.
Pushing agentic QA trust. Still the real problem, not "self-healing scripts": let the agent decide what to verify, make its uncertainty legible, keep the human as the safety net.

Publishing on this site more regularly. Recent pieces on giving Claude Code computer use, the deck my agents work on, and how agent frameworks are quietly converging on the plain-markdown setup I've run for months.
Rebuilding the site itself in public as I go: quieter, more personal, less landing-page.

TermDeck: a browser workspace for people who run coding agents in terminals. Live session grid, clickable file paths, a tweak mode that sends CSS selectors back to the agent.
Kensa: a QA copilot browser extension. Capture a repro, hand it to an agent, get a clean report back.
browser-harness / computer-harness: the forked tools I use daily to let agents drive a browser and the Mac without stealing my focus.
COG: the Obsidian + Claude Code + Git second brain that runs all of the above.

A personal briefing system that reads across Slack, Jira, GitHub, and meeting transcripts each morning and surfaces the two or three decisions I actually need to make. Rough at first; now load-bearing.
Sourcing public-domain museum art instead of AI-generated imagery for anything that needs a real picture, including this site.