Descriptions:
Wes Roth walks through a full hands-on build of “AI Pulse,” an automated AI news monitoring system constructed in a single afternoon using Oz — a cloud coding agent platform from Warp, the AI-powered terminal with over 700,000 developer users. Unlike tools such as Cursor or Claude Code that assist in real time, Oz runs agents in isolated Docker containers in the cloud, with cron-style scheduling, Slack and GitHub triggers, and live mid-task steering from a web panel or the Warp terminal.
The tutorial begins with creating three foundational reusable agent skills: a meta “skill creator” that generates correctly structured skill files automatically, a browser automation skill using Playwright and Puppeteer for headless browsing and dynamic scraping, and a YouTube summarizer using yt-dlp that handles captions, timestamps, and key entity extraction. These markdown-based instruction files serve as permanent playbooks, meaning every subsequent agent automatically inherits all three capabilities.
The main build spans three components: a backend API that researches AI stories via web search, scores them by trend velocity using Google Trends and Twitter/X signals, and generates summaries and tweet drafts; a Next.js frontend dashboard with heat-scored stories; and three scheduled Oz agents running every 3-6 hours. A standout Oz feature — cross-repo task execution within a shared environment — lets a single agent make coordinated changes across the API and frontend simultaneously. The finished system runs 24/7 with no human involvement after setup, illustrating cloud-native asynchronous agent orchestration as a practical, accessible workflow.
📺 Source: Wes Roth · Published February 20, 2026
🏷️ Format: Hands On Build
