Warp Launches Oz, Enabling Autonomous AI Agent Workflows
The rapid pace of AI development often leaves professionals scrambling to keep up. In a single afternoon, one developer leveraged Warp’s new Oz platform to build a sophisticated, autonomous AI news monitoring system, dubbed ‘AI Pulse.’ This system researches, ranks, summarizes, and drafts social media updates on AI advancements, all before the workday even begins.
What is Oz?
Oz is a cloud-based coding agent platform developed by Warp, the company behind the popular AI-powered terminal. Unlike local AI coding assistants that work alongside a developer, Oz is designed for autonomous tasks. It allows users to spin up multiple AI agents in the cloud, orchestrate their actions, schedule them to run independently, and trigger them via platforms like Slack or GitHub. The platform offers a centralized management panel for all agent activities.
Key Differentiating Features of Oz:
- Cloud Environments: Agents run in isolated Docker containers, providing access to necessary repositories without burdening local machine resources.
- Scheduling: Agents can be programmed to run at specific intervals (e.g., hourly, daily), performing tasks and delivering results autonomously.
- Steering Capability: Users can intervene mid-task, connect to an agent’s session via Warp or a web interface, provide course corrections, and then let the agent resume its work.
Building Custom Agent Skills
Oz’s functionality is enhanced by its ‘skills’ system, which allows developers to create reusable sets of instructions for agents. These skills act as playbooks, enabling agents to perform specific tasks consistently without repetitive prompting.
Skill 1: Skill Creator
Demonstrating a meta-capability, the first skill built was ‘Skill Creator.’ This agent is designed to generate new skills based on descriptions. It understands Oz’s skill file specifications, best practices, and directory structures. For example, when prompted to create a skill for checking stock prices, it generates the necessary files, including a skill definition, helper scripts, and a README, adhering to Oz’s standards. This effectively teaches the AI agent to build other AI agents.
Skill 2: Browser Automation
The second skill enables agents to interact with the web like a human user. Using headless browsers with tools like Playwright, agents can log into websites, fill out forms, take screenshots, and scrape dynamic data that is inaccessible via standard APIs. This skill opens up possibilities for automating tasks on websites that lack robust APIs or require interactive navigation.
Skill 3: YouTube Summarizer
To address the challenge of consuming video-first AI news, the ‘YouTube Summarizer’ skill was developed. This skill utilizes YouTube DLP to extract video transcripts and then employs AI to generate structured summaries. The summaries include a concise overview, section breakdowns with timestamps, key announcements, product mentions, and resources. For videos longer than 30 minutes, a ‘top five takeaways’ section is included. This skill allows for rapid assimilation of information from lengthy video content.
Constructing the ‘AI Pulse’ System
With these foundational skills in place, the ‘AI Pulse’ system was built. It comprises three main components:
- Backend API: An Express TypeScript API responsible for researching AI news via web searches, scoring stories by trendiness, generating summaries, drafting tweets, and sending alerts. It uses SQLite for storage and Twilio for SMS notifications.
- Frontend Dashboard: A Next.js application displaying the latest AI stories, their ‘heat scores,’ and ready-to-use tweet drafts.
- Scheduled Oz Agents: Three autonomous agents manage the system: one researches new stories every 3 hours, another generates tweet drafts every 6 hours, and a third performs daily maintenance, code cleanup, and dependency updates.
A key advantage demonstrated was the ability to set up a single Oz environment containing both the backend and frontend repositories. This allows agents to seamlessly work across both codebases, making changes that affect the API and the UI in a single operation. This cross-repository capability significantly streamlines development and coordination.
Autonomous Operation and Real-time Collaboration
The development process highlighted Oz’s power. While the backend agent was scaffolding the API, a second agent was simultaneously building the frontend dashboard. Both operated in the cloud, consuming no local resources. The ‘steering’ feature was showcased when the developer added a real-time search bar to the frontend after the agent had already started its work. This collaborative intervention allowed the agent to adjust its task without requiring a restart.
AI Pulse in Action
After 24 hours of operation, ‘AI Pulse’ successfully aggregated news from various sources like Reddit, Google DeepMind, MIT Technology Review, and TechCrunch. The system identifies trending stories, provides summaries, and offers pre-written tweet drafts for social media engagement. Breaking news, such as Claude Sonnet 4.6 outperforming benchmarks, triggers real-time alerts via SMS or Slack. The system’s ‘About Us’ page even notes its recursive origin: built by AI agents for AI news monitoring.
Why This Matters
Oz distinguishes itself not by having ‘smarter’ agents, but through its robust infrastructure for managing them. The ability to run multiple agents in parallel across different repositories, schedule autonomous workflows, and intervene with steering capabilities provides a powerful new paradigm for software development and automation. For developers hitting the limits of local agent execution, Oz offers a scalable cloud-based solution. The ‘AI Pulse’ project, built in an afternoon, now saves the creator significant time daily by automating the monitoring of AI news, proving the platform’s real-world utility.
Warp is offering Oz as a cloud coding agent platform, though specific pricing details were not provided in the demonstration. Setup is reported to take under 10 minutes for most repositories. The developer hinted at future plans, including a voice briefing agent, underscoring the platform’s potential for further automation.
Source: How to Build ANYTHING with Oz by Warp (YouTube)
