My Jev Pi Durable Agent CURATES News: HYPE, FACT, or SIGNAL
Proactive agents don't wait for you to work. They work for you, then wait. While everyone else is stuck prompting back and forth in the loop, the best engineers are moving from in the loop, to near the loop, to completely out loop. 🚀 PUSH YOUR AGENTIC ENGINEERING TO OUT LOOP Phase 3 early sign-up: https://agenticengineer.com/ia3-pre-launch?y=OLqP8jxJ0Wc 🎥 VIDEO REFERENCES • Durable News Agent codebase: https://github.com/disler/pi-durable-news-agent • Durable Pi video (previous video): https://youtu.be/DNDYYHsypZg Most engineers are still babysitting AI in a terminal while drowning in clickbait, slop, and attention theft. Stop waiting on your agents, and stop letting your agents wait on you. 🔥 Proactive agents don't wait for you to work. They work for you, then wait. While everyone else is stuck prompting back and forth in the loop, the best engineers are moving from in the loop, to near the loop, to completely out loop. 🤖 In this video, IndyDevDan breaks down how to filter through AI hype, facts, and actionable signal using a custom Durable News Agent built on Pi Durable and the open-source pi agent harness. Instead of acting like another chat window, this system runs nonstop every single minute as a cron agent on an M4 Mac mini AI agent box and inside an exe.dev AI agent sandbox. 🔥 A proactive agent is a durable agent plus a trigger inside a sandbox. Because the agent lives behind a SQLite-backed server API, both the browser generative UI and the terminal TUI act as thin, detached clients synced in real time. My hands are off the keyboard, and the system is gathering, filtering, ranking, and curating high-signal engineering news for me without me. 🛠️ Code is now infinitely cheap and copyable. What separates elite builders today is taste, judgment, and system architecture. To protect your time and token budget, we put a fast, dirt-cheap jev classifier in front of our language models. At sub-half-second latency, jev gates irrelevant stories, deduplicates coverage, verifies claims across sources, and scores every item across three distinct buckets: hype, facts, and true actionable signal. 💡 Key takeaways: In, Near, Out Loop: The ultimate measure of your agentic engineering skill is how many autonomous threads of work run on your behalf around the clock. Proactive agents are the missing bridge between in-loop prompting and full out loop autonomy. Hype vs. Facts vs. Signal: Hype is unfounded excitement. Facts are objectively true (like a massive math proof dump) but not always actionable today. Signal is high-leverage information you can act on right now (like a new state-of-the-art model rollout). The Jev Pipeline: Using a jev classifier at every stage (gather, gate, dedupe, rank) keeps tokenomics razor thin so a cron agent can sweep every minute without burning cash on expensive LLM calls. Detached Client Architecture: Built in layers with SQLite, Pi Durable, a custom pi agent harness, and a unified SDK that powers the web generative UI, terminal TUI, CLI, and cron triggers across a local Mac mini AI agent or a cloud exe.dev AI agent sandbox. Data Flywheel and Taste Tuning: Upvote or downvote news cards, dynamically adjust keyword and source weights in plain chat, and let two-way prompting continuously sharpen what your durable agents surface next. 🌟 Don't let AI noise steal your focus, and don't lose faith in your software engineering fundamentals. Build the system that builds the system. Clone the Durable News Agent repo, wire up your own taste profile, and start transitioning your workflow from in the loop, to near the loop, to out loop. Stay focused and keep building. - IndyDevDan 📖 Chapters 00:00 Proactive Agents - In, Near, Out Loop 01:23 Durable News Agent 05:15 Build With Taste - Hype, Facts, Signal 06:58 Software Architecture Still Matters 08:25 Building Software in Layers 10:26 Jev Everywhere, Results Everywhere 11:17 Measure Your Agentic Engineering 12:25 Building Useful Product 18:27 Durable Agents Are Powerful #agenticengineering #softwareengineering #aiagents
