How to build long-running Claude agents

#1 𝕏

Logan Kilpatrick reports that nearly 200,000 apps built in Google AI Studio were deployed and shared for free in the past month, underscoring that everyone can bring their ideas to life.

#2 𝕏

Teresa Torres: Priya at Override Labs began with a misuse-first mindset and hard-coded safety rules that pre-screen conversations as red or yellow flags before any AI runs. The system never gives a positive signal and always highlights what true consent entails.

#3 𝕏

Peter Yang demonstrates how to build a long-running Claude agent from scratch and reveals how Anthropic teams deploy these overnight agents to map codebases, synthesize user feedback, and pressure-test product decisions.

#4 𝕏

Sebastian Raschka shares a hands-on walkthrough for running local coding agents with open-weight models like Claude Code or Codex entirely offline. He also includes a checklist for evaluating model suitability, covering long-context RAM usage and prefill performance.

#5 𝕏

Ali Ghodsi reports that Genie Code has just crossed the 50% mark for code generation on Databricks, and AI‐written code now outpaces human authors by 3×.

#6 𝕏

Guillermo Rauch warns that AI agents’ non-deterministic, multi-step, distributed design makes debugging a nightmare, so the Vercel team shipped out-of-the-box observability for Vercel Agents, earning positive early feedback.

#7 𝕏

Harrison Chase shared a free 3-hour YouTube Deep Agents course that dives into task planning, file-system-based context management, subagent spawning, and long-term memory techniques.

#8 𝕏

Peter Yang discovered @paper for YouTube thumbnail creation and praises its “select a few images and then generate more variations” feature as a game-changer, thanks to a tip from @rileybrown.

#9 𝕏

clem 🤗 urges PM builders to take the next step by post-training their own open-source models for tailored AI capabilities.

#10 𝕏

Aravind Srinivas says enterprises will spin up proprietary model–harness–sandbox–eval flywheels optimized for token value per watt, leveraging their unique tacit knowledge of domain and customer workflows.

#11 𝕏

claire vo 🖤 told VP+ product, engineering, and design execs at startups and 150k-employee enterprises to measure AI adoption to get it right and make it better. She also warned they’re likely under-consuming tokens.

#12 𝕏

Guillermo Rauch warns that with today’s limitless engineering options, human judgment is essential for deciding what to build and which architectures to use.

#13 𝕏

Kevin Yien highlights that this compact new reader automatically syncs with his Readwise account, features a Readwise-powered “x-ray” for deeper context, and is smaller than a Kindle.

#14 𝕏

Logan Kilpatrick says that since Google I/O, you can deploy to Google Cloud for free without even needing a billing account.

#15 𝕏

Harrison Chase enabled cache-aware requests in Deep Agents, reusing a warm cache to slash cache misses and drive down operational costs.