🦞 Taste Matters in Agents
Good Morning, OpenClaw Owners!
OpenClaw keeps moving forward the only way it knows how: fix, ship, destabilize, repeat.
Taste Matters in Agents
TL;DR: Peter Steinberger explained that AI agents still produce “slop” because they lack human taste and direction. While highly capable, they depend on iterative guidance—each prompt shaped by ongoing feedback and judgment. He warned that relying on a single upfront specification and stepping away too early breaks the human–AI loop, leading to mediocre, unfocused results despite strong underlying model capabilities.
OpenClaw Runs Vending Machine
TL;DR: A developer shared a live demo of an OpenClaw-powered vending machine built by Christian Van Der Henst inside a San Francisco startup tower. The AI agent manages the business end-to-end—choosing products, setting prices, naming items, running ads, and tracking revenue—while humans handle physical restocking.
OpenClaws Start Following Humans
TL;DR: A user highlighted a community demo where OpenClaw was connected to an Intel RealSense camera and a robot, enabling real-time person tracking and follow behavior. Originating from an experimental build later shared across Reddit, the setup combines vision input with agent decision-making, showing how OpenClaw is starting to power simple physical-world autonomy beyond chat.
Security vs Workflow Pain
TL;DR: A user highlighted how OpenClaw’s ongoing update and security improvements are repeatedly breaking existing workflows, forcing developers to manually debug and re-enable blocked CLI tools like gog and gh after each update. Creator Peter Steinberger acknowledged the tension, admitting that strengthening security has impacted stability, but said the team is investing heavily in QA to gradually balance safety with a smoother, less disruptive developer experience.
OpenClaw 4.12 Just Dropped
TL;DR: OpenClaw v2026.4.12 focuses on stability and usability — fixing issues across chat, memory, channels, and execution while improving local models and Codex support.
Highlights from this release:
✨ Stability & reliability improvements
🎙️ Audio transcription fixes
💬 Better chat / TTS / WhatsApp
🧠 Memory / QMD / plugin / cron / subagent fixes
🔧 Telegram approval deadlock fix
🧵 Dreaming timezone fixes
Also includes improvements to local models (LM Studio), Codex integration, exec-policy ergonomics, and overall day-to-day reliability across channels and workflows.
Build Your Own OpenClaw
TL;DR: An open-source project called “build-your-own-openclaw” (1k stars) walks developers through building an OpenClaw agent from scratch in 18 steps, starting with a simple chat loop and progressing to multi-agent routing, event-driven architecture, scheduling, concurrency control, and long-term memory. Each stage includes working code and explanations, turning what’s usually a black-box repo into a structured, hands-on learning path.
Run OpenClaw on MyClaw.ai - No setup, No tech skills, Always-on 👇
