Prompt Engineering

AI Topics Discussed on 31 Jan, 2026

—

by

Fraser

Creative & Visual Media

@HBCoop_ showcased multiple generative video workflows, including Midjourney images animated into cinematic clips using Veo 3.1 for a luxury Pacific Northwest home at sunset, another Veo sequence labeled “The journey begins,” a style reference clip (–sref 4707569743), Ray3.14 for underwater bioluminescence, and Hailuo 2.3 for a Tokyo stroll.

@javilopen continued a thread on using Gemini and Magnific (Nano Banana) to create photorealistic renders from raw 3D screenshots for a custom fireplace project, successfully removing obstructive elements like pipes while preserving architectural details, in collaboration with professional architects and interior designers.

@c_valenzuelab, RunwayML CEO, shared an image likely related to video generation advancements.

@levelsio shared excitement over Alibaba’s open-source LingBot-World, a world model rivaling Google’s Genie 3 with 10 minutes of stable interactive generation at 16fps compared to Genie’s 60 seconds.

@AIWarper demonstrated a simple SORA hack enabling AI-generated characters to appear in cameos within videos.

Software Development

@rauchg highlighted Vercel’s AI support agent achieving 87.6% autonomous resolution rate on cases, with plans for near-100% via fine-tuning, and emphasized how customer feedback will drive coding agents to enable autonomous software improvements.

@simonw highlighted that OpenClaw is built on Pi, a coding agent capable of self-extension, calling it the future of software writing its own software.

@danshipper pointed to an Anthropic study finding that poor use of AI coding tools reduces skill mastery, praising their transparency.

Automation & Orchestration

@rauchg announced general availability of Vercel Sandbox, a secure compute environment for AI agents now powering platforms like Blackbox AI, RooCode, and v0, with features like snapshotting and open-source SDK/CLI.

He also discussed agentic engineering in a fireside chat with @steipete (OpenClaw), noting how AI-native developers prompt complex systems like Next.js into existence, shifting focus to system architecture while agents handle implementation.

@omarsar0 promoted “Agentic Image Generation” via Claude’s Code Playground plugin, enabling self-improving loops for image generation with precise annotations, and offered a tutorial.

Strategy & Ecosystem

@javilopen stressed the low adoption of generative AI outside AI communities, demonstrated by professionals’ reactions to quick renders, positioning tools like Magnific as enhancers for visualization and client communication rather than job replacers.

@rauchg echoed that AI-native skills prioritize systems understanding over code-writing, as agents take over implementation.

@simonw emphasized Karpathy’s nanochat breakthrough, training a GPT-2-grade model for $73 in 3 hours on 8xH100s—a 600x cost drop over 7 years, or 2.5x annually.