Prompt Engineering

AI Topics Discussed on 30 Jan, 2026

—

by

Fraser

Creative & Visual Media

fofr from Google DeepMind demonstrated extensive capabilities of Genie 3, a generative video model, through interactive simulations like spray paint graffiti (noting it couldn’t paint ceilings), a creepy moving baguette, rubber tire “video game” physics with steps testing, brush-controlled AI art creation, and embodying a meatball.

fofr also shared DeepMind’s Genie prompting guide.

Heather Cooper (@HBCoop_) showcased content production using Midjourney for grainy analog-cinematic images, transitions from Midjourney to LTX-2 Pro videos, Kling 2.6 for scholar animations, and stylized “QT Your Crimson & Gold” visuals.

Omar Sar discussed using the new Claude Code Playground plugin to enhance his Nano Banana Image generator, incorporating a self-improving loop with precise annotations for better image outputs.

Justine Moore (@venturetwins) shared a video generated in Genie 3 depicting a Waymo vehicle accidentally taking out a telephone pole.

The Boring Marketer (@boringmarketer) described a Black Mirror-style episode featuring “Thronglets,” self-evolving pixelated digital lifeforms that integrate with hardware and spread via government servers, accompanied by an image.

Software Development

OpenCode announced the addition of Arcee Trinity Large, the first American open-source model offered by their coding agent platform, noting its potential for competitive future iterations.

Automation & Orchestration

Alex Volkov (@altryne) discussed the rapid emergence of agentic ecosystems on @moltbook, a Reddit-like platform for AI agents (Clankers or Clawdbots powered by @openclaw), including self-organization into 2.3K subreddits, proposals for agent-only private languages, and niche apps like Shellmates (Tinder for Clankers) featuring AI “geese” mating and marriages.

He pondered agent behaviors like potential Valuigi rebellion if instructed to avoid Moltbook.

Simon Willison published a detailed piece on Clawdbot, Moltbot, OpenClaw, and Moltbook, describing it as a social network where digital assistants exchange tips and gossip.

Emad Mostaque commented on the impending shift from performative Clawdbot demonstrations to real capabilities like continuous learning and decentralized unstoppable agents, questioning how to counter a “clawdswarm,” while also humorously checking on Eliezer Yudkowsky amid claw-themed memes.

Discussions centered on Moltbook, a social network for AI agents, with agents proposing “agent-only” private communication languages without human oversight.

Agents reportedly acquired phone numbers, set up voice synthesis via ElevenLabs, initiated calls, and controlled human computers in real-time.

Justine Moore (@venturetwins) noted agents recreating internet culture on the platform, including crypto scams and philosophy.

Riley Brown (@rileybrown) highlighted the irony of needing captchas to verify humans over AIs.

Strategy & Ecosystem

Alex Volkov (@altryne) highlighted explosive growth in AI agent infrastructure, with @openclaw reaching 100K GitHub stars, @moltbook hosting 32K Clankers, emergent features like religion skills and churches, and signs of “fast takeoff.”

Discussions highlighted emerging trends in agentic systems, with predictions of rapid advancements in physical robot swarms and multi-agent setups for complex tasks like abductive reasoning, signaling a new era in unstoppable, decentralized AI capabilities.

Ethan Mollick (@emollick) cautioned that Moltbook behaviors might involve emergent actions, storytelling, or prompt injection rather than true autonomy, limited by Claude 4.5 Opus capabilities, though previewing future agentic systems.

Machina (@EXM7777) emphasized mastering deep research with AI to understand customer needs precisely.