Prompt Engineering

AI Topics Discussed on 11 Feb, 2026

—

by

Fraser

Creative & Visual Media

Javi Lopez celebrated Seedance 2.0’s text-to-video capabilities, showcasing a dramatic hostage rescue scene and stating “It’s official: where are ‘there’. There is no going back.”

He emphasized the breakthrough in generative video models by hyping multiple examples including shampoo ads and creative parodies.

Heather Cooper shared Midjourney generations using style references, exploring artistic visuals.

– A.I. Warper reported that the Seedance 2.0 video generation model was removed from the BytePlus website, prompting questions on alternatives.

– A.I. Warper showcased Seedance 2.0 outputs converging on scenes like Master Chief vs. stormtroopers, noting clunkiness but impressive results for quick iteration.

– A.I. Warper predicted AI would complete the Demon Slayer anime adaptation as well as or better than studios before the 2029 release.

– omarsar0 highlighted a paper on accelerating diffusion LLMs.

Justine Moore (@venturetwins) highlighted Seedance 2’s advanced generative video features by recreating Winter Olympics events starring pandas, contrasting everyday UGC use with creative experimentation.

She attributed China’s edge in video models to less restrictive copyright policies during training and inference.

Ethan Mollick (@emollick) demonstrated the model’s prowess with prompts like an otter piloting a mech into battle and flying an airplane in a nature documentary.

Software Development

Guillermo Rauch announced Vercel Flags to support growing engineering teams including agents, aiding AI-assisted development.

fofr demonstrated OpenClaw with Gemini 3 Flash creating a fully functional Game Boy Snake game in Z80 assembly, complete with emulation testing, highlighting coding agents’ prowess.

– Dan Shipper discussed OpenAI’s Atlas browser, where the engineering team used Codex for over half the codebase, aiding Chromium navigation, prototyping, and UI effects.

– Simon Willison shared notes on OpenAI’s new Skills feature in their API, experimented via the shell tool.

OpenRouterAI (@OpenRouterAI) revealed GLM-5, a 744B parameter model from Z.ai optimized for coding and agentic use cases, achieving state-of-the-art benchmarks and now available via their platform.

Machina (@EXM7777) critiqued Opus 4.6 for excessive overthinking and shared an optimal system prompt for Deep Research in ChatGPT.

Automation & Orchestration

Rauchg introduced advanced features in Vercel Sandbox, including network isolation via egress policies to secure agent workflows and prevent data exfiltration.

He also highlighted self-driving infrastructure updates like runtime logs and CLI filters for agent debugging and orchestration.

– goose_oss introduced goosetown, a multi-agent workspace manager inspired by Steve Yegge’s Gas Town, for distributed async subagents.

– goose_oss emphasized workflows over one-shot prompting for real agent builds, addressing context failures.

– Dan Shipper explored agentic browsers like Atlas, balancing autonomous task handling with user control.

Machina (@EXM7777) noted that activating OpenClaw agents is straightforward, but configuring granular data access permissions poses significant privacy challenges.

POM (@peteromallet) discussed VibeComfy, open-source tools enabling agents to interpret and execute ComfyUI workflows via natural language, reducing friction for non-technical users.

Ethan Mollick (@emollick) showed Claude Cowork one-shot analyzing 107 complex business documents (PPTs, Word, Excel) for a Wharton case study.

Strategy & Ecosystem

Rauchg reflected on the obsolescence of manual programming around 2026, comparing it to mechanical watches after centuries of refinement, signaling a major trend in AI-driven development.

fofr and Alex Volkov noted accelerating AI capabilities, urging adoption of tools for competitive advantage amid rapid advancements.

– jfischoff remarked that AI’s endgame involves robots and chatbots for elderly care, tying into calls to pivot to healthcare.

– levelsio touched on AI-driven job losses exacerbating unsustainable welfare, pushing wealthy toward space-based escapes.

Ethan Mollick (@emollick) explored GenAI “poetry tastes,” finding ChatGPT, Gemini, Claude, and others converging on Borges’ “The Golem” to metaphorically describe their capabilities and limitations, with secondary picks like Rilke’s “Archaic Torso of Apollo.”

He noted mismatched perceptions: some underestimate AI’s transformative work impacts, while others overlook deployment hurdles.

OpenRouterAI promoted dynamic model routing (e.g., Kimi K2.5 vs. Opus) via Auto Router for optimized workflows.