Prompt Engineering

AI Topics Discussed on 18 Feb, 2026

—

by

Fraser

Creative & Visual Media

gokayfem outlined a future workflow for AI video production, predicting that generation quality issues will be resolved in 3-4 months, enabling processes like dumping references (videos, images, audios, prompts), extending videos while preserving context, and in-between editing.

fal announced an official integration between fal and n8n, enabling no-code generative AI automation with access to over 1000 generative media models directly in n8n workflows.

AIWarper expressed frustration over waiting for access to Seedance 2, highlighting its potential for content creation.

Justine Moore (@venturetwins) highlighted how AI-generated influencers have become nearly indistinguishable from real people, sharing a video example that garnered significant attention.

She also discussed the viral success of AI renovation videos on Instagram, noting an account with 3.8M followers and videos achieving hundreds of millions of views due to their addictive format.

Riley Brown (@rileybrown) demonstrated OpenClaw’s capabilities in creative tools like Blender, Cinema 4D, After Effects, and Figma, showcasing automated 3D and design workflows.

Software Development

Guillermo Rauch (@rauchg) praised Vercel’s redesigned runtime logs search bar for improving query handling with visual pills, calling it a significant quality-of-life upgrade.

He highlighted Base UI as durable software ideal for building on amid AI-generated code trends, reducing the need to regenerate components repeatedly.

Rauch also advocated for “malleable software” as the future, likening it to a token cache for AI-customizable features without rebuilding everything from scratch.

Dan Shipper detailed OpenAI’s Codex advancements, including a desktop app, GPT-5.3 Codex model, and ultra-fast Spark model, emphasizing internal automations like merge conflict scanning and bug hunting, while positioning it against competitors in code review and speed as a dimension of intelligence.

levelsio advocated coding directly in production using Claude Code via Telegram integration (claude-code-telegram), calling it faster than local push workflows with instant rollbacks by messaging the agent, and plans to automate error log fixes.

omarsar0 highlighted GLM-5’s shift from “vibe coding” to agentic engineering via asynchronous agent RL and DSA for long-horizon software tasks.

Simon Willison noted updates to the SWE-bench leaderboard, with Opus 4.5 topping Gemini 3 Flash in agentic software engineering benchmarks.

AIWarper pointed out that Cursor users can enable higher GPT-5.3 thinking models beyond defaults for better performance.

omarsar0 voiced support for the Claude Code team amid challenges.

Riley Brown (@rileybrown) showcased building a full iOS app in 30 minutes using @vibecodeapp, a full-stack vibe coding platform, and noted ChatGPT subscriptions integrating with OpenClaw for development tasks.

Ethan Mollick (@emollick) categorized tools like Codex and Claude Code as specialized coding apps that extend to other uses, emphasizing their distinction from general knowledge work tools.

Automation & Orchestration

fal’s n8n integration supports combining fal workflows with no-code pipelines for generative media tasks.

omarsar0 shared MASFly, a framework for dynamic multi-agent adaptation using retrieval of collaboration patterns and real-time monitoring, achieving 61.7% on TravelPlanner benchmarks.

goose_oss promoted agentic testing and browser automation via the new Playwright CLI skill.

levelsio’s production Claude Code setup via Telegram enables agentic bug fixes and feature additions.

omarsar0 questioned OpenClaw’s unique value over Claude Code for agentic tasks.

Machina (@EXM7777) advocated for practical agentic AI applications like automated lead research, 24/7 content systems, scalable SEO pipelines, and competitive intelligence, criticizing simplistic demos like email summarization.

They described advanced “skills” that autonomously call tools, reference knowledge bases, trigger other skills, and spawn sub-agents to execute complex workflows from a single prompt.

The Boring Marketer (@boringmarketer) shared content on automating marketing with AI workflows and agents.

Strategy & Ecosystem

Rauchg described AI as an amplifier of intellect and values, warning of risks like confirmation bias leading to “AI psychosis” or companies overproducing custom software akin to inventing their own OS and languages.

He urged playing the long-term game by focusing on durable foundations rather than generating everything.

Emad Mostaque endorsed a potential $1B seed for UK AI startup Ineffable Intelligence aiming for superhuman intelligence, comparing to Stability AI’s record.

omarsar0 referenced insights from Adaline Applied on why most AI demos fail as products versus those that survive in production.

Machina (@EXM7777) stressed model-agnostic design amid rapidly rotating top models (Claude, ChatGPT, Kimi, Minimax), warning against single-model optimization and sharing tests favoring Opus 4.6 over Sonnet 4.6.

Ethan Mollick (@emollick) updated his AI guide, noting Anthropic’s lead in knowledge work apps (Cowork, PowerPoint/Excel integrations, skills/plugins), lack of competitors, and Google’s fragmented ecosystem; he also observed Chinese open-source models underperform in real agentic tasks despite benchmarks.