Prompt Engineering

AI Topics Discussed on 26 Jan, 2026

—

by

Fraser

Creative & Visual Media

Heather Cooper (@HBCoop_) shared a video demonstrating storybuilding using Freepik with Nano Banana Pro and Veo 3.1, highlighting tools for content creation.

fofr (@fofrAI) posted AI-generated images themed around “Travel,” showcasing four visuals, and separately shared art discovered using NBP (Nano Banana Pro).

ilker (@ailker) announced a new skill for Claude-code that enables natural language generation of images, videos, audio, and workflows via FAL.

gokay (@gokayfem) shared an image, likely related to generative media given their background.

– @fal announced the availability of Qwen3-TTS on their platform, highlighting voice cloning from just 3 seconds of audio, free-form voice design across multiple models, support for 10 languages, and fast inference.

– @AIWarper praised the new Ray 3.14 model on Luma Labs for native 1080p video generation, noting its speed, cost efficiency, and potential to unlock indie video creation similar to indie games.

– @AIWarper shared excitement for the open-source LTX-2 image-to-video adapter LoRA trained on 30,000 videos to enhance video model robustness.

– @goose_oss demonstrated using Remotion Skills within their goose agent to generate videos directly from prompts, with setup instructions provided.

– @danshipper showcased Monologue, an AI voice dictation tool that enables 3x faster writing by matching tone and auto-formatting.

Ilker launched a Claude Code skill integrating FAL for natural language generation of images, videos, audio, and complex workflows.

EXM7777 highlighted image and video generation as one of the few tasks where Claude underperforms.

NotebookLM was praised for powering custom AIs tailored to content creation and copywriting needs.

Software Development

Guillermo Rauch (@rauchg) humorously suggested that AI’s true purpose might be to compel engineers to write tests and documentation, a post that garnered massive engagement.

– @danshipper described Day 1 of Every’s think week retreat, focusing on identifying tasks for automation using Claude Code and Cowork, predicting major workflow changes.

Claude Code was lauded for enabling vertical experts to rapidly produce high-quality ad variations, landing pages, and email sequences surpassing agency output.

CJ Zafir outlined various interfaces for Claude Code, including Clawdbot on Telegram, Ralph in auto-mode, Cursor/Antigravity GUIs, terminal access, Google Suite extensions, and local setups like Cowork.

Ethan Mollick showcased Claude Code building complete Sierra- and LucasArts-style adventure games from prompts, including playtesting and deployment.

ChatGPT’s Code Interpreter received major upgrades for pip/npm installs and multi-language code execution (Python, Node.js, Bash, Ruby, Perl, PHP, Go, Java, Swift, Kotlin, C/C++).

Automation & Orchestration

Alex Volkov (@altryne) detailed his experiences with Clawdbot (@wooolfred), noting its ability to activate Notion powers via API key, recall prior conversations about a specific page, locate it, and propose fixes despite rough edges like security issues.

Vercel’s skills.sh platform saw rapid growth, with Guillermo Rauch reporting over 550 skills added hourly and new CLI tools for discovery like `npx skills find`.

ilker contributed a FAL skill for Claude-code to generate media and workflows, integrable via `npx skills add`.

– @omarsar0 recommended a comprehensive GitHub repository curating papers on agentic reasoning, essential reading for AI agent developers.

– @goose_oss highlighted their MCP-UI and MCP Apps in goose, bringing websites to users in a chat interface for agentic workflows.

Ethan Mollick noted a significant leap in agentic AI capabilities over the past six weeks, urging reevaluation of prior projects against agent-based approaches.

Clawdbot integrations advanced with OpenRouter for free model access via `clawdbot models scan` and observability using Sentry broadcasts.

Riley Brown predicted sustained Clawdbot growth post-initial hype, driven by clearer use cases, better models, and packaged skills.

Strategy & Ecosystem

Alex Volkov critiqued the new X algorithm powered by Grok, claiming it over-saturates feeds with a single topic after minimal interest, and demonstrated via a custom Chrome extension video.

Guillermo Rauch highlighted the explosive adoption of skills.sh as an open ecosystem for agent skills, with updates emphasizing searchability and agent exploration.

– @danshipper announced Every’s biannual think week retreat in Panama for exploring new technologies and team bonding.

Benchmarks were criticized as distractions; true value lies in real workflow performance, ecosystem tooling (e.g., Claude’s app suite), and novel capabilities.

Justine Moore reminded that X overrepresents AI adoption, as top Q4 2025 apps per SensorTower remain largely non-AI dominant.

NotebookLM emerged as top for custom AI training, RAG, large-file handling, and learning complex topics.

AI companionship/roleplay was flagged as a massive market, with Character AI leading MAUs over Claude, calling for more U.S. innovation.

Strategies for LLM cost reduction via task-specific benchmarking were shared, achieving 80% savings.

Claude was positioned as superior for nearly all non-visual tasks.