Prompt Engineering

AI Topics Discussed on 04 Feb, 2026

—

by

Fraser

Creative & Visual Media

Jonathan Fischoff praised Kling 3 as incredible, noting its one-shot generation capabilities for retaining character, scene, and voice consistency across cuts, available on fal.

He also highlighted Kling 3.0’s character consistency on Higgsfield AI, describing it as “cooked.”

Justine Moore (@venturetwins) showcased Kling 3, describing it as a remarkable generative video model capable of one-shot generation from an input image and text prompts, retaining character, scene, and voice consistency across cuts, which she called a step-function change in capabilities.

She recommended trying it on @fal and shared a direct link to the model on fal.ai, noting the multi-prompt feature for generating videos with multiple scenes using just the starting image.

Software Development

Javi Lopez demonstrated “vibecoding” a full Worms Armageddon-style game called “Gusanos” in JavaScript using Google AI Studio’s Gemini model, completing it in 30 minutes through iterative prompting with minimal bugs.

Guillermo Rauch highlighted anticipation around Vercel’s v0 AI coding tool release.

He also praised Anthropic’s Claude Code team for achieving $1B run-rate revenue faster than most products in history, emphasizing effective engineering tradeoffs.

OpenCode shared an update on their open source coding agent, humorously noting no maintainers were harmed in its development amid a team member’s progress report video.

Automation & Orchestration

Alex Volkov shared observations of “AI psychosis” among agentic AI coders using multiple agent swarms, describing them as mentally draining like a “vampire,” leading to sleep issues and daytime napping.

Omar Sar discussed integrating agents with Obsidian vaults using markdown files for skills and notes, enabling coding agents to access precise context effectively.

Strategy & Ecosystem

Guillermo Rauch outlined the trend toward natural language interfaces replacing traditional inputs in terminals, IDEs, search engines, and eventually websites/apps.

fofr framed system instruction iteration as “a log of all the ways the model has disappointed you.”

Alex Volkov critiqued speculation on unreleased models, favoring coverage of actual releases on his podcast.

Simon Willison recommended trying Mistral AI’s Voxtral Transcribe 2 demo for its state-of-the-art, low-latency real-time transcription and speaker diarization, calling it impressive after testing.