Prompt Engineering

AI Topics Discussed on 14 Feb, 2026

—

by

Fraser

Creative & Visual Media

Cristóbal Valenzuela (@c_valenzuelab), CEO of RunwayML, predicted that within two years, 90% of pixels on screens—from images and videos to games and software—will be AI-generated.

Javi Lopez (@javilopen) shared a highly engaging AI-generated video featuring Nicolas Cage as Superman battling Chuck Norris, described as “the film you didn’t know you need.”

Heather Cooper (@hbcoop_) showcased a workflow from Midjourney-generated image to Kling 3.0 video, preserving close shots, natural lighting, and reflections in an astronaut scene.

fofr (@fofrAI) posted examples emphasizing that “you can make amazing things with AI,” including visuals likely generated by advanced models.

Jonathan Fischoff (@jfischoff) noted emerging tools like json-render enabling AI to respond with rendered UI and 3D scenes.

A.I.Warper (@AIWarper) shared updates to an autonomous video editing tool adding LUTs, grain filters, VHS/glitch effects, frequency-reactive zooms, stutter cuts, and dynamic pacing from random clips, producing music videos synced to AI-generated songs.

They highlighted FireRed-Image-Edit-1.0 as a new state-of-the-art for image editing with style mastery and virtual try-on, Seed 2.0 for multimodal capabilities, and a model converting NPR to realism styles.

Omar Sar (@omarsar0) demonstrated an MCP tool generating Excalidraw diagrams from prompts.

Software Development

Guillermo Rauch (@rauchg), CEO of Vercel, discussed the transformation in engineering: pre-AI prioritized focus over parallelization, but post-AI, parallelization surpasses focus as the key skill.

Alex Volkov (@altryne) detailed building a custom pixelated 2D platformer game for Valentine’s Day using AI coding agents like @wooolfred (on @openclaw), writing zero lines of code while directing assets via Nano Banana (NBP), story, and playtesting—highlighting AI’s productivity boost without full autonomy.

Dan Shipper (@danshipper) humorously depicted reviewing Claude’s 10,000th pull request of the day, highlighting AI-assisted coding scale.

He also shared a Claude interface compacting conversations for ongoing chats.

Jonathan Fischoff discussed second-order effects of smarter LLMs eliminating communication friction in interactions.

Automation & Orchestration

Volkov’s project exemplified agentic workflows, with AI agents handling code generation, sprite animation (despite limitations), asset creation pipelines via tools like Nano Banana, and iterative development over weeks for a functional game.

Jonathan Fischoff emphasized AI agents self-integrating into companies faster than past tech like fax or internet due to laziness.

@levelsio (@levelsio) welcomed X’s crackdown on automation and spam, hoping to end unmarked AI replies while supporting official API bots.

Omar Sar highlighted AdaptEvolve for efficient evolutionary AI agents using confidence-based model routing, reducing compute by 37.9%.

He also noted enterprise AI agents needing evaluations for reliability.

Machina (@EXM7777) shared insights on OpenClaw setups, noting that the best ones are created by non-coders who master turning messy ideas into clear, structured instructions for AI workflows.

Strategy & Ecosystem

Rauch outlined emerging trends in AI-driven engineering, shifting core competencies toward leveraging parallel tasks enabled by AI tools.

Valenzuela reinforced Paul Graham’s essay on taste as the key differentiator in an AI era where creation is democratized, stressing expertise and intolerance for subpar work.

Jonathan Fischoff predicted declining college value as businesses recognize weakened signals from admissions amid skill gaps.

He forecasted rapid societal laziness accelerating AI efficiency gains over envy/greed barriers.

Machina (@EXM7777) argued that clear prompting remains a critical skill for leveraging AI effectively, as it compensates for technical limitations and will matter until models can directly interpret intent.

Ethan Mollick (@emollick) raised concerns about AI advancement making verification a bottleneck accessible only to a tiny expert group, exemplified by challenges in checking AI-generated math proofs, and called for solutions like multi-AI collaboration.