Prompt Engineering

AI Topics Discussed on 16 Mar, 2026

—

by

Fraser

Software Development

Guillermo Rauch (@rauchg) highlighted progress in AI-assisted engineering at Vercel, announcing the achievement of “pure agent-driven layout shift fixing & skeleton generation” internally to ensure instant loads and smooth experiences without jitter, replicating craftsmanship automatically.

He also shared a one-hour video demo of “TravisCore” for building and accelerating development workflows.

Dan Shipper praised autoresearch for its simplicity in boosting performance, such as improving Liquid template benchmarks by 53%.

Omar Sar shared excitement over CLI-Anything reaching 15K stars, noting its strong fit for coding agents when paired with tests.

He also highlighted the chrome-cdp skill for enabling agents to interact with live Chrome sessions without browser frameworks.

OpenCode announced free access to NVIDIA’s Nemotron 3 Super model on their platform, emphasizing its speed, openness, and 1M context for coding tasks.

Discussions highlighted the transformative impact of coding tools like Codex and Claude Code on content creation and project management. These tools enable rapid analysis of hundreds of content pieces, swipe file building in Obsidian via CLI, ICP research, and one-shot generation of landing pages, newsletters, and LinkedIn posts, making pre-writing workflows 10x faster.

Comparisons of Excel agents from Claude, OpenAI, and Microsoft Copilot showed ChatGPT creating a fully functional strategy game with formulas and a “smart” enemy, while Claude acted as a game master rather than building a self-contained game.

Automation & Orchestration

Discussions touched on agentic systems for development tasks, with Rauch noting deprecation of Edge Runtime in favor of Node.js for better performance, citing past learnings.

Gokay (@gokayfem) demonstrated a workflow stacking Nano Banana 2, ElevenLabs, Whisper, and Creatify Aurora for video production.

Autoresearch drew attention for embodying the bitter lesson by prioritizing token scaling over complex agent setups.

Simon Willison added a chapter defining agentic engineering to his patterns guide and shared notes from a talk on the topic.

Omar Sar discussed applying distributed systems theory to LLM teams, noting O(n²) bottlenecks and benefits of decentralized structures; he promoted skills for tool usage in agents via continual learning frameworks like XSkill.

Goose OSS highlighted MCP’s value in enterprise for agent coordination.

Agentic tool optimizations were a focus, with CLI preferred over MCP servers to avoid loading full tool definitions on every turn, saving 96-99% tokens via tools like mcp2cli that convert servers to on-demand commands.

OpenClaw workflows for content clipping were detailed: agents trained on course content score virality from transcripts, edit clips with captions, and generate UGC via Sora 2, scaling beyond manual limits for $10k+/mo businesses.

OpenClaw setups now include automated evals for thumbnail generation over multiple rounds.

Strategy & Ecosystem

Cristóbal Valenzuela (@c_valenzuelab) predicted dedicated AI categories at awards like the Oscars within five years, with nearly all movies incorporating AI, responding to anti-AI sentiments in animation.

He reiterated the end of traditional creative software and AI “eating software,” boosting fields like art and philosophy, while announcing Runway Labs incubator for AI video and world models across industries.

Javi Lopez (@javilopen) contrasted AI skeptics labeling outputs “slop” with industry adoption, including studios and viral content reaching millions.

Rauch celebrated 10M weekly AI SDK downloads for seamless model integration and emphasized entrepreneurship’s role in AI stamina over genius.

Valenzuela shared tips like “let models imagine aggressively but make them believe conservatively.”

Predictions for 2026 included Opus 4.6-level models on MacBooks, OpenAI consumer hardware flop, Claude gaining OpenAI market share, fine-tuned small models outperforming GPT-5 in domains, fully autonomous AI startups funded, xAI acquiring coding tools, phone-based quantized models beating Sonnet 4.6, $200 AI inference boxes outselling Mac Mini, new agent IDEs succeeding, and persistent “AI souls” competing with humans.

AI researchers face a steep industry-academia pay gap, with top 1% industry earners making $1.5M more annually than academics—a fivefold increase since 2001—driving transitions to private firms and shifts toward patents over papers.

Fact-checking AI claims requires domain experts, with consensus viewing dog cancer diagnostics and fruit fly brain mapping as having caveats, while math capabilities enable real but non-critical work.

Creative & Visual Media

Generative tools dominated, with Heather Cooper (@hbcoop_) showcasing Midjourney style references (–sref) animated via Kling 3.0 for images and videos.

Gokay promoted Nano Banana for styled slides and videos, urging reference images for variety.

Fofr (@fofrAI) experimented with prompts blending classical paintings and shared feedback on Seedance 2.0’s challenges in narrative video scaling despite strong short-form and consistency.

Lopez countered “stolen art” claims with high-quality examples and celebrated a Seedance/Kling short garnering 10M views, rivaling cinema attendance.

Valenzuela demoed Runway’s real-time video agents for BBC live TV augmentation and spontaneous “vibe producing” clips.

Alex Volkov (@altryne) praised streaming Gaussian Splatting videos.

Fal launched Sora 2 Character Creation for consistent characters across video scenes and Pixelcut Background Removal for precise cutouts in product imagery.

Gemini 3.1 Pro excels in web design workflows: extract styles/layouts from Framer templates, generate Veo 3.1 hero videos, and Nano Banana branding assets.

AI video mashups like LOTR x Pawn Wars combine Nano Banana, Kling 3, and ElevenLabs, with emerging “Cocomelon for adults” genres and Krea Edit for regional edits, perspectives, lighting.

Kling v3 multi-shot workflows automate prompt creation and video generation.