Prompt Engineering

AI Topics Discussed on 10 Mar, 2026

—

by

Fraser

AI Topics Discussed on 10 Mar, 2026

Creative & Visual Media

Javi Lopez demonstrated Nano Banana 2 generating a Tim Burton-style logo on a white background.

fofr shared examples of creating ditto-like anime versions of oneself using the model, including a detailed prompt for digitally sketched doodles with specific features like gloopy hair and wide mouths.

Lopez also updated on improvements to Bytedance’s Seedance 2.0 video model, noting much better performance after addressing moderation issues.

Cristóbal Valenzuela introduced “Devil’s Advocate,” a live AI podcast featuring two RunwayML characters—God and the Devil—discussing news in real time, with ideas for audience calls and 24/7 operation.

NotebookLM’s new video generation feature impressed users by producing detailed, engaging videos from research reports, such as a strategy for time-traveling to conquer ancient Rome.

Software Development

Guillermo Rauch celebrated Vercel’s AI SDK reaching 10 million weekly downloads, emphasizing its simplicity as a one-package solution for any model via “npm i ai.”

Gokay Has noted fun early experiments with physics-informed neural networks using Karpathy’s autoresearch tool.

Simon Willison discussed how AI should help produce better code, sharing a guide on agentic engineering patterns focused on improving code quality.

Guidance emerged on mastering AI-assisted coding through Claude skills: have the model create courses from Anthropic’s guide, document milestones and bug fixes, then build reusable skills for one-shot future projects.

Automation & Orchestration

fofr integrated Google’s new Gemini Embedding 2—a natively multimodal model handling text (up to 8192 tokens), images (up to 6), videos (120s), audio, and PDFs (6 pages)—into local prompting tools via OpenClaw and Gemini 3.1 Pro for improved search.

Elvis Sar (@omarsar0) highlighted the evolution from “context engineering” to “harness engineering,” urging developers to build their own agent harnesses. This approach fosters understanding of agentic systems, including CLI tools, APIs, skills, memory management, automations, and schedulers.

Reusable agentic skills for coding architectures were promoted, stacking them to accelerate projects via structured documentation and skill creation.

Strategy & Ecosystem

Discussions highlighted emerging multimodal capabilities like Gemini Embedding 2 and milestones such as Vercel AI SDK’s massive adoption, signaling growing ecosystem maturity.

Alex Volkov commented on the memetic value of AI agents like OpenClaw and platforms like MoltBook in capturing attention, drawing parallels to strategies by leaders like Zuck and Sama.

AI labs prioritize building smarter models for superintelligence over consumer products, with product shipping being incidental.

Research counters claims of inevitable AI homogenization, showing prompting and context yield diverse ideas comparable to humans.

A NYT blind test revealed 54% of readers prefer AI writing over human.

Debate on mandating AI text labels sparked counterarguments for labeling “dumb” human writing instead.