AI & Agents
Technical articles, field notes, and systems troubleshooting filed under AI & Agents.
Gemini 4 Hydrogen, xhigh Thinking, and Native Orchestrator Mode in Google Antigravity
Planning on a frontier model, then manually flipping the picker to a fast model just to save quota was getting old. Antigravity’s Gemini 4 drop finally automates the split with native Orchestrator mode and Hydrogen (xhigh).
RIP Google Jules: Why Local IDE Agents Don't Replace Closing Your Laptop on Cloud VMs
Google just put Jules into read-only mode and told everyone to move to Antigravity. Antigravity is great while your screen is open, but it misses the one killer workflow Jules actually nailed: dispatching a batch of tasks to isolated cloud VMs from the CLI, closing your laptop lid, and coming back two hours later to finished PRs.
Inside RUN World Creator Studio - Pure Vibe Coding, Model Testing, and the AI Coach
A hands-on look inside RUN.world's browser Creator Studio: vibe-coding limits, model comparisons, the internal CLAUDE.md template, and the AI Coach.
MCP v2 and the 2026-07-28 Spec - Running Stateless MCP Servers on Cloudflare Workers
Breaking down why MCP v1's stateful session model was painful on the web, how the 2026-07-28 spec makes Streamable HTTP stateless, and how to deploy on Cloudflare Workers without Durable Objects.
The Vibe Coders Checklist - Things LLMs Silently Break in Production Code
Why LLMs write untyped defensive JavaScript inside .ts files to silence errors, and how Prism, Vigil, and Craftsman clean it up.
The AI Dopamine Rush Is Wearing Off
From CES kitchen gadgets to multi-billion mega-rounds, the initial AI gold rush has ended. What the numbers say about venture fatigue, subsidized compute, and what comes next.
The 30-Minute Game Paradox and the Reality of AI Game Jams
Building Ogre Smash Sports in 30 minutes on RUN.world, and the resulting existential realization about single-use casual games and marketing jams.
Gemini 4 Argon vs. Gemini 3.8 Flash — Reasoning, Execution, and Copywriting
How to balance Gemini 4 Argon and Gemini 3.8 Flash in Antigravity: saving scarce reasoning quotas for architecture and technical copywriting while letting Flash chew through code execution.
Building Abomination Arena on RUN World
Entering RUN.world's Monster Mash game jam: building a fast-paced bullet-heaven horde game with heart upgrades and Doctor Mittens.
Building Games on RUN World - An Honest Creator Review
An honest, builder-first evaluation of RUN.world as an AI web game publishing platform: Game Studio vs rundot CLI, rev share, and trade-offs.
The Prompt Anchoring Trap in AI Coding Agents
Why exhaustive checklist prompts cause autonomous coding agents to hyper-optimize single functions into microsecond extremes, and how Zero-Anchor traversal solves it.
Gemini 3.8 Flash in Agentic Mode — Throughput Beats TTFT
Why sustained 310 tok/s output and execution discipline beat low TTFT for real autonomous coding agent work.
Google AI for devs
Practical breakdown of Jules, Antigravity, Colab, AI Studio, and Genkit in daily software engineering workflows.
Gemini 3.1 Pro vs. Gemini 3.6 Flash in Antigravity — Reasoning vs. Raw Speed
Strategic task allocation between planning and muscle: comparing Gemini 3.1 Pro and Gemini 3.6 Flash in Antigravity.
Vibecoding a Cosmic Arcade
Creating an open-source browser physics simulator with PixiJS, D3.js, GLSL shaders, and disciplined memory management.