https://t.co/gWa67YYNL3

So to the wanderer, and to the tried, To those laid low, to those in new employ, To those who ship’d their miracles with pride, To those who ache, and yet refuse to cloy, I send this wish, like starlight on the snow: Peace in your rooms, and brighter days to come.

Autonomous AI has arrived. Congrats, your backlog just gained: things reality does to your roadmap. The next AI moat isn’t a model, it’s execution infrastructure: fresh context, contradiction handling, agent-speed coordination, verifiable trust, and budgets that don’t melt

The math behind AI inference margins is about to change. Reuters: Google’s internal TorchTPU is aimed at making PyTorch run cleanly on TPUs. The moat isn’t FLOPS, it’s the risk of touching a working CUDA pipeline. If switching becomes “just another Tuesday,” we move from

Unity dev is getting a new superpower, and it isn’t AI writing scripts. MCP is turning the Unity Editor into something agents can operate + verify, not just autocomplete. I wrote up what that changes about iteration, governance, and authorship. https://t.co/96FqYKr8By #Unity

Used #Gemini3Flash to get a ridiculous amount of information about music, in @GoogleAIStudio

Few updates: - switched to #Gemini3Flash - added AI chat to entity graph - added intelligent graph refinement

I swapped the brain in my agent framework to Gemini 3 Flash today. It stopped performing and started operating. Why? Loop Density. When intelligence becomes this fast and cheap, verification becomes a reflex instead of a luxury. You stop designing for scarcity and start

Gemini 3 Flash shipped today, so my extremely "scientific" experiment asking 4 frontier models for a ship date just got a result. Claude 4.5: Dec 15 (off by 2 days) Grok 4.1: Dec 16 (off by 1 day) Hard misses: GPT-5.1: Dec 30 Gemini 3 Pro: Feb 2026 Closest priors win. Now the

Cleaned this up, added a few features, and released it under the MIT license:

185 seconds in Google AI Studio: I asked Gemini 3 Pro for a web app that uploads Markdown, extracts entities, and renders an interactive 3D knowledge graph. It one-shot a typed React 19 + TS codebase (UI, state, graph, schema JSON) with only a tiny TS cast fix. The bottleneck is

Shipped MIT open-source Olympus Odyssey: a voice-controlled MUD where Gemini is the intelligent GM. Speak or type commands, Hermes narrates in real time, and new Greek-myth locations generate on-demand as you explore. https://t.co/JMej3JbA5s

Single-pass refinement of a vibe-coded experiment in Google AI Studio to a production-ready stack with Google Antigravity is simply phenomenal. #GenAI #GoogleAIStudio #Gemini3 #GoogleAntigravity

Spent the last few days letting Google Antigravity + Gemini 3 Pro refactor my AI Agent Framework. Verdict: we’re moving from AI in the sidebar to AI as staff engineer. Deep dive on what that feels like + what it means for engineers: https://t.co/0tUweBV6Tp #AI #DevTools

Hit the wall with prompt engineering, so I built an Advanced Agent Framework: an AI coworker that remembers, researches, and runs its own sims. It took NPC LLM dialogue from ~$1,125/hr to ~$1.50/hr. Story here:

I asked Gemini 3 Pro to help me build an OS (inspired by Tron Legacy). I described the world (charcoal canvas, wireframe windows, AI co-pilot). It generated a browser OS: windowing, terminal, firewall, Neural Link. I describe the universe. The model builds the scaffolding. #AI

Ran a very dumb, very fun experiment: asked 4 frontier models, “When will Gemini 3 Flash drop?” Claude 4.5: Dec 15 2025 Grok 4.1: Dec 16 GPT-5.1: Dec 30 Gemini 3 Pro: Feb 2026 When it ships, we’ll see whose priors (and whose worldview ) were closest. #AI #Gemini3

Today my agent stopped being a demo and started being a system: LangGraph runtime + ADK OODAL with checkpoints. Plans pause for consent, ACT sees all governed tools, outputs read like outputs, memory compounds. One ask in, work out. #AgenticAI #GoogleADK #LangGraph #OODALoop

Wharton says 74% of companies see positive ROI from gen AI. MIT says 95% of pilots fail to reach production. Both are right. One measures assistant level wins, the other systems that move the P&L. #GenerativeAI #AIStrategy I unpack the split screen here:

Amazon just proved consumer AI sells. Rufus is an in-flow shopping agent on track for $10B+ in incremental sales. In my new piece I argue this is the clearest market proof yet for AI-native games: commerce and play are cousins in the agentic loop. #AI #GameDev

Today I used my AI Video Studio to turn 2 velociraptor shots into 1 seamless sequence: jungle-at-night → drinking at a waterfall. No timeline, no keyframes—just Gemini 3 + Nano Banana Pro + Veo 3.1 acting as camera, editor, and compositor. #AI #GenerativeVideo #Veo31 #Gemini3