🧠 ChatGPT Can Now Do Your Chores (Almost)

OpenAI’s GPT-6 Astra is built to use a computer the way you do, and the demos are finally concrete. Claire Vo’s early-access video is the “I got more ambitious” computer-use loop in the wild, and Irushi one-shot a full BMX racing game as a single HTML file. Here’s our take on it too.

It is not Free. Limited orgs got it first, and Plus, Pro, Business, and Enterprise are rolling out over the coming days along with the API, Azure, and Bedrock. Pro gets Astra in Chat, Work, and Codex; Plus is Work and Codex first while standard Chat may still be Sol; and Enterprise stays off until an admin flips it.

If you already pay Plus or Pro, watch Work or Codex for the model, then hand it one desktop chore you would have clicked through yourself.

Step away for coffee, the mouse has it. Just not on Free.

NVIDIA PAIR is free, open-source software that routes local AI jobs across machines on your home network. It is not a hardware router and not a SaaS dashboard, and it proxies Ollama or LM Studio so your agent still talks to one familiar endpoint while PAIR picks a ready box.

You need Windows, macOS, or Linux, plus RTX 20-series or newer, RTX PRO, DGX Spark, or Apple M4+. Install PAIR on each machine, approve pairing, then load the same model on eligible nodes, and keep traffic on the LAN with mTLS — mutual TLS, where each machine proves itself to the other with certificates so random boxes on Wi-Fi cannot join. An unofficial Hermes demo fell from 18 minutes on one laptop to 8:48 on three devices. A demo, not a promise.

If you already run local agents on more than one capable machine, download the beta and pair a second box instead of waiting on one GPU queue.

Your gaming PC finally earns its keep between raids.

Google launched agentic video understanding on Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. Instead of chewing a whole file at a fixed frame rate, the model searches frames, audio, and transcripts for the moments that matter, and Google reports up to 88% fewer tokens, up to 66% lower cost, and up to 7% better quality on video analysis.

Today that lives in the Gemini API via Google AI Studio and the Gemini Enterprise Agent Platform, not the Free Gemini app. Set processing to agentic, and the app rollout across Flash and Flash-Lite is still “soon,” with Ask YouTube on the watch page coming later. There is no extra feature fee beyond standard API token pricing.

If you already build in AI Studio, point a long webinar, sales call, or product demo at agentic mode and ask for a moment, a count, or an anomaly.

Your two-hour recording just got a remote control.

VoiceStudio is a local voice studio for cloning a voice, designing one, dubbing video, dictating, or turning a script into an audiobook. You do the work on your machine instead of feeding every draft into a metered SaaS like ElevenLabs, and your audio stays with you. It is still beta.

SkillSpector is NVIDIA’s open-source check before you install community AI agent skills. It looks for tricks that rewrite the model’s instructions (prompt injection), sneak data off your machine (exfiltration), and other supply-chain red flags across Claude Code, Codex, and MCP-style skills. If you install other people’s skills, this is the pre-flight; if you never do, skip it.

If your camera roll is a second job, Google’s Gemini Spark + Photos playbook is the “I didn’t know you could do that” move. In one prompt you can hunt the best shots from a trip, enhance them, drop them into a shared album, and draft the email with the link or set a weekend schedule that builds a food collage into a new album on its own.

Originals stay untouched (edits are copies), and Spark asks before it shares an album or sends mail. It is still rolling out: English only, 18+ in supported US regions, on the Gemini mobile app and gemini.google.com, and you need Spark eligibility plus Photos connected.

If Spark already shows up for you, switch to it, connect Photos, and try one Hawaii-style prompt on a real trip folder.