🧠 Everyone's Hiring AI Now

xAI (now folded into SpaceXAI after the Cursor deal) launched Grok Bot, a beta product that hands AI agents their own cloud machine to work from. Instead of you copy-pasting between tabs, a Bot logs into your actual tools, including ones with no API, and works through a task end to end.

The pitch is blunt: Bots are teammates, not assistants. They keep running after you close your laptop, remember context between tasks, and can even coordinate with other Bots to split up work. You only get pinged when something needs your sign-off.

Access is limited for now to top-tier SuperGrok and Cursor subscribers, priced well below what a human hire would cost. This is the clearest signal yet that the AI agent race has moved from "helps you work" to "replaces a line item."

OpenAI is rolling its all-in-one ChatGPT desktop app out to Linux, folding three separate jobs into one window: Chat for quick questions, Work for research and deliverables, and Codex for actual software builds.

The Work mode is the interesting bit. With permission, it can dig through your local files and desktop apps, and its built-in browser can act on real websites and web tools instead of just reading about them.

While everyone chases smarter models, NVIDIA released Nemotron 3.5 Lightning, an open model built purely for volume. It's a 30-billion-parameter model that only activates 3 billion at a time, which is the technical way of saying it's built to run fast and cheap.

The logic is simple once you see it: a long-running AI agent doesn't need a genius for every single step. Most of its time goes to grunt work like calling tools, checking results, and handing tasks to sub-agents. NVIDIA built Lightning specifically for that grunt work, claiming up to 4x faster output than similarly sized models.

It's also fully open, meaning anyone can download and fine-tune it. NVIDIA is pairing it with a routing tool called NeMo Switchyard, so a bigger "smart" model can handle the planning while Lightning handles the busywork underneath.

Before you hand anything to AI, map out what your team actually does day to day, then sort those tasks by how much judgment they need. Start with the pure "playbook" tasks that need zero judgment calls. Once that's solid, move to low-stakes judgment tasks like drafting captions or scripts. Only after that should higher-stakes work go to AI, and even then, keep a human reviewing before anything ships.

If you've been agonizing over the perfect prompt, try this instead. Built-in recording features such as ChatGPT’s "Record and Replay" or Claude’s "Record a Skill" let you just do the task once while it watches. Walk through your actual process step by step on your computer, and the AI picks up your real intent instead of your best guess at describing it in words. It turns a repeatable task into an automated skill far faster than writing it all out manually.

Here’s how to do this.

Are you looking to automate your business with AI? We’re hosting live build sessions with Practicaly AI for Entrepreneurs members every Friday.

Not a member? Join now to get 20% off one month

Join us live, build in real time, and get your questions answered on the spot.

👉Save your spot today!

Can’t make it live? Don’t worry, we’ll send out the recording after the session to everyone who RSVP’ed.

Event details:

📅When: Friday at 10AM PT/1PM ET
📍Where: Virtual, Google Meet

After you RSVP, you’ll be sent a link with the event details. Hope to see you there!