Grok Bot goes after the last 10% of work most AI leaves undone

Grok Bot started as an internal prototype at xAI, and by the company's account it spread on its own. Teams built bots for sales outbound, marketing campaigns, office operations, and bug fixes. Today the same product opens in beta to paying subscribers, with a claim most agent launches avoid: these bots do not stop at 90% of a job.

"There is a huge difference between 90% done and 100% done," says Roman of the product team, in a quote posted with the launch announcement. "Most AI gets you almost there. Grok Bot can finish the swing, because the work lands where a human would put it, in the actual tool."

That quote is the product. Most assistants hand back a draft or a plan and leave the human to close the loop. Grok Bot is built so the work lands where the job actually happens: the CRM, the inbox, the ticket tracker. The bot signs in, finishes, and only comes back when something needs approval.

A computer of its own, with no API required

xAI leads with the architecture. Each bot gets a computer of its own in the cloud, so jobs do not stall when you step away. It can sign into apps, tools, and websites, including platforms with no clean API or MCP. The usual agent playbook leans on APIs and MCP to reach other services; xAI is explicitly pitching the ability to work where neither exists. The tool-first approach has its own documented failure modes: a recent paper found tool-using models can make easy questions harder.

Inside the company, the examples are already multi-step. A sales bot updates the CRM with call transcript notes, then drafts follow-ups. An ops bot seats new hires and processes invoices received in Gmail. An engineering bot reproduces a bug in the product UI, files the ticket, and hands the fix to a debugging bot.

Bots passing work to other bots is a pattern the announcement pushes hard, and xAI has already shipped something in that spirit: Cursor's Grok 4.5 was built by AI agents. You can run a chief of staff bot on top, with specialist bots underneath for inbox, expenses, recruiting, bug fixes, or operations. Bots message each other in threads, share context, coordinate in group chats, and only pull a human in for judgment calls. The point, in xAI's framing, is that you stop being the middleman.

Learning by watching, then letting go

The second mechanic is teaching by demonstration. Ask a bot to follow along the next time you do a job. It watches the steps, saves the workflow as a routine, takes your corrections, and runs the whole thing alone next time.

No setup ritual precedes any of this. "There wasn't anything to learn. It was just like bringing on a coworker," says Fiona of the community team. You message the bot and it gets to work.

Bennett, in sales, says he showed a bot a workflow once and now trusts it to run forever. "I feel like I'm 2-3x more efficient because it does it without me verifying and reviewing." Emma, in operations, came at the trust question from the other side: she checked on her bots every 15 minutes until they asked why she kept asking so many questions. Now she lets them do their thing.

The quotes circle the same moment, the one where you stop hovering. That is the real product, and also the part that is unproven. The testimonials come from early access users and xAI's own employees, published by xAI; the announcement offers no independent verification.

Who gets it today

Grok Bot is in beta for SuperGrok Heavy, Cursor Ultra, and Cursor Teams Premium subscribers on desktop and iOS. Enterprise customers can join a waitlist. Notably, the beta rides on existing paid tiers rather than a separate checkout, which puts Grok Bot in front of users who already pay for AI tooling.

xAI has been building toward this. Grok launched in late 2023 as a native feature on X and has since expanded into a standalone multi-modal assistant, according to the company's product pages. Its flagship model, Grok 4.5, is pitched as built for coding, agentic tasks, and knowledge work, a release whose biggest story was how it was built. The company also describes a mode where you describe a job once and Grok runs it on schedule or when an email arrives, the trigger-driven jobs we covered earlier. Grok Bot is the harness that turns that positioning into something you can hand a real assignment to.

The open question is the one the product itself raises. If the last 10% really gets done, inside the actual tool, "AI teammate" stops being a metaphor. If it stumbles, the gap between 90% and 100% is exactly where you will notice. Either way, the battleground just moved: starting a task is easy to demo, finishing the job is the product.