xAI's Grok Bot will now offer to do jobs before you ask. Early on Friday morning, Adelaide time, the official Grok Bot account on X announced: "Grok Bot can now suggest ways to help without you needing to ask." Your main assistant, which xAI calls your primary Bot, watches for work it can take off your plate and offers to handle it.

The same day, Grok 4.7, the model xAI released to developers on 21 September, turned up inside the everyday Grok app. The model picker on grok.com now lists Grok 4.7 behind its two most capable modes, Expert and Heavy.

What Primary Bot does

Grok Bot is xAI's agent product. The company has joined SpaceX and now goes by SpaceXAI. You get a set of named AI helpers, called Bots, and each one works on its own computer in the cloud. They sign into the apps you connect, keep working when your laptop's closed, and check back when they need a decision.

Tap to enlarge

Until now, you gave a Bot a job and it did that job. Primary Bot flips that around. The announcement went out at 17:35 UTC on Thursday 1 October, which was 3:05am Friday 2 October in Adelaide. In a follow-up post in the same thread, the account added: "Your primary Bot will spot work it can take off your plate and offer to handle it. It's rolling out over the next few hours, and suggestions don't count against your usage."

That last bit matters if you watch your allowance. xAI's Grok Bot page says the service is included with eligible SuperGrok and Cursor plans, with weekly Grok Bot usage built in and anything extra billed by token cost. xAI says a suggestion is free. It hasn't said whether a task you accept is free too. Until it does, it's safest to assume the work counts like any other job you hand a Bot.

The AI tracking site TestingCatalog posted screens from the iOS app on Thursday, hours before the official post. The welcome screen reads: "Introducing Primary Bot. Primary Bot does work proactively. It is your go-to for everyday tasks, unblocks your work, and checks in when it needs your help." You can then pick "Use an existing Bot" or "Create Primary Bot". So if you've already trained a Bot on your inbox or calendar, you can promote it rather than start again.

In the same screenshots, a new Primary Bot introduces itself like this: "I'm here to be your primary bot. I can manage your other bots. Task me with anything. I'll make sure it gets to the right bot." In practice, it's a front desk. You tell one Bot what you need, and it passes the job to whichever specialist Bot suits it best.

Grok 4.7 reaches the everyday app

xAI launched Grok 4.7 on 21 September for developers, in Cursor, in its Grok Build coding agent and through its API. It calls the model "our most capable model for coding and knowledge work," and says it "works longer on difficult tasks, checks its own work more carefully, and comes with our best-calibrated safeguards to date."

Ten days on, everyday users have it too. When we checked grok.com on Saturday morning, Expert mode was described as "Thinks hard · Grok 4.7" and Heavy as "Team of Experts · Grok 4.7". Fast mode just says "Quick responses" and doesn't name a model, while Auto "Chooses Fast or Expert". TestingCatalog has reported that Grok 4.7 now sits behind every mode. The app itself only names it on those two, and xAI hasn't posted about the switch.

The link to Primary Bot is right there in xAI's launch notes. "We also trained Grok 4.7 to natively understand the Grok Bot harness, making it better at conversational tasks and general knowledge work," the company wrote. Put simply, the model was built to drive the kind of always-on helper xAI has just made proactive.

The price hasn't moved. xAI lists Grok 4.7 for developers from US$2 per million input tokens and US$6 per million output tokens, the same as Grok 4.6. There's also a faster version that runs at twice the speed for twice the price. In xAI's own benchmark table, Grok 4.7 scored 46.3% on CursorBench 4.0, up from 40.4% for Grok 4.6, and 37.6% on Terminal-Bench 4.0, up from 20.3%. Those are the company's figures. The same table shows Anthropic's Fable 5.1 still ahead on both, at five times the input price. We covered Fable when Anthropic put it out on 1 September.

On safety, xAI says Grok 4.7 was "built with an entirely new safeguard stack" and is "the strongest model we've tested on refusals and jailbreak resistance." It reports a score of 62.4% on LatchBio's biosafety benchmark. It also says only 3.3% of risky dual-use prompts got through on its own HackerBench v0.3 test.

A busy week for business customers

Primary Bot caps a run of launches aimed at workplaces. On 28 September xAI launched Team Bots, shared Bots that a whole team can work with. "Although the Bot is shared, each person's conversations with it remain private," the company wrote. A Team Bot can sit in a Slack channel with its own handle, and xAI says it keeps separate memories for each user.

xAI gave one customer example. Dakotah Rice, chief executive of the small-business insurer Harper, is quoted in the post: "We built a Team Bot in 24 hours to automate this, saving our customers over $120,000 from hundreds of policies." That's Harper's own figure, and we haven't checked it.

On 1 October, xAI also released Grok for Intune, a separate iPhone app for companies that manage staff phones with Microsoft Intune. 9to5Mac reported that the app "supports Intune Mobile Application Management, honors your organization's app protection policies, and signs in with your work email," in the company's words. The company also lists no in-app purchases. 9to5Mac noted that OpenAI released a dedicated enterprise version of its own app in May.

Developers got more ways in as well. Google's Gemini Enterprise Agent Platform now lists Grok 4.7 as a managed model, with the model name grok-4.7, on a page last updated on 1 October. Amazon Web Services added Grok 4.7 to Bedrock on 28 September, with its 500,000-token context window and four levels of reasoning effort.

What xAI hasn't said yet

Primary Bot arrived through social posts and an app update, not a blog post. As of Saturday morning, xAI's news page still led with Grok 4.7 and the Team Bots post, and neither mentions Primary Bot.

That leaves a few practical questions open. xAI hasn't said how often Primary Bot will chime in with ideas, how to narrow what it watches, or how to switch the suggestions off. It also hasn't said which model each Bot runs on.

What already exists are Grok Bot's approval controls. xAI's Grok Bot page says Bots come back when your approval is needed, and that sensitive actions can go through a check it calls Auto Review before they run. With a Bot that now goes looking for work, those settings are worth a look before the first suggestion lands.

Rivals are heading the same way. Earlier this week OpenAI launched its own always-on agents, called dots, running on its GPT-6 Astra model. OpenAI says you start with your "primary dot", and that dots can bring you work "sometimes before you even think to ask." xAI's answer is to put its agent into plans people already pay for, and now to have it put its hand up first.

What comes next

xAI said Primary Bot would reach users "over the next few hours" after Thursday's post, so most Grok Bot accounts should have it by now. Grok 4.7's move into the consumer app is still rolling out.

The next model is already on the record. On 14 September, Musk wrote on X: "Grok 4.8 will be a noticeable improvement." He was replying about Grok 4.7. xAI hasn't given Grok 4.8 a date, a price or an API model name.

For now, the change most people will notice is the simplest one. Open Grok Bot, and it may already have a job lined up for you.