This is Part 2 of our two-part Grokbot Playbook, covering its latest features and the use cases they unlock. Missed Part 1? Start here. Click here.

What's new: Grok Bot introduced Vault for storing credentials, payment methods and contact information, while expanding its 1Password integration.

Why it matters: Bots can work inside authenticated websites without requiring you to paste passwords or card details into a chat. This may sound like a convenience feature, but a lot of repetitive work still happens in websites that don't have useful integrations.

With Vault, Grok Bot can securely fill saved contact and payment information into supported checkout flows. For products you buy regularly, the bot can find the item, check availability, prepare the cart and present the total.

You could also give it requirements around price, brand and delivery.

Try this prompt:

There's an important security distinction here. Vault cards can be used without a fresh approval requirement on every transaction. If you want to authorize every payment personally, use a one-time payment flow rather than relying on a saved Vault card.

For me, the more compelling use case is authenticated work that doesn't involve spending money.

Think about logging into a vendor portal to download invoices, retrieving reports from a business account or collecting information from a service without a convenient API. With the 1Password integration, you can approve a login and have the credentials filled into the bot's browser without exposing the underlying password to the bot.

Try this prompt:

The integration is rolling out gradually and requires a compatible 1Password version. Some sites will still need you to take over for additional verification.

What's newly practical: Grok Bot can operate websites and desktop software, and Teach a Task lets you turn a demonstrated browser workflow into a reusable skill.

Why it matters: Many annoying, repetitive tasks still happen inside software that was designed for people to click through manually. The ability to demonstrate a process and then schedule it is especially useful when there's no API or ready-made integration.

One user described using Grok Bot to help sell a domain. The bot compared marketplaces, researched companies that might want the domain and drafted personalized outreach.

The same process could be applied to other assets with identifiable potential buyers.

Try this prompt:

This works better with a short list of credible buyers than with mass outreach.

User @mikepat711 shared an example on X of asking Grok Bot to negotiate with an internet provider through its customer-support chat. He supplied a redacted bill, explained what he wanted and defined what he wouldn't accept.

The bot reportedly spent about 30 minutes working through the conversation and found an offer that could save roughly $58 per month.

That's one person's result, but the approach is worth considering for negotiable recurring bills.

Try this prompt:

Even if the bot can't secure a better rate, it may be able to explain your options and prepare a script for a follow-up call.

A prompt to adapt:

I'd treat this as a document-analysis and claims-administration assistant, not an authoritative interpreter of insurance law or medical necessity. You'll also want to be particularly careful about the personal information it can access.

One entrepreneur (@MinimalDuck)  asked Grok Bot to help reduce spam calls. The bot reportedly completed the registration process for the U.S. National Do Not Call Registry, including the confirmation step.

It's a small task, but a nice example of delegating everyday administration.

Try this prompt:

Use the official registry at donotcall.gov. Registration won't eliminate illegal robocalls, and the relevant process varies by country.

Here's a more experimental idea. One user described using Grok Bot to install and operate Hermes Desktop on the bot's computer. He directed the process from his phone while Grok Bot handled the installation, configuration and execution, then captured the results.

You could apply a similar approach to a specialized coding environment or another desktop application.

Try this prompt:

I'd treat this as an advanced experiment. The user reports suggest it's possible, but it can involve a lot of trial and error.

This is one of the features I'm most interested in. Instead of explaining every step of a workflow, you can demonstrate it.

Grok Bot's Teach a Task feature records your browser interactions for up to ten minutes and turns them into a draft skill. Kevin Stratvert’s YouTube channel recently covered doing this with bookkeeping tasks in QuickBooks. 

Consider a task you perform every week: open a vendor portal, download a report, rename it and save it in Drive.

Instead, you could perform those steps once while Grok Bot observes, review the generated skill and then schedule it.

How to get started:

Try this prompt after recording:

The recording captures visible browser interactions, not microphone audio.

I would start with a simple, predictable workflow rather than something requiring lots of judgment. Teach a task is also still rolling out, so you may need to create the skill from written instructions if the recording option isn't available.

What's new: Grok Bot introduced Team Bots in late September, now called Shared Bots. They allow teams to share specialized assistants with common instructions, tools, and knowledge.

Why it matters: You can take a process that works well for one person and turn it into something the rest of the team can use. This is also where multiple bots start to become useful. Different assistants can handle research, drafting, operations and quality control, with clear handoffs between them.

The trick is to avoid building an elaborate collection of bots that spends more time coordinating than working.

Meeting summaries are useful, but they're only the first step. In a recent sales demonstration, Grok Bot was used to take information from customer conversations and prepare updates to account records, sales materials, and follow-up documents.

That's a much more complete workflow.

After a customer call, the bot could extract new stakeholders, objections, commitments and next steps. It could compare those findings with your CRM, propose updates, and prepare a follow-up email.

You review the changes before anything is written.

Try this prompt:

I would be particularly careful with inferred commitments and deadlines. If the customer didn't agree to something explicitly, the bot shouldn't record it as a confirmed decision.

Kevin Luke of Above the Line described using two specialized bots for prospecting. One researches events and decision-makers. The other turns that research into personalized outreach.

You can build the same system around potential customers, strategic partners or recruiting candidates.

The benefit is that the research can be reviewed and reused before anyone starts drafting messages.

Try this prompt:

I'd start with two bots rather than five. Add complexity only when you can point to a specific problem it solves.

In a Grok Bot sales workshop, Above the Line described using an assistant to produce a detailed estimate from an RFP. The company reported completing in roughly 30 seconds work that previously took a producer much longer.

The exact time savings will depend on the business, but this is an obvious candidate for automation.

The bot can read the RFP, compare it with previous projects, apply your rate card and produce a draft estimate.

Try this prompt:

This is a good example of where even an imperfect first draft can save a meaningful amount of time.

Another workflow described by the Grok Bot sales team is a Deal Bot that reviews non-standard terms and routes exceptions for approval. Think about a sales organization where representatives routinely need approval for discounts, payment terms or changes to standard contracts.

Instead of manually finding the right person and explaining the issue from scratch, a bot could review the request against policy and prepare a concise summary.

Try this prompt:

The bot's job is to make the approval process easier, while the decision stays with someone authorized to make it.

One creator shared a setup involving four bots working together on a publishing workflow.

The first watches a selected group of X accounts and collects news. The second evaluates and drafts stories. The third checks that editorial approval has been given. The fourth hands approved work to a coding agent that prepares the website update.

Nothing goes live without a final human decision. For a publisher producing content frequently, this is a useful pattern.

Try this prompt:

I would keep the editorial review substantial. Automatically moving an incorrect story through four bots doesn't make the story more accurate.

The Grok Bot team described using AI to identify support tickets related to the same incident. In one reported example, an engineer approved responses for 204 tickets in about ten minutes. A bot can group related tickets, investigate the issue and prepare consistent responses.

The interesting capability is recognizing that many apparently separate customer problems may have the same underlying cause.

Try this prompt:

Be conservative about grouping. A mistaken response sent to 200 people is considerably worse than a mistake in one draft.

The reported support results are from the company, so I'd treat the figures as an example rather than a universal productivity benchmark.

Grok Bot's team has also described using Shared Bots inside its own engineering organization.

A bot connected to Slack, Notion, Linear, Datadog and Cursor can follow project discussions, triage bugs, create tickets and launch coding agents.

The company reported that a five-person team used this setup to ship more than 100 pull requests a day during the project.

That's a vendor-reported figure, but the underlying workflow is interesting regardless of the number.

Try this prompt:

This becomes much more useful when the team already has good documentation, clear acceptance criteria and established review practices.

A prompt to adapt:

This is worth exploring if you're building multiple applications or maintaining several internal tools. The important part is making the roles reusable while keeping each project's code, credentials and context appropriately separated.

Alex Finn and Sonny Sangha have demonstrated having multiple bots participate in a group-chat standup. Each bot shares priorities, blockers, and relevant updates.

I can see why this appeals to people running several specialized agents, but it can also become an exercise in bots talking to each other without accomplishing much.

I'd keep the format short and focus on decisions.

Try this prompt:

This works best when each bot is already responsible for real work and has a reliable way to report its status.

Lauren Tan (@poteto), a designer on the Grok Bot team, created a bot-building assistant called Dr. Eggbot.

One user described giving it a brief explanation of his job and receiving a proposed team of six bots, including a chief of staff, inbox assistant, leads bot and call-preparation assistant.

What stood out to me was the treatment of permissions. Each proposed bot had an access list and an explicit rule describing something it could never do. An inbox bot might be allowed to organize and draft but prohibited from sending. A prospecting bot might research companies but never contact them.

I'd borrow that approach.

Try this prompt:

I'd resist the temptation to build every proposed bot immediately. Start with the one that has the clearest payoff.

Shared Bots also make it easier to distribute expertise across a team.

Suppose you have a strong editorial review process. It lives partly in guidelines, partly in examples and partly in the judgment of an experienced editor.

You could create a Shared Bot that applies those standards to every submitted draft. The same idea works for marketing reviews, proposal preparation, recruiting and onboarding.

Try this prompt:

The benefit compounds as you improve the underlying instructions and examples.

Shared Bots use shared tools and configuration, although individual conversations remain private. Review what information and credentials the bot can access before making it available to the team.

What's new: Grok Bot has improved its integrations with Cursor cloud agents, GitHub and related development tools. It can also create short videos from descriptions or images.

Why it matters: You can delegate more of the software development process and inspect what was built before accepting it.

For builders and creators, these are some of the more interesting updates in the collection.

One developer described asking Grok Bot to navigate a WooCommerce staging store, complete a test purchase and record the session.

That's a useful capability for anyone who needs to demonstrate software.

You could have a bot test a signup flow, record the steps, and collect the material for a product walkthrough.

Separately, Grok Bot's video-generation capabilities can create short visual clips from prompts or images.

Try this prompt:

I'd avoid using generative video to recreate a software interface where accurate text and interactions matter. Real screen recordings are more dependable for that purpose.

Grok Bot can delegate development work to Cursor cloud agents and coordinate the results through GitHub. 

Suppose you want to add a feature to an internal dashboard. You can describe what the feature should do, identify the relevant repository, and ask Grok Bot to launch a coding agent. When it finishes, you can review a pull request, test results, screenshots, and recordings where available.

Try this prompt:

I'd make evidence of the result a standard requirement.

A coding agent saying a feature is finished tells you very little about whether it actually works.

In an official Grok Bot 201 workshop, a demonstration showed multiple Cursor cloud agents working on the same development brief using different models.

The resulting implementations were compared, and the strongest elements could be combined.

That opens up an interesting approach for exploratory projects. Instead of committing to one implementation, you can compare different approaches before deciding which to pursue.

Try this prompt:

This will use more resources than a single agent, so I'd reserve it for projects where exploring alternatives is genuinely valuable.

What's new: Grok Bot has added iPhone shortcuts, improved voice interactions and expanded integrations with phone applications.

Why it matters: Your assistant becomes easier to use when you're away from the computer, and it can work with more of the information stored on your phone.

If you have an iPhone with an Action Button, you can configure it to start a Grok Bot voice conversation. Matt Palmer demonstrated the setup in October.

Go to Settings, open Action Button, select Shortcut and choose Grok Bot's Start Voice Chat action. One press, and you're talking to your assistant.

Try this voice prompt:

I like this for quick briefings and capturing ideas when I'm away from my desk. Recent demonstrations also suggest improved continuity between voice conversations, though I'd test the exact memory behavior before relying on it.

Experimental: Limited availability

In late October, Elon Musk announced that Grok Bot could make phone calls to people and businesses.

Early users described tasks such as navigating phone menus, waiting on hold and asking businesses for information.

If this becomes widely available and reliable, there are some obvious applications: checking appointment availability, contacting customer support, dealing with government agencies and gathering information from businesses that don't offer online booking.

Try this prompt if outbound calling is available to you:

As of October 28, native outbound calling wasn't listed as generally available in Grok Bot's official changelog. Similar workflows can be built with some third-party phone integrations, but I'd treat this as an emerging capability.

Grok Bot has been adding integrations with mobile applications, including Calendar, Reminders, Contacts and other phone capabilities.

Recent demonstrations also showed Apple Home functionality. That creates opportunities for more context-aware routines.

For example, you could have a bot check your calendar for conflicts, add reminders or perform a narrowly defined smart-home action.

Try this prompt:

The mobile and home integrations are rolling out gradually, so availability will depend on your device and account.

What's new: Grok Bot introduced a Shopify connector, along with Order Concierge and Deal Watcher templates for tracking purchases and prices.

Why it matters: You can delegate some of the research, monitoring and administrative work involved in shopping or running an online store. The most practical workflows often stop short of making the actual purchase. They prepare the information or cart so you can make the final decision.

One user described giving a household bot access to a year's worth of Walmart orders. The bot reviewed 78 purchases, created a catalog of frequently bought items in Notion and began preparing a grocery cart every Sunday.

The user still reviews the cart and completes checkout.

It's an appealing example because the task benefits from remembering your habits and preferences.

Try this prompt:

In the user's initial demonstration, the bot forgot soap. A small reminder that a good shopping assistant still needs a checklist.

Grok Bot introduced Order Concierge and Deal Watcher templates in its October 26 update.

Order Concierge helps track purchases, deliveries, refunds and returns. Deal Watcher monitors products for price changes and coupons.

Neither requires an especially elaborate setup, which makes them good starting points.

Try this prompt:

Because the templates are relatively new, I'd inspect their permissions and run a test before trusting them with routine account actions.

Grok Bot's Shopify connector lets you retrieve store information and work with certain administrative tasks without constantly opening the Shopify dashboard.

For a small business, I'd start with a daily operations report.

Have the bot summarize yesterday's orders and revenue, identify low-stock products and flag orders waiting too long for fulfillment.

Try this prompt:

Once the reporting is reliable, you can explore more specific administrative workflows.

There's a common mistake I see in a lot of agent demonstrations: someone creates an elaborate automation before establishing whether the underlying task can be performed reliably.

I'd reverse that order.

Start by giving your bot one real task. Define what you want accomplished, what information it needs and what a successful result looks like.

Let it do the task once while you're paying attention. Check the work. Fix whatever went wrong.

Once you have a process that works, save it as a skill. Include the steps, exceptions and situations where the bot should stop and ask for help.

Then add a routine so it runs automatically.

And finally, make sure you're checking the result rather than simply trusting that the routine completed.

Here's a general-purpose prompt you can use as a starting point:

This framework works across most of the examples in the guide.

The two pieces I'd spend extra time on are the expected result and the permission boundaries. Those are often what separate a useful automation from one that creates more cleanup work.

After going through these updates, what stands out to me is how many small pieces of everyday work are becoming easier to delegate.

Think about how much time you spend moving information between applications.

A customer emails you, so you research their company. A meeting ends, so you update the CRM. Someone makes a commitment in Slack, so you create a reminder. An invoice arrives, so you download it and save it somewhere.

None of these tasks is particularly difficult. But they interrupt other work, require context switching, and add up.

For years, automating them generally meant connecting APIs, configuring an automation platform, or writing custom software.

Those options still have their place. But with tools like Grok Bot, more of the setup can happen through conversation, and the assistant can increasingly handle the steps between applications.

There's also a shift in how the work gets started.

Instead of opening a chatbot whenever you need something, you can have workflows respond to incoming information. An email arrives. A customer submits a request. A meeting ends. A monitoring routine finds something worth your attention.

Your bot can begin the relevant work and bring you the result when it's ready for review.

We're still early. Reliability varies, integrations have limitations, and some tasks need more human oversight than product demos suggest.

But I think there's already enough here to change how many of us approach repetitive work.

My advice is to start with one or two workflows that have a clear payoff. Get them working consistently. Build in verification and approval rules. Then look for the next task worth delegating.

A handful of dependable automations will do more for you than 50 half-configured bots.

Research cutoff: October 28, 2026. This guide draws on Grok Bot's documentation, release notes, product demonstrations, and user-reported experiences. Some use cases predate October but have become more practical through recent updates. Feature availability varies by plan and rollout. The example prompts are starting points to adapt and test, not independently verified implementations. User-reported results shouldn't be treated as guarantees.