A practical playbook for delegating your inbox, research, schedule, and recurring admin
your inbox, calendar, and personal admin have been sharing one employee: you.
every recurring task charges twice. once when you do it, and again when you remember that it needs doing. the meeting prep, the reply you owe, the receipt you still have to find.
Grok Bot lets you give that work an owner. named bots use a persistent cloud computer, connected apps, and saved routines to carry assignments through several steps.
start with one task you can check. then build a small team around the work that keeps returning. the aim is to open a result you can use and see exactly which decisions still need you.
01. install it, then meet the Chief
install Grok Bot from x.ai/bot, complete the account setup, and create a bot called Daily Chief. the desktop app is available for macOS, Windows, and Linux with an eligible plan. start with one named bot and one real assignment.
Daily Chief becomes the person you message when work needs an owner. later, it can coordinate specialists. for now, give it something small enough to check against the source material: tomorrow's morning brief.
the first chat defines the finish line. the brief needs AGENDA, SOURCES, and OPEN QUESTIONS. include all three. the highlighted Open questions row draws attention to the part most briefs miss: what is still unknown, missing, or contradictory.
- agenda - meetings, commitments, and preparation
- sources - links back to the calendar, messages, and notes
- open questions - missing inputs and decisions still waiting for you
check the result before handing over a bigger job. can you trace the important details to their sources? are the gaps visible? correct one real output, then use that feedback in the role you define next.
02. give it a job title and a role contract
a recurring job needs a clear owner. name the bot after the responsibility it carries: Daily Chief, Inbox Lead, Research Scout, or Life Admin. keep returning to the relevant conversation so corrections build on the work already done.
write its charter like a first-day brief. the role contract in the illustration has four fields: ROLE, INPUTS, OUTPUT, and WHEN TO ASK. together, they explain what the bot owns, where it works, what it returns, and which decisions come back to you.
be specific about the last field. a missing attachment should be named in the brief; a missing fact that changes the whole assignment may need a question before work continues. a draft can be prepared while its eventual sending waits for approval.
refine the charter when a real result exposes something you forgot. a correction such as “put deadlines first” should become a lasting instruction for this role.
03. connect the tools once
give the job access to its actual sources. open Connect apps, add the supported connectors you need, and complete authentication. use the browser for a service or step the connector cannot handle.
the illustration shows the change in everyday work. finding files, opening apps, and carrying context between new chats takes your time. a persistent workspace gives the bot somewhere to keep the files and sessions behind the assignment, so you can return to the result.
your bots share one account-level cloud computer, including files and signed-in browser sessions. connectors are account-wide too. separate roles keep responsibilities focused; they do not create separate access boundaries. choose the sources the team actually needs and give shared files clear names.
connect deliberately at the beginning. when you add a specialist later, it can use the existing workspace without making you assemble the same context again.
04. hand off the login, then let the bot resume
a site may need a sign-in or verification step before the bot can continue. when that happens, open Agent Computer, take control, complete the blocked step, and return control to the bot.
the chat shows the handoff: the Chief waits at the sign-in page, you authenticate in the browser, and “Signed in. Continue.” returns the assignment to the bot. it resumes the brief from the point where access blocked it.
keep passwords and one-time codes in the authentication flow. the conversation needs the blocked task and its status. if the session expires later, use the same handoff again.
a useful pause is precise: which source is waiting, what you need to do, and which piece of work will resume afterward.
05. show it once, then save the method
the best workflow to teach is one you repeat, whose inputs are clear and whose result you can check. opening the calendar, checking the inbox, and preparing a brief is a useful starting point.
when Teach a task is available, demonstrate the browser workflow and review the skill it produces. if the control is unavailable, ask the bot to capture the method from the completed task and written instructions.
the illustration follows one good run into a saved skill, then into Monday, Tuesday, Wednesday, and Thursday briefs. each run needs fresh inputs. save the decisions behind the clicks: what to check, what counts as complete, and how to handle a missing source.
run the saved skill on another day's inputs. that test tells you whether it captured a reusable method rather than the details of one successful example.
06. turn useful work into a routine
once the output is worth repeating, give the saved method a schedule. confirm the owning bot, time zone, sources, and next run. inspect a test result before relying on the routine.
the morning example uses 07:30 Europe/Moscow. start with Daily Chief doing the brief. after you add the specialists in the next step, the same routine can collect updates from Inbox Lead, Research Scout, and Life Admin, with the Chief assembling the result.
cloud work can continue while your laptop is closed. the schedule starts the run; allow time for it to finish. if you need the brief ready by a fixed deadline, choose an earlier start.
schedules cover the morning brief and the weekly close. event-triggered work can come later: for example, an incoming pricing inquiry can start a draft. supported event triggers may need their own integration setup, so define a narrow matching rule and verify it before relying on it.
07. hire specialists with clear ownership
add specialists when the work splits into distinct responsibilities. an inbox role, a research role, and an admin role each give you a focused conversation to review and improve.
the team in the illustration has one owner for each kind of work. Inbox Lead handles drafts and correspondence. Research Scout handles research and sources. Life Admin handles paperwork, calendar logistics, and practical details.
Daily Chief owns the finished result: assemble the specialist outputs, check them, and deliver one brief. create the specialists, give each the relevant role brief below, and keep the final coordination instructions with the Chief.
split by responsibility before adding more bots. every specialist should have a deliverable you can recognize, and every shared assignment should have one final owner.
08. put the team in a group chat
bring the relevant bots into a group conversation when a job crosses their responsibilities. give them an objective and a final owner, then let them divide the work and make the handoffs visible.
an upcoming client visit is a useful example. the finished packet needs correspondence, research, and logistics in one place. each handoff should include the completed work, its sources, and the questions the next bot still needs to resolve.
use the group for coordination and the direct conversation with Daily Chief for the final review. the chat shown here ends with a Client Visit Packet, source links, and open questions. “Review the packet” is the next step; outgoing messages and bookings remain on hold.
the shared thread should reduce the notes you copy between conversations. when the work reaches you, the useful pieces should already be assembled into one result.
09. draw the approval line
decide which work the team should finish on its own and which actions need your judgment. drafting, researching, summarizing, and preparing a packet can happen within the access and scope you have already given it.
the approval path in the diagram returns from Daily Chief to YOU / FINAL CALL. require a concrete proposal: ACTION, TARGET, and REASON. a draft reply should arrive with its recipient and source thread; a booking proposal needs dates, price, and terms.
match the approval settings to the role instructions. review requests belong in the relevant bot's direct conversation. when a source blocks progress, the Chief should name the missing input and the affected task so you can resolve the actual problem.
finish the preparation before asking for the decision. “approve?” becomes useful when you can see exactly what will happen next.
10. review weekly and keep what earns its place
put fifteen minutes on the calendar for a weekly review. check whether each routine ran, whether its output was right, and whether you would miss it if it stopped. inspect at least one completed result against its sources.
use the three categories in the chat: USED, REWORKED, and IGNORED. the daily brief may be useful as it stands; the research notes may need clearer instructions; duplicate alerts may be adding work instead of removing it.
give the bot your feedback where it cannot see what you actually used. in the example, the next changes are to tighten the research instructions and pause duplicate alerts, while keeping the daily brief. make those choices explicit, then check the next run.
a routine earns another run when it reduces the work you carry, including the effort required to review it. keep the useful outputs, repair the ones you keep rewriting, and pause the noise.
start with the job you want off your plate
your first useful team might be a Chief, an inbox specialist, a researcher, and a bot that keeps the paperwork moving.
start with one real assignment. define the role, connect the sources, and review the finished work. save the method that works, then give it a schedule. add specialists when there are clear responsibilities for them to own.
the skill is learning to delegate a complete job: what should be finished, which sources matter, and where the bot should bring a decision back to you.
save this guide. choose the task you would otherwise do again tomorrow. hand it to your first bot, then judge the result before you add the next routine.
