PlaybookExecutive1 week · 6 stepsUpdated 1 November 2026
Pick the agent that takes the most repeated, low-risk work off your team: a Reply Drafter if follow-up emails pile up, a Scheduler if meetings take days of back and forth. Run it for one week on one kind of task and approve everything it does. Then count what closed and decide to widen, adjust or stop.
Who it is for, and how long it takes
Who it is for
How long
What you need
What you end with
Executives and team leads handing routine work to an AI agent for the first time.
About an hour to decide, then one week of running with approvals.
A week of your team's sent email and calendar, one person to approve, and one clear first task.
One agent doing one kind of task, a count of what it closed, and a decision to widen, adjust or stop.
Key takeaways
Choose by volume and risk: plenty of repeated work where a mistake is easy to fix.
Start with one kind of task, such as supplier follow-ups or buyer meeting requests.
An agent knows how to do the task. It may not know whether the task still matters, so the handover must carry that context.
An AI agent may know how to execute a task. It may not know whether that task still matters to the company. A first agent goes well when it gets a narrow job, clear context and a person who approves every output for a week. This playbook helps you pick that job, and the same test applies to whatever agent you use.
Worked example. The examples use Vaultex, a safety-footwear manufacturer in Greater Noida and GeniOS's first design partner. The buyers, orders and dates are for illustration only.
PrepBefore you start
Have these ready before the agent starts its first task.
A week of your team's sent email and calendar, to count the repeated work.
One approver with ten minutes a day for the week.
A one-line first task, narrow enough that a new hire could do it on day one.
The context the task needs: current prices, open orders and who decides what.
Your first approval rules, even if they are three lines: what the agent may draft and what stays with people.
A simple edit log: what you changed in each draft, and why.
A date at the end of the week to decide what happens next.
StepsThe steps
1 week · 6 steps
Count a week of repeated work
Look at last week's sent email and calendar. Count the replies and follow-ups that followed a pattern, and the meetings that took more than two messages to book. Write both numbers down.
Mark the risk of each kind of task
For each kind of task, ask what happens if a draft is wrong and the approver misses it. A supplier follow-up is easy to correct. A price sent to a buyer is hard to take back. Start where the worst mistake costs an apology.
Pick the agent
If repeated replies are the bigger number, start with a Reply Drafter. If scheduling is, start with a Scheduler. At Vaultex, Harsh sends many supplier and buyer follow-ups, so a Reply Drafter comes first, and Anushka's quality-test bookings could make a Scheduler the second.
Define one task and its context
Write the first task in one line, such as "follow up with suppliers on open price questions". List what the agent must know: the current price, the open order and who decides. With GeniOS, Delegate hands the situation to the agent with that context attached, such as the steel toe-cap supplier's 8% price rise and the Bata order it affects.
Approve everything for one week
Nothing goes out until a person approves it. Read every draft or invite, edit when needed and note why. The same edit twice means the agent is missing context, so add it.
Count, then decide
At the end of the week, count what the agent carried to done with only an approval, and how often a person had to step in or rewrite. If people rarely stepped in, widen to a second task. If they often did, fix the context first.
TemplateCopy this task handover
Give the agent the same brief you would give a new hire.
Task: [one line, for example: follow up with suppliers on open price questions].
Why it matters now: [the order or goal it affects].
What the agent should know: [current price, open order, quantity, deadline, who decides].
Tone and limits: [for example: polite and short; never agree a price or a date].
Approver: [name]. Nothing is sent until they approve.
Done means: [the proof, for example: the supplier has replied with a confirmed price].
ExampleWorked example: Harsh's first week with a Reply Drafter
How the choice and the first week might go in sales at Vaultex.
The count. Harsh looks at last week. Many of his emails followed a pattern: supplier price questions, sample confirmations and buyer reminders. Meetings were fewer, and most were booked in one or two messages. Replies win, so he starts with a Reply Drafter.
The task. "Follow up with suppliers on open price questions." Prices to buyers stay with Harsh. With GeniOS, Delegate hands the first situation to the agent with its context: the steel toe-cap supplier's 8% price rise and the Bata order for 12,000 pairs it affects.
Days one and two. Harsh edits both drafts to add the open Bata quantity and the date Vaultex needs an answer by. He adds those two facts to the handover.
Days three to five. The drafts need only small edits, mostly tone. Harsh approves each in about a minute. One supplier replies with a revised price, and the item closes on that email.
The decision. Harsh widens to a second task, buyer sample confirmations. Anushka's quality-test bookings take several messages each, so a Scheduler is the next agent to try.
MistakesCommon mistakes
Five ways a first agent disappoints in its first week.
Starting with the riskiest work. A wrong price to a buyer is hard to take back. Start where a mistake costs an apology.
A task that is too broad. "Handle my email" cannot be judged in a week. Pick one kind of task.
Thin context. An agent may know how to write a follow-up and still miss the order behind it. Put that in the handover.
Approving without reading. A quick yes teaches nothing. Read, edit and note why.
Running two agents at once. You cannot tell which one caused a bad week.
What good looks like after
One agent handles one kind of task, and the approver trusts most of what it drafts.
Edits fall between the first day of the week and the last.
You know how much work the agent closed and how often a person stepped in.
The approver spends a few minutes a day on it, far less than the task used to take.
There is a written decision about the second task or the second agent.
The first approval rules exist, even if they are three lines long.
If the approver rewrites most drafts, the task is too broad or the context is thin. Narrow the task and run another week.
No. GeniOS thinks and delegates: it works out what needs doing and hands it, with context, to an agent such as a Reply Drafter or a Scheduler, which drafts. A person approves before anything is sent.
Can we run both agents from the start?
You can, but one at a time teaches you more. With one agent you can tell whether a bad week came from the agent, the task or the context.
When can we stop approving every output?
Only after the approver has stopped editing a kind of task for several weeks. Even then, write that permission down as an approval rule, and keep anything involving money or commitments to buyers with a person.