The most common mistake of week one is handing the agent the hardest thing and being disappointed immediately. Company strategy, hard negotiations and creative decisions will not be the first task. Boring routine will.
Week one: what you already do by hand
- Triaging incoming mail: what needs an answer today, what can wait, what to delete.
- Meeting notes: a recording becomes decisions and a task list.
- Reminders about commitments: who you promised what and when it is due.
- Searching your own files and notes when you remember the gist but not the location.
These tasks are good because a mistake breaks nothing while the benefit is visible immediately. In a week the agent accumulates context about your work and stops being a stranger.
Week two: numbers and reports
Once the agent knows your files, hand it what you assemble every Monday: revenue for the period, the list of debtors, project status. The key difference from a chatbot in a browser is that it takes numbers from your data instead of inventing plausible ones.
Week three: external systems
Next come the calendar, spreadsheets, CRM and servers. Here the agent starts not only answering but acting: creating tasks, drafting emails, checking that a service is alive.
A rule we recommend to everyone: anything involving money, outside people or deleting data happens only after your confirmation. Not because the agent is stupid, but because the cost of a mistake is asymmetric.
What not to expect
An agent does not replace the person who makes decisions. It removes the preparation for a decision: gather, reconcile, verify, remind. Expecting otherwise guarantees disappointment — not because the technology is bad, but because the task was framed wrong.