This is the post I wish existed when I started. Not what agents could theoretically do in a perfectly set-up enterprise environment. What they can actually do for people running real businesses, right now, with the tools that actually exist.
Here are the five categories that cover most of what I've seen work in practice.
🔍 Research & summarise
Given a topic, a set of links, or a folder of documents — an agent can read through everything, extract what's relevant, and produce a structured summary. What used to take two hours of tab-switching takes twenty minutes of reviewing the output and pointing to anything it missed.
✉️ Draft & send communications
From a set of inputs — a brief, a list of names, a previous email thread — an agent can draft personalised communications at volume. Not mail-merge fill-in-the-blank. Actually contextualised drafts that sound like you wrote them. You review, adjust, approve, send.
🔄 Move information between systems
Copy data from a spreadsheet into a form. Pull information from emails into a tracker. Reformat data from one structure to another. These are tedious, error-prone tasks when done manually. An agent can handle them faster and more reliably than a human doing the same repetitive thing for the fortieth time.
🔔 Monitor & alert
Check something regularly and notify you when it changes. A competitor's pricing. Your own inbox for specific types of messages. A shared document for updates. Agents can sit in the background watching things you'd otherwise have to check manually on a schedule.
📅 Scheduling & follow-ups
Draft follow-up emails triggered by a time window. Remind you — or someone else — when something hasn't happened that should have. Schedule and prepare materials for recurring meetings. The stuff that falls through the cracks when you're managing a lot of moving pieces.
The common thread: tasks that are low-skill but high-friction.
None of these require a technical background to use. They require knowing what you want automated, being able to describe it clearly, and reviewing the output critically. The AI does the execution. You do the thinking about what's worth automating and whether the result is good enough.
Start with one. Pick the most annoying recurring task in your week. That's your first agent experiment.