AI assistant vs AI that automates: drafting vs doing
Prices and features checked 2026-11-13
The line between the two is blurring, because assistants are gaining the ability to act. But the difference still decides how much of your week you get back.
Drafting vs doing
Picture a supplier's bill arriving by email.
- An assistant reads it and tells you the amount, the due date and the job it relates to. It might draft a reply. You then log into Xero, enter the bill, code it and file the PDF.
- An AI that automates reads it, enters the bill into Xero against the right job, files the PDF and tells you it's done. If the amount doesn't match the purchase order, it stops and shows you why.
Both are useful. The assistant saves thinking time. The automation saves the clicking, copying and remembering that fill up a small business owner's evenings.
For writing a proposal, an assistant is all you need, because the hard part is the thinking. For the fortieth invoice of the month, the hard part is the doing.
Approvals: the AI should ask before it acts
An AI that can send email or change records needs a way to check with you. The good ones let you decide what needs a yes.
- Claude asks for approval before each email it sends through Gmail by default. On Team plans, owners decide whether members can let those actions run without asking.
- ChatGPT documents app permission levels: always ask, ask on any changes, ask on important actions, or never ask.
- The Everything asks before anything important goes out. It confirms before any ServiceM8 change, and in Word each change is approved.
A sensible starting point is to approve everything for the first few weeks. Loosen it for routine jobs once you've seen the AI get them right. Never loosen it for money, contracts or anything going to a customer for the first time.
Triggers and schedules: work that starts without you
An assistant waits to be asked. Automation can start on its own, in two ways.
- On a schedule. A morning brief at 7am, a weekly report on Friday, a monthly spend audit. Claude Team's scheduled tasks run hourly, daily, weekly or on weekdays, even with your computer asleep. ChatGPT tasks can run at a set time or on a recurring schedule. Microsoft's Autopilot is described as continuing work in the background, billed by usage. The Everything sends a morning brief in Teams.
- On an event. A new email arrives, a quote comes back, a booking is cancelled. Ask each vendor which events in your own systems can start a job, not just which timers it offers.
Scheduled work is where the time savings add up, because nobody has to remember to ask.
Connectors: the AI can only do what it can reach
To enter a bill in Xero, the AI needs a connection to Xero. To book a meeting, it needs your calendar. This is often the deciding factor.
- Copilot has Copilot connectors and Copilot Studio for building your own.
- Claude has connectors including Gmail, Google Calendar, Google Drive, GitHub, Microsoft 365 and Slack, plus remote and local connectors.
- ChatGPT has plugins, connectors and MCP hooks.
- The Everything has 22 live integrations, including Outlook, SharePoint, Xero, Trak, ServiceM8 and Tradify. Two are in early access and 12 are “Ask us”, such as MYOB and QuickBooks. See the integrations list.
A connector you have to build is a project. A connector that's already there is a job you can hand over today. Check the systems you actually use before choosing.
When each makes sense
An assistant makes sense when…
the work is thinking and writing; each task is different; a person needs to own the result anyway; or only a few people need it. Copilot, Claude Team and ChatGPT Business are all strong here.
Automation makes sense when…
the same jobs come round every day or week; they touch several systems; mistakes come from things slipping, not bad judgement; and you can describe what “done” looks like.
Many businesses end up with both. See how the four products compare on specific jobs in the job-by-job comparison, and what each costs for your team size in pricing compared.
Questions people ask
What is the difference between an AI assistant and AI automation?
An assistant produces drafts, summaries and suggestions that a person then acts on. Automation carries out the task in your systems, such as sending, booking or entering into Xero, usually after asking for approval.
Is it safe to let AI send emails for my business?
It can be, with approvals. Start with the AI asking before every send, then relax it only for routine messages once you trust the results. Keep approval on for money, contracts and new customers.
Do I need a developer to automate with AI?
It depends on the product. Tools such as Copilot Studio, Claude connectors and ChatGPT plugins can reach many systems with some setup. Products with ready-made integrations for your software need less.
Can ChatGPT or Claude run tasks on a schedule?
Yes. Claude Team's scheduled tasks run hourly, daily, weekly or on weekdays. ChatGPT tasks can start at a set time or recur. Microsoft's Autopilot covers similar ground and is billed by usage.
Will AI automation replace my office staff?
It's better at taking the repetitive work off them, such as data entry, chasing and filing, so they spend more time on customers and judgement calls. Someone still needs to approve and oversee it.
See whether it does your jobs
The Everything has a 14-day free trial, with no credit card. After that it's $400 AUD a month + GST for the whole company, with unlimited users.
Try The Everything free