Custom AI agent or off-the-shelf copilot? How to decide

A question that comes up in almost every call: "we already have Copilot in Microsoft, why do we need you?" Honest answer: sometimes you really don't. The difference is between a personal assistant and a digital worker - here is how to decide.

When an off-the-shelf copilot is enough

Summarizing emails and meetings, drafting text, asking questions about documents - personal work of an employee at their own screen.
When the information lives entirely in one ecosystem (Microsoft 365, Google Workspace) and the task ends in a suggestion the employee copies.
When the tool doesn't need to act in your systems - only to talk.

When you need a custom agent

When the process crosses systems. An invoice that starts in email, moves into the ERP and ends in a WhatsApp alert - no off-the-shelf copilot will run that route.
When you need execution, without a person in the middle. An agent that works in the background on every incoming invoice - instead of an assistant waiting to be asked.
When the data is sensitive. A custom agent runs in your environment, with your keys. You decide which model, where, and what it sees.
When the business logic is yours alone. Your approval rules, your exceptions, your order of operations - things that have no checkbox in an off-the-shelf product.

My rule of thumb: if the task ends in text a human reads - an off-the-shelf copilot is probably enough. If the task ends in an action inside a system - you need an agent built on your process.

And it's not either-or

In most organizations I meet, the right answer is both: a copilot for personal productivity, and custom agents on the 2-3 processes where the pain is biggest. Start with one, prove ROI, expand.

References

Microsoft Copilot agents - what's on the shelf
Why most AI agents never reach production