Skip to content
Book a meeting Flash Diagnostic

BlogGetting started

ChatGPT, Copilot or a custom agent: which is right for your company?

An off-the-shelf tool helps each person; a custom agent does a specific job in your tools, with your rules. How to choose, and when to combine the two.

By Stéphane Barrio Published on Updated on 2 min readFor SMEsMid-sized companies

"Why pay for a custom agent when we already have ChatGPT?" The question is fair, and the honest answer is: sometimes you do not need one. Here is how to decide.

Two families of tools, two uses

Off-the-shelf tools (ChatGPT, Microsoft Copilot, Mistral Vibe from Mistral AI and others) are assistants. A person asks them a question, gets an answer, and decides what to do with it. They are excellent for writing a first draft, summarising a document, rewording an email, finding an idea.

A custom agent is designed for a specific job: processing supplier invoices, preparing quotes, qualifying incoming requests. It is connected to your tools, follows your rules, acts without being chased, and hands over to a person when a decision commits the company.

The decision table

Your need Off-the-shelf tool Custom agent
Help everyone write, summarise, search ✓ ✗
Do recurring work without being chased ✗ ✓
Act in your tools: CRM, accounting software, email limited ✓
Apply your rules: prices, deadlines, tone every time, by the person once, in the agent
Measure a specific gain difficult agreed before starting
Initial cost low, per user a project, priced in advance

When the off-the-shelf tool is enough

  • The need is individual: everyone saves a few minutes on their writing.
  • The work does not come round in a comparable form.
  • You first want to get your teams familiar with AI.

In that case, the essential thing is to write rules of use: which data can be entered, who reviews, which tools are allowed. Without these rules, uses multiply without control. See Your teams already use AI without saying so.

When a custom agent becomes useful

  • A task or a process comes round every week, and costs hours.
  • The work requires your rules and your data: a generic tool gives an answer that is right for everyone, and wrong for you.
  • You want a measured result, not just usage.
  • The data is sensitive, and you want to decide where it is processed.

For a single task, our Task Automation puts it into service in one to two weeks, at a price agreed with you before we begin: no surprises.

The two can be combined

The most common case: teams use an off-the-shelf assistant for their writing, and one or two agents take on the processes that cost the most. It is not one against the other.

What we do when the off-the-shelf tool is enough

We tell you. It is one of the possible conclusions of our free Scoping Meeting: if an off-the-shelf tool meets your need, you leave with that answer. And if your teams already use these tools without a framework, the AI Usage Audit takes stock and gives you a usage charter.

Share on LinkedIn All articles

Read next

In the same category

Free · no commitment

Where to start? Two minutes to find out.

Get your AI maturity level in a few clicks, the offer to start with and your first use cases.

  1. 2 minthe Flash Diagnostic, online
  2. 1 hthe Scoping Meeting, free
  3. 72 hyour written report
Flash Diagnostic 2 min, free

Scoping Meeting

In 1 hour, find out what to hand over to AI and what you stand to gain; a written report within 72 hours.

Calendar provided by Google Calendar. Our privacy policy