Build or buy an AI agent? When off-the-shelf is the better call

We build custom AI agents, and we still tell some people to buy one. Support chat, meeting notes, inbox triage and basic WhatsApp replies all have decent off-the-shelf tools at a monthly price. If one of them does the job, it is cheaper and faster than anything we could build.

Buy when

• The job is common and the same in most companies: FAQs, meeting summaries, ticket tagging.
• Your data lives in one popular tool the product already connects to.
• A wrong answer is cheap and a person checks the important ones.
• You want to learn whether an agent helps at all before you invest in one.

Four signs you have outgrown it

• It needs data from your own systems that the product cannot reach, such as live stock, custom pricing or your back office.
• It has to take actions, not just answer: change an order, book a slot, update your CRM.
• Your process is what makes you different, and the tool forces you into its process.
• Per-seat or per-conversation pricing now costs more each month than owning the agent would.

The middle path

Many teams keep the bought tool for the common part and add a custom piece where it falls short: an integration with their own system, a workflow for one tricky case, or an agent that handles only what the product cannot. That is often the cheapest way to get most of the value.

Switch without starting over

Whatever you buy, export your conversations, keep your FAQs and rules in a document you own, and note every case where the tool failed. That record is the scope and the test set for a custom agent later, and it saves weeks.

If you cannot name three things the bought tool does wrong for your business, you do not need a custom agent yet.