Does my WhatsApp agent forget the conversation? What Claude's new feature solves and what it does not
Anthropic has launched Message Threads in beta. It helps in long conversations, but it expires after 5 minutes. On WhatsApp, the memory of your customer support is still your system's job.
A customer asks if you have the trainers in a 38. Three hours later, they reply "put a pair aside for me". If the agent does not know what they are talking about, the sale goes cold right there.
Last week, Anthropic launched a feature that seems to solve this. It solves part of it. It is worth understanding which part.
What Message Threads is
On 2 November, the Claude API gained Message Threads, in beta. Until now, every message from an agent had to resend the whole conversation. With a thread, the conversation is stored on the server, and later requests carry only what is new.
According to the documentation, the time to first response becomes shorter and more stable as the conversation grows. That is good news for agents that do a lot of work in sequence, such as those that read documents or operate a system step by step.
Why this is not the memory your customer support needs
A thread expires 5 minutes after the last request and is deleted within 1 hour at most. On WhatsApp, customers reply when they can: at lunchtime, in the evening, two days later.
The documentation itself tells the application to keep its own copy of the conversation. When the thread no longer exists, the system resends the history and starts another one.
There are two more limits to consider. Threads are not covered by zero data retention agreements, which matters in regulated sectors. And a thread does not create a cache on its own: whoever develops the agent still has to configure that to keep the cost from rising.
What gives an agent real memory
The memory a customer notices is made of three layers, and they all live in your system:
- Conversation history, stored in your database and linked to the customer's number.
- Business data, such as order, stock and customer record, looked up in the ERP, the CRM or the shop at the moment of the reply.
- Relationship summary, with what has already been agreed, so the agent does not ask again.
A thread comes in as an optimisation when the customer is talking in sequence. It improves speed, but it does not replace any of these layers.
What this changes for your company
If you already have an agent, ask whoever maintains it where the history lives and what happens when the customer comes back the next day. The answer needs to be concrete.
If you have yet to start, start small. In our pilot AI agent on WhatsApp, the agent uses your company's knowledge base and one integration, and hands off to a person at the right time, with success criteria agreed upfront. In 2 weeks it is live for you to test with real customers.