An early WhatsApp tool began with repetitive work: open a conversation, prepare text, check the number and save the outcome. I explored this with ChatGPT Codex and Claude so I would not have to reconstruct each contact’s history manually every day.

That task grew into my early WhatsApp tool. With ChatGPT Codex and Claude, I worked through preparing outreach, keeping contacts nearby and remembering whom I had already spoken to after each working day.

Memory matters more than speed

If a system sends quickly but does not know a contact’s history, it only accelerates trouble. One person receives another approach, another has opted out, a third replied — yet the table still marks them as new.

Status should answer a simple question: what actually happened? Prepared text has not yet been sent. An opened conversation does not mean delivery. The operator’s intention is not the action’s outcome.

These distinctions gradually produce a data model, then an interface. The reverse is tempting: make an attractive list with green checks, then work out what they actually confirm.

Why this is not about bypassing limits

Automation does not make an unwanted message welcome. No delay between sends makes outreach guaranteed safe or justifies promising freedom from blocks.

For a working tool, limiting the audience, making opt-out clear and retaining operator control matter more. Speed without these is like adding an engine to a car and forgetting to check the brakes.

In the broker CRM, preparing contacts and actions is separate from manual sending. Repeat processing is protected against duplicates; opting out blocks further outreach.

What grew out of it

Today I distinguish three tasks: prepare a contact, help a person reply and preserve conversation history. They may need different tools.

My broker CRM handles working contacts and follow-up actions. The separate WhatsApp CRM project explores a manager’s work with team conversations. They grow from one pain point but are different systems, with different data and responsibilities.

In the early tool, AI helped development. A language model’s presence did not determine whom it was permissible to message or what counted as a successful send.

How I frame this task now

I start with states: contact cleared for work, outreach prepared, action completed, response received, person opted out. Then I check the transitions.

What happens on a double click? A network failure? Reopening a tab? These questions are where an impressive demo ends and a useful program begins.

What remains from the early tool

I can help turn it into a small controlled process: contacts, history, tasks, message preparation and clear limits.

This post preserves the early WhatsApp stage. The standalone broker CRM and team-conversation project came later; their functions should not be attributed to the first tool. A separately verified repository for this version has not yet been found.

Related stories: Broker CRM · WhatsApp for a team.