One email a day on where software breaks. Written in Spanish.

The emails are in Spanish. For readers without Spanish, the talks and papers are in English.

For the last couple of years everyone has been asked to put AI into what they build. An assistant here, a plug-in there, a chat on top of the customer data. It works in the demo. Then the first bad day arrives: someone types what they should not, or by the tenth message it answers as if it never read the previous nine, or it does something nobody asked with a permission nobody remembers granting.

That has an explanation. Almost everything that breaks does, and it is rarely in the new tool. I have spent twenty years opening systems their owners considered closed: medical records of millions of people sitting open on the internet, back doors that had been inside company systems for months, an eight hour job that dropped to a tenth of a second once someone looked at where the time went. Later I was Principal Engineer in the AI department at a bank with 185,000 employees. Now I work on my own, inside companies, with the system running.

What I see there on a Tuesday, I write about on Wednesday. No names, the whole failure.

A ten minute audio, also in Spanish, arrives the moment the signup goes through. In it I explain why ChatGPT, Claude or any other model starts brilliant and turns dumb around the sixth message. It has an explanation and a fix, and almost nobody who uses these tools every day knows it.

Every day? Yes, I send one every day and it takes three minutes to read. On a day the topic is not interesting, it can go straight to the bin.

Is anything for sale? Sometimes. Ongoing technical advisory and team training. I say so when it applies and do not dress it up as advice.

And if I do not like it? Unsubscribing takes one click.

Daniel Alfocea