The moment of truth
Imagining systems is lovely. Designing your framework, debating ontologies with a machine that reasons, writing articles about graphs. Programming has always been part play, and more so now, with an AI taking the tedious hours of writing and testing code off your hands. But this trade is not won in the laboratory. It is won on a Tuesday at nine in the morning, when the phone rings.
— Hi, Joel. My PDFs won't open.
— Hi, Pepe. They won't open... from some folder? When you download them from the internet? The ones in your email?
— My PDFs won't open.
Anyone who has ever done support knows this conversation. You can spend twenty minutes trying to narrow the problem down with the best will in the world: Pepe knows exactly one thing, and it is that his PDFs won't open. It is not Pepe's fault. Pepe has his business, his clients and his day; the PDFs are your department. That is why he calls you.
The history of this trade — the real one, the one that never makes it into conference talks — is the history of what happens after that call. And as it happens, I have lived through all of it.
The 80s: the switchboard operator
In the eighties, Pepe did not call you. He called your office's switchboard operator, who left you the message. You called back, Pepe was in a meeting, his secretary took a message for you. When you finally connected, the good part began: a couple of hours on the phone with you playing the hands and Pepe playing the screen.
— Type c, then a colon.
And Pepe would type «c..».
After that impossible operator-to-technician séance, you arranged to visit his office — whenever his diary and yours allowed, which meant days later. The problem, meanwhile, sat there, breeding.
The 90s and 2000s: the mobile phone
Mobile phones arrived and the switchboard operator disappeared. No more crossed messages, no more days-long pursuit: Pepe called you directly, in your pocket, at any hour — progress and a life sentence in equal parts. But the process ended the same way: at Pepe's office, in a couple of days. We had sped up the beginning of the story; the ending was still the same.
The 2010s: finally, seeing the screen
Remote-access tools were the first true revolution. At last you could see what was happening from your own computer: look, test, touch. Half the mysteries dissolved in ten minutes once you saw with your own eyes what Pepe could not describe.
And for the other half there was the trade's old ritual: the format-and-reinstall. Because there were problems you never quite understood, but that were cured by a good wipe and a fresh start. Mind you: backups, formatting, programs, settings, tests... a couple of days with Pepe all over again. The liturgy had changed; the penance, not so much.
And then came the AI
And then the AI arrived, with its improbable ability to read and interpret Windows logs. Where you saw twenty thousand log lines, it saw the sentence: "this happened, because of this, and it can be fixed like this". The first time I watched it do that, I understood the trade had just changed more than in the previous forty years combined.
But of course: the AI had to know which computer is Pepe's. And how to reach it. And how to get in. And which program is acting up, and since when, and what was touched last time. All of that you explained to it, every single time, because the following week — another call, another mystery — the AI was born again remembering nothing.
The obvious solution was to write it all down for it. And there began the disease every veteran recognises: the AI read more and more files and added more and more annotations to them; everything filled up with reminders, readmes, notes about notes. Every session started slower, and reaching Pepe's computer cost more — in time and in tokens, which are the new billable hours. We had invented paper memory for an electric mind, and paper memory does what it has always done: grow, contradict itself and rot.
And one night I asked myself the question that started everything I write about on this blog:
How can the smartest tool humanity has ever invented be... this dumb?
It was not dumb
The answer I already told in detail in the loom article: it was not dumb. It was brilliant and amnesiac, which is a different thing. And amnesia is not cured with more notes: it is cured with real memory — structured, queryable and pruned.
Today Pepe's computer is a node in the loom. Its card says how to reach it, how to get in and what runs inside. Last time's fix is a procedure with its steps and its verification. Today's breakdown is a job that gets opened, logged and closed with its result in writing. When Pepe calls because his PDFs won't open, the AI does not start from zero: it consults, reaches the machine, reads the logs, recognises last time's trap — it is written down — and goes straight to the point.
The conversation with Pepe is still the same; no technology will ever change that one, and I am almost glad. What has changed is everything else: what used to be "let's meet Thursday at your office", and later became "give me a couple of hours and I'll connect", now often fits between one coffee and the next call.
The moment of truth
Forty years stand between the switchboard operator and the loom. Every leap in this story — the mobile phone, remote access, the AI, the memory — fixed the previous leap's bottleneck and uncovered the next one. The mobile's was locating people. Remote access's was seeing the screen. The AI's was understanding the logs. And the last bottleneck, the one almost nobody looks at because it is not photogenic, was this: making sure the one who understands the logs knows where Pepe's computer is without being told every week.
That is why I distrust architectures that have only ever been proven in demos. Systems do not show their true face when you design them, full of excitement, on a Sunday afternoon. They show it on a Tuesday at nine in the morning, with Pepe on the phone and the PDFs refusing to open. That is the moment of truth, and it has been asking me the same question for forty years, wearing different costumes.
The difference is that now, at last, there is someone else in the room who remembers the answer.
— an old programmer · 64 years old · rss