From Punch Cards to GPT-5: Four Decades of AI in One Career
On this episode of João C. Sousa podcast we talked about something I rarely get to do in one sitting: trace the evolution of AI through an actual working life — from 1980s symbolic AI and punched cards, to building a semantic search company in Barcelona, to raising money in Silicon Valley, to watching transformers and ChatGPT change the rules again.
Watch the conversation
Highlights from the conversation
1. A very Catalan start: Sabadell, big family, two flats
I told Joao about growing up in Sabadell in a flat where we couldn’t all fit — so part of the family lived in the apartment below. Two grandmothers, an aunt, five siblings. It felt like one big household stacked on top of itself. That kind of environment makes you comfortable with people, noise, and improvisation — useful later when you manage teams across countries.
2. Getting into computers in the early 80s
My older brother told me, “this computer thing might have a future.” Back then there were no screens, no keyboards — you programmed with punched cards, waited two days, and got back a printout saying “syntax error.” That era forces precision and respect for computing.
3. AI keeps disappearing once it works
We talked about AI’s funny property: once something works, people stop calling it AI. In the 80s, finding the shortest path was AI. Today it’s just Google Maps. Maybe in 10 years we’ll say the same about ChatGPT: “that wasn’t AI — AI is what we have now.”
4. From services to product: building semantic search
We started as a services company helping companies improve their search, because keyword search didn’t really understand language. Then we realized we could turn it into a product and we built our own semantic search using Meaning–Text Theory — that’s the piece we patented.
5. Why timing in Spain helped
In the late 90s big consultancies in Spain were busy with Y2K and the euro conversion, so we grew in the “modern” space — internet, Java, Oracle — while they were fixing COBOL dates. Sometimes growth is just being available when others are busy.
6. The Silicon Valley leap
We won a contest that gave us space at Plug and Play in Silicon Valley. Suddenly we were in front of U.S. customers, investors and even partners from Chile and Japan. Being there unlocks conversations. That’s how we raised, and later partnered with NTT.
7. What being CEO actually means
One of our board members told me: “Jordi, everything is your fault.” If sales stall, if people are unhappy, if there’s no electricity — it’s on the CEO. The job is making the whole orchestra play in sync, even when it’s U.S. execs plus a European team.
8. Selling and still thinking “I could have done better”
We sold the company and it kept operating under new ownership. Yes, there’s pride — but also that founder voice that says “I could have done better.” That’s normal. There’s always someone who did it faster or bigger. The point is to build something real.
9. The two AI tribes — and the one that won
I explained to Joao the old “war” between symbolic AI (rules, grammar, lexicons) and connectionist/neural approaches (copy the brain: lots of neurons, let it learn). Transformers + OpenAI’s scaling made the neural side win so completely that today when people say “AI” they mostly mean big neural networks.
10. GPT is just “General Pre-trained Transformer”
Nothing mystical: GPT literally means General Pre-trained Transformer. It’s the same architecture that first did German→English, but trained on almost everything.
11. The wizard riddle GPT-5 finally solved
I had a logic riddle from university that I knew was not on the internet. GPT-3.5, GPT-4, even 4.5 couldn’t solve it. GPT-5 did — in about two minutes. For me that was a personal milestone: we’re past “fancy autocomplete” and into actual structured reasoning.
12. What happens to office work
I said it clearly: a lot of office work is just people moving digital information between systems. That’s exactly what AI will eat. It’s uncomfortable, but humans have always created new work once old work was automated. I’m optimistic, even if we can’t name the new jobs yet.
