Torras.ai

On The Visory Podcast, host Tyler Hart at Visory AI and I talked about what happens when artificial intelligence moves beyond being a feature and becomes part of how a company actually operates. We traced that shift from the AI systems I studied in the 1980s to today’s agents, inexpensive software, and the new company I am building, Guanta AI.

Watch the conversation

Highlights from the conversation

1. I believed in AI before customers did

My path into artificial intelligence began at university in 1985, when natural language processing and neural networks were tiny versions of what we know today. Years later, when I told potential customers that we were building AI, their reaction was often like watching a bad old science-fiction film: “That is not a thing, and it never will be.” The technology was premature, but the need never disappeared. We still wanted computers to understand humans and work for us.

2. ChatGPT was an industry-reset moment

Like many people, I first opened ChatGPT and simply typed “hello.” A few questions later, it was obvious that the rules had changed. This was not a small product improvement; it was one of those rare moments when an industry is reshaped. Large language models and transformers were here to stay, and I had the privilege of studying and teaching that transition while it unfolded.

3. Humans became the middleware

Every company has important processes living in spreadsheets, Word documents, inboxes, and people’s daily routines. Humans copy information between systems, reconcile versions, send reminders, and keep the workflow moving. Much of that activity could be software. I call it human middleware: people performing the integration layer that the company’s technology never implemented.

4. Every company has a corporate operating system

A business already behaves like a digital system. Its CRM, ERP, financial models, documents, and informal routines together form a corporate operating system. Some of it is encoded in software; much of it still runs in people’s heads. The companies that build the best operating systems will reach more customers, deliver faster, and offer more for less. AI is becoming the layer that can connect and automate those processes end to end.

5. Guanta AI is built around automation, observability, and adaptation

At Guanta AI, we are working on three connected ideas. First, automate real customer processes from end to end. Second, make agentic work observable, so a company can understand what happened and why an AI made a decision. Third, adapt the platform to the customer. When software is becoming dramatically cheaper to create, a platform should be able to grow new capabilities instead of forcing every company into the same rigid template.

6. Knowledge workers will become workflow designers

If your work happens mainly in front of a laptop, it is digital—and some part of it can probably be done by AI. That does not mean every knowledge worker disappears. I expect something more interesting: every knowledge worker will start to look like a software engineer. People will stop running repetitive workflows with their own attention and instead define, improve, and supervise the systems that run them.

7. A recurring spreadsheet is software waiting to be built

When the same spreadsheet appears every week or every month—with the same columns, slightly different data, and perhaps a few new colors—that spreadsheet is a symptom. There is a software process underneath it, but humans are making it run. In the past, building a dedicated application was too expensive. Now it may soon be easier to build the application than to maintain the formulas, manual steps, and fragile file that imitate one.

8. Fast-moving AI changes the build-or-wait calculation

AI infrastructure is improving so quickly that a project started three months later can sometimes finish sooner, more cheaply, and with a simpler design. It resembles the old interstellar “wait calculation”: a later, faster spacecraft can overtake one launched earlier. That does not mean doing nothing. Early experiments teach us where we want to go—but founders should be careful about building an entire moat around temporary access to the same models everyone else can use.

9. The practical first step is to start now

For anyone who feels behind, my advice is deliberately small: get access to a capable AI tool and experiment today. Automate a personal administrative task. Explore how an agent handles a simple workflow. Learn through use instead of waiting for a five-year corporate plan. The era of unlimited free AI will not last, and practical fluency will matter. Fortunately, learning it is also genuinely fun.

10. A new AI company with a thousand-year-old name

Guanta takes its name from a valley, creek, and castle near Barcelona whose written history reaches back more than a thousand years. I liked the contrast: a very old local name for a very new kind of company. The technology changes at extraordinary speed, but it is still useful to remember where we come from while we build what comes next.