What this episode is about
Íñigo Medina is CPO at Dcycle. He is also, unusually, a philosopher of technology. He does not study code; he studies the layers of culture, history, and human behavior that explain why we build what we build, and why organizations adopt it the way they do.
The episode starts with a provocation: Salesforce is now investing primarily in agentic layers rather than user interfaces. Notion is doing the same. The “end of interfaces” narrative is everywhere. Íñigo is skeptical, and his argument is historical. Nothing dies on the internet. What we actually get is more layers, more cognitive load, and a medium that keeps accumulating until it becomes what he calls a “gazpacho.”
Internet as a gazpacho
The metaphor is exact. A gazpacho is not a smoothie, not a soup, not a salad. It is all of those things blended together in a way that should not work but does. The same goes for the internet: ARPANET, browsers, mobile, voice, APIs, LLMs, terminals in full renaissance. None of these killed its predecessor. Each one added a new layer.
This matters for how we think about AI. The organizations that are investing in agentic interfaces while still running ERPs that look like they were designed in 1993 are not going to wake up one day to find the old layers gone. They are going to find one more layer on top of the others. The question is whether anyone has thought through what that means for the humans who still have to use the software today.
Íñigo points to one number that makes this concrete: agentic AI usage grew 8,000% in the last five to six months. But the same data shows a sharp increase in review times, because neither technical nor non-technical profiles fully trust the outputs they receive. The human review layer, especially in legal, quality, and finance, where a 1% error is serious, is gaining value, not losing it.
Organizations as adolescents
The second thread is organizations and how they respond to technological change. Íñigo’s diagnosis is direct: organizations behave like adolescents following a trend, with intensity and without nuance. The antidote, from Aristotle to Spinoza, is education.
Not LinkedIn education. Not conference-circuit education. The kind of education that means reading a paper a week, understanding the historical series of technological shocks on the labor market, and touching the material yourself rather than delegating that touch to a slide.
When economists study the historical record of technology’s impact on the labor market, the finding is not “AI will destroy all jobs” or “AI will destroy none.” It is more nuanced: when motor cars appeared, some professions were displaced, not eliminated. There was no mass unemployment, only a shift. The same pattern has repeated across every major technological wave. The organizations that understand this can calibrate their response. The ones that do not end up either paralyzed or reckless.
News from the radar
Salesforce goes agentic. The CEO declared that Salesforce is developing products for non-user interfaces, betting on the agentic layer as the future of software. The question worth asking: does this bet help the humans who still have to use the software today?
8,000% increase in agentic AI usage. In the last five to six months, agentic AI usage grew 8,000%. The same period brought a sharp rise in review times. Adoption spreads faster than trust.
Who builds software when AI can build it? The “everyone is a builder” thesis sounds plausible but has never happened in technology. History shows the emergence of organic marketplaces around each new capability layer instead. WordPress did not turn everyone into a developer; it created a new tier of small-business service providers. AI may follow the same pattern.