There is an old English proverb that keeps surfacing in my mind as the AI debate gets louder: the proof of the pudding is in the eating. You cannot judge something by its appearance, its promises, or the presentations made about it. You judge it by the result.
The AI conversation in 2026 is dominated by two camps. The first tells you AI will replace most knowledge workers — that if you cannot prompt your way through the next decade, you will be left behind. The second tells you the whole thing is overhyped, that nothing fundamental has changed, that the productivity gains are marginal. Both camps spend most of their time talking. Neither has shown you the pudding.
This is my attempt to show you the pudding.
The Problem
In May 2026, I found myself in a situation familiar to many senior executives: multiple significant opportunities arriving simultaneously, a personal brand that had not been updated in years, no website, outdated documents, and a need to go to market fast — without six weeks and a branding agency.
Twenty-five years of career capital. Zero infrastructure to present it. Three active opportunities at companies including a VC-backed US deep tech startup, a global solar tracker manufacturer, and a top-tier executive search firm. Each requiring a different positioning, a different register, a different commercial story.
I decided to do the entire thing with AI — and to document it honestly.
The question was never whether AI could produce content. The question was whether it could produce the right content — and what the difference between those two things actually looks like in practice.
What Was Built
Across two working sessions, the following was produced, deployed, and in several cases already sent to real hiring managers at real companies:
Outputs — Two Sessions · May 2026
The outputs were real. The applications were sent. The website is live.
That is the test that matters.
Where the Human Made the Difference
Here is the part that most AI case studies skip — because it is uncomfortable. It reveals what AI cannot do. And that, paradoxically, is the most useful information of all.
AI is extraordinarily capable at the production layer. It writes, formats, codes, designs, researches, and iterates faster than any human and at a fraction of the cost. But across these two sessions, there were specific, verifiable moments where 25 years of commercial judgment changed the output in ways that no prompt could have produced.
| The Judgment Call | Why It Mattered |
|---|---|
| Knowing which question to ask before asking it | The single most important discipline in this entire process was forcing alignment before generation — establishing exactly what the output needed to achieve before a word was written. AI produces confidently in any direction. Knowing which direction is worth pursuing is the experience layer. It is also the most transferable skill from 25 years of complex negotiations: you define the objective precisely before you open your mouth. |
| Reading the buyer, not the brief | AI defaults to the most recent and prominent roles — the obvious answer. The correct answer for one application was a role from 15 years earlier that matched the buyer's specific commercial challenge exactly. Identifying that required understanding both what the buyer was actually trying to solve and which moment in a 25-year career contained the proof. No dataset contains that pattern recognition. That is what deep domain experience looks like as a cognitive asset. |
| Calibrating register across audiences — instantly | Three different applications. Three completely different registers. US VC-backed startup: direct, confident, outcome-led. Spanish executive search firm: warm, competency-framed, IESE alumni register. Global tracker manufacturer: track record first, numbers second, relationship third. AI can execute any register when directed. Knowing which register to direct it toward — without being told — is not a prompting skill. It is a career's worth of reading rooms. |
| Knowing what to leave out | Several powerful career narratives were deliberately excluded from specific applications. The reasons: awareness of how certain deal structures from the 1990s read through particular legal and cultural lenses in target markets. What is left out of a document is often more important than what is included. AI has no instinct for strategic omission. It fills the page. The judgment to leave things out cannot be prompted. |
| Knowing when not to act | Several proposed outputs during the process were correct in form but wrong in moment. A document ready to send but requiring one more piece of information. A message drafted but better held until a connection was accepted. The right content delivered at the wrong moment is the wrong content. AI has no concept of timing. That calibration is pure experience — and it is the difference between a process that moves and one that stalls. |
| Verifying what confident looks like versus what accurate looks like | AI produces with equal confidence whether it is right or wrong. Across these sessions, several facts required direct verification against the primary record before they could be included in any document sent to a real hiring manager. The discipline of treating AI output as a first draft requiring expert review — not a final answer requiring light editing — is a calibration that most people have not developed. It is the difference between a process that builds trust and one that erodes it. |
Senior engineers are not irreplaceable because they know syntax. They are irreplaceable because they understand architecture — what to build, what not to build, and why the difference matters. The same is true in commercial leadership. The judgment layer is not what AI replaces.
The Verdict
What This Actually Demonstrates
AI without experience is noise. It produces content — sometimes impressive content — without the judgment to know whether it is the right content, for the right audience, framed in the right way, with the right things left out. The outputs look good. They may even read well. But they will fail at the moments that matter most.
Experience without AI is slow. The same work — the website, the ebook, the three application packages, the research, the deployment — would have taken weeks with an agency and cost significantly more. Speed and cost are genuine advantages that compound over time.
The combination — when the human brings genuine domain expertise, exercises real judgment, and corrects the AI when it is wrong — produces something neither could achieve alone. Faster than experience alone. More intelligent than AI alone. Verifiably better than both.
That is the pudding. It is live at christianherrero.com. The applications have been sent. The results will follow.
A Note on the Companies Eliminating Judgment
There is a narrative running through the technology industry that deserves scrutiny. Major companies are eliminating tens of thousands of roles and citing AI as the reason. Some of this is genuine — certain production roles are being automated. But a significant portion, in my view, is something else: using AI as a restructuring narrative that plays well with investors and improves EBITDA multiples. The hype provides cover for decisions that would otherwise require more uncomfortable explanations.
What these companies are actually eliminating — in many cases — is not redundant production capacity. They are eliminating the judgment layer. The people who know which problems are worth solving. Who understand why a particular solution will create new problems downstream. Who have the institutional memory to know what has been tried before and why it failed.
That layer is very difficult to replace. And the companies that discover this will do so at the worst possible moment — when a competitor with both AI capability and retained judgment moves faster, positions better, and closes the deals that matter.
The proof, as always, will be in the pudding.