Somebody posted on LinkedIn about applying for AI jobs, and they said a word.

The account went like this. They had spent years teaching themselves. Workflows, application programming interfaces, prompt engineering, automated systems, actual products that actually shipped. They finally started interviewing for AI-related roles, expecting to sit down and talk about what they had built. What they got instead was a sequence of questions. Are you an expert in AI. Do you hold formal machine learning credentials. Have you worked at an AI company before. Have you led an AI team.

Every one of those questions is a question about provenance. Not one of them is a question about the work.

I understood the frustration immediately, and I want to be precise about why, because this is not a post about one person's difficult job search. It is a post about one of the strangest contradictions currently operating in the labor market, which is that organizations are announcing at volume that this technology is the future while simultaneously restricting access to the roles that build it, using standards that do not yet meaningfully exist.

The goalposts are moving while people are running toward them.

What stood out most in that post was the point about shifting rules. People are being told to adapt quickly, learn the tools, become fluent, stay ahead, innovate, upskill, and reinvent themselves continuously. That instruction is delivered constantly, from every direction, with an edge of threat attached to it.

And then the people who follow that instruction, who genuinely do the thing they were told to do, run directly into hiring systems that have no functioning method for evaluating what they now know. So a loop forms. Companies announce a talent shortage in the same breath that they filter out curious, capable, self-taught people who do not resemble a credentialing pattern that was built for a different kind of work entirely.

That disconnect is not a small operational inefficiency. It is a statement about how unprepared many organizations remain for a transition they are publicly leading.

There are no long pipelines here, because there has not been time to build any.

This is the part that the interview questions cannot accommodate. There are no decades-long credentialing pipelines for most of these roles, because most of these roles did not exist at commercial scale a handful of years ago. We are watching an entirely new labor category emerge in real time, and we are inside it while it emerges.

People are experimenting deeply right now. They are learning through curiosity, communication, open source projects, video tutorials, Discord servers, trial and error, and building things independently because there was nowhere else to learn them. That is how people have always learned in emerging industries, long before institutions caught up and long before anyone issued a certificate confirming that the learning occurred.

Meanwhile a large number of hiring processes are still operating as though this were accounting, or corporate law, or medicine, fields with mature and universally recognized credentialing pathways. That mismatch produces a particular kind of harm, and the harm compounds because many hiring managers are not yet fluent enough themselves to distinguish between someone casually prompting a chatbot and someone who understands architecture, iteration, debugging, integration logic, workflow design, data handling, agent orchestration, and the discipline of building toolkits and manifestos that other people can actually use.

Those are not adjacent skill levels. They are different practices entirely. But when the evaluator cannot perceive the distinction, the distinction collapses, and the candidate absorbs the cost of a gap in the evaluator's fluency. The person being assessed pays for what the assessor does not know.

Fear is doing more work here than anybody wants to admit.

I do not think this is primarily malice, and I do not write from the assumption that hiring managers are acting in bad faith. I think a great deal of this is fear.

The technology is moving quickly enough that executives and hiring teams are attempting to manufacture certainty in a situation that does not contain any. And when certainty is unavailable, institutions reach reflexively for whatever is measurable. Credentials are measurable. Titles are measurable. Pedigree and institutional affiliation are measurable. A portfolio of self-directed work built at night across two years is not measurable in the same way, so it gets treated as though it is not evidence at all.

But emerging technology has rarely worked that way at the beginning. The people who become genuinely influential during a technological shift are frequently the people who were experimenting outside the established systems long before institutions understood what was being built.

There is an emotional dimension to this that I refuse to skip past. The labor market feels unstable. Layoffs are happening across sectors while expectations rise. More people are competing for fewer roles, and companies are asking for years of expertise in technologies that barely existed commercially at scale a few years ago. That produces a feeling of impossibility, and the people carrying that feeling are trying, seriously and continuously, spending nights and weekends and every spare hour building projects, watching tutorials, testing systems, learning interfaces, reinventing themselves in real time, and still finding themselves outside the gate.

The frustration is real, and it is proportionate to the situation.

Every technological transition produces gatekeeping, and this one is producing it early.

Every emerging industry eventually develops status hierarchies. Certain people become recognized as legitimate. Certain institutions become signals of authority. Certain credentials become shorthand for worthiness, and the shorthand eventually replaces the assessment it was meant to summarize.

What makes this particular moment unusual is that the field is still early enough that those hierarchies are unstable, inconsistent, and in some cases straightforwardly arbitrary. That instability is exactly why curiosity may matter more than pedigree right now. The people experimenting deeply are frequently learning faster than the institutions attempting to evaluate them.

To be clear, and I want this understood because the argument gets misread otherwise, expertise matters. Technical rigor matters. Safety matters. Machine learning depth matters. Engineering depth matters. I am not arguing for a field with no standards. I am arguing that many organizations remain unclear about the difference between hiring for highly specialized research and hiring for practical implementation, workflow design, operational integration, and emerging applications. Those are not identical jobs, and they do not require identical backgrounds. When a company cannot articulate the distinction internally, it unintentionally filters out precisely the adaptable talent it claims to need in order to execute.

The gate is frequently itself an AI system, which is where this becomes a governance question.

Here is the layer that the LinkedIn post did not reach, and it is the layer I work in.

A significant portion of this filtering is not being performed by a human being at all. It is performed by automated screening, keyword matching, ranking systems, and increasingly by models trained on the profiles of people who were previously hired into similar roles. Amazon famously abandoned an experimental resume screening tool after finding it had learned to penalize signals associated with women, because it had been trained on a decade of hiring decisions in a field where men dominated. That episode is not an outlier. It is the default behavior of any system trained on historical outcomes and asked to predict future fit.

So consider what that means in this specific context. A model trained on who previously held AI roles will learn that people who hold AI roles have machine learning degrees, prior employment at AI companies, and team leadership at AI companies. It will then rank candidates accordingly. It cannot recognize a self-taught practitioner with a serious portfolio, because the pattern it learned was assembled before self-taught practitioners existed in sufficient numbers to appear in the training data.

The system is not malfunctioning. It is functioning exactly as designed, and the design encodes a moment in time as though that moment were a permanent truth about competence.

This is a Translation Layer problem.

I have been building an AI governance framework I call the Translation Layer™, which I define as the governance infrastructure that enables communities most affected by AI systems to exercise actual decision-making authority over the development, evaluation, deployment, and nurturing of those systems. Its central distinction is between participation and power, and hiring is one of the cleanest places to see why that distinction is not academic.

Run the diagnostic.

Participation infrastructure. The people being evaluated hold no standing anywhere in the design of the evaluation. Candidates do not participate in defining what counts as AI fluency, what evidence a portfolio must contain, or which signals a screening system weights. The criteria are written entirely upstream of them and applied entirely to them.

Evidence redefinition. A candidate's demonstrated work is not admitted as evidence with the same authority as a credential. The shipped product, the working automation, the documented system, the toolkit somebody else is using in production, all of it is received as a claim requiring validation by a credential, rather than as evidence in its own right. That is the same move I have documented elsewhere, where the testimony of the affected party is heard, acknowledged, and then set aside as anecdote while the institution's own metrics are treated as fact.

Enforcement authority. There is no mechanism by which any candidate can require an employer to evaluate demonstrated work, to disclose whether an automated system screened them out, or to correct a determination that misread them. You can rewrite the resume. You can add keywords. You can apply again. What you cannot do is compel a review or require the criteria to change.

Three components, and not one of them exists for the person on the other side of the application.

What this is actually about.

That post was interesting to me because it was not simply somebody venting about a job hunt. It was describing a collision between institutional hiring habits and a labor market changing faster than the institutions attempting to sort it.

We are going to keep seeing this, because the future of work conversation is no longer only about whether this technology changes jobs. It has become a conversation about who gets recognized as qualified to participate in the economy being built around it, and, more importantly, about who holds the authority to do the recognizing.

Right now the rules are shifting underneath people who were told to move, and a substantial number of the gatekeepers do not yet understand the game they are guarding.

Naming that precisely is the precondition for changing it.