For 500 years, Guilds were how expertise worked. Then we forgot. Now we're remembering.

In November 2025, Palantir, one of the most influential (and controversial) technology companies in the world, announced it had hired 22 high school graduates and given them something universities could not: a real apprenticeship.

The programme began not with code, but with Lincoln's Gettysburg Address. With Frederick Douglass. With the question of what it means to do consequential work in a free society. After a month of seminars, the fellows moved into live client deployments alongside senior engineers, doing the work that full-time employees do, evaluated not by examination but by the holistic judgement of practitioners who had already mastered the craft.

Alex Karp, Palantir's CEO, was blunt about what this was. Universities, he said, had become an "industrial complex". They are a system designed to process students through a predetermined sequence of prerequisites rather than develop the kind of contextual, evaluative, tacit knowledge that actually makes someone good at their work. "Skip the debt," Palantir's recruitment copy reads. "Skip the indoctrination. Get the Palantir degree."

What Palantir has built, without using the word, is a Guild.

And they are not alone.

What a Guild Actually Was

The craft guilds of medieval Europe such as goldsmiths, weavers, stonemasons and apothecaries are often remembered as protectionist trade associations. That reputation is not entirely undeserved. But it obscures something more important: they were the most sophisticated knowledge transmission institutions the pre-modern world produced.

At their peak in the 14th century, the great craft guilds of London, Florence, Bruges, and Cologne had solved a problem that still defeats most modern organisations: how do you transfer the knowledge that makes someone truly excellent at their work from the person who holds it to the person who is developing it?

Their answer was the three-tier structure of apprentice, journeyman, and master. This provided a progression not through examinations but through sustained, guided immersion in real work, culminating in the masterpiece: a single piece of work evaluated holistically by the community of masters who would decide whether the journeyman had earned the right to join their ranks.

The standard was not written down. It did not need to be. It lived in the collective judgement of the master practitioners — an internalised, shared, continuously-refined sense of what excellent work looked like, transmitted through years of working alongside those who already held it.

D. Royce Sadler, one of the most cited researchers in educational assessment, gave this a name in 1989: Guild Knowledge. "The knowledge and skills required to produce work of a given quality," he wrote, "reside largely in unarticulated form, inside practitioners' heads as tacit knowledge." The only way to transmit it is through the kind of evaluative immersion that apprenticeship provides. Not rubrics. Not mark schemes. Not criteria. Practice, comparison, calibration, and the judgement of people who already know.

The guilds had been doing this for five centuries before Sadler named it.

The Interruption

Then came the Industrial Revolution and with it, a different philosophy of knowledge.

Newton had described a universe governed by deterministic laws. Brunel had demonstrated that complex systems could be precisely specified and reliably built. Frederick Taylor applied the same logic to human labour: decompose the work, define the optimal method, eliminate variation, enforce the standard. The factory needed reliable, interchangeable workers, not craft masters with decades of tacit knowledge, but operatives trained to a specification.

Assessment followed the factory. If quality in a product could be specified and inspected against a standard, so could quality in a person's knowledge. The mark scheme was born. The standardised test. The competency framework. The rubric. These are instruments of a specific philosophical moment during the brief period, roughly 1780 to 1980, when it genuinely seemed as though the most important things about quality could be fully described in advance and reliably measured after the fact.

The craft guilds were, broadly, swept away in this period. Not because they were wrong about knowledge, but because the factory system required a different kind of worker and a different kind of assessment. The guild's holistic judgement was replaced by the inspector's checklist. Tacit knowledge was replaced by written specification. The apprenticeship was replaced by the examination.

This was not progress. It was a category error, one that took two centuries to fully expose.

What AI Revealed

The exposure came from an unexpected direction.

When large language models began generating essays that ticked every box on a mark scheme experienced teachers could immediately sense something was missing. The model had assumed that what matters about quality can be fully specified. AI demonstrated, empirically and at scale, that it cannot.

At the same time, something else was becoming visible in the workforce data. The people who worried least about AI were the most experienced ones. Ask a seasoned professional what AI has done for their work and you hear a consistent word: amplifier. It handles the routine. It frees them to focus on the work that requires genuine judgement.

Ask someone two years into their career and the word is different: threat.

This is not a confidence gap. It is a Guild Knowledge gap. The experienced practitioner has an internalised standard against which they can immediately evaluate everything AI produces. They can see what is right, what is subtly wrong, and what direction to push the model next. The novice, trained in a system built on criteria and specifications rather than judgement and immersion, has not yet developed that standard.

A Stanford University study tracking millions of workers found that 22–25 year-olds in the most AI-exposed roles saw a 13–16% relative employment decline following the widespread adoption of generative AI. Entry-level job listings fell by approximately 35% between 2023 and 2025. A 2026 global CEO survey found 43% of leaders planning to cut junior roles, nearly double the year before.

What is being eliminated, role by role, are the positions that traditionally built Guild Knowledge. The junior analyst who read reports before they could write them. The trainee teacher who marked scripts before they could calibrate their own sense of quality. The graduate lawyer who reviewed contracts before they could spot the risk. These weren't inefficiencies in the pipeline. They were the pipeline. They were how organisations reproduced their own expertise.

AI has automated the codified, procedural parts of those roles. In doing so, it has also removed the experiential pathway through which tacit standards were built. Most organisations have not yet noticed the compounding consequence: when today's senior experts retire, there will be no one behind them who has developed the same internalised standard and therefore no one capable of directing AI intelligently in their absence.

The Return

Palantir noticed. Their Meritocracy Fellowship is a direct response to the recognition that the university system, built on the industrial model of prerequisites, examinations, and certified credentials, is no longer reliably producing people with the evaluative, contextual, tacit knowledge that consequential work requires.

They are not alone in noticing.

Professional bodies across medicine, law, and teaching are grappling with the same question: why do some certified practitioners consistently demonstrate excellent judgement while others, with identical credentials, do not? The answer, which the assessment research has been converging on for decades, is that the credential (eg the examination, the rubric, the competency framework) measures the specifiable surface of expertise, not necessarily the tacit depth of it.

The research on evaluative judgement in higher education, led by David Boud, Margaret Bearman, Phillip Dawson, and Joanna Tai at Deakin University, has made this increasingly explicit. In a 2024 paper responding directly to the rise of generative AI, they argued that developing evaluative capability and the ability to judge the quality of work, including AI-generated work, is "a uniquely urgent educational and professional challenge." It is, in their formulation, "a uniquely human capability at a time of technological acceleration."

They are describing Guild Knowledge. And they are saying the same thing the guilds understood five centuries ago: you cannot develop it by specifying it. You develop it by doing what guild apprentices always did - making, explaining, and calibrating evaluative judgements, repeatedly, in the company of those who already hold the standard.

What Comes Next

The industrial age was not wrong to want rigour, consistency, and accountability in how we develop and assess expertise. Those are genuine goods. It was wrong to believe that rubrics and mark schemes could reliably deliver them for the kinds of knowledge that matter most.

We are living through the correction. AI has made the category error visible. The workforce data is showing the consequences of the pipeline problem. And the most forward-looking organisations, from Palantir to the researchers at Deakin to the professional bodies quietly rethinking their CPD frameworks, are arriving at the same place from different directions.

The guild was not a primitive precursor to modern knowledge management. It was a sophisticated solution to a hard problem that the industrial model never actually solved, only deferred.

We are remembering that now. And this time, for the first time, we have the technology to do it at scale.