Codifying the craft

Every office has someone everyone goes to with questions. They have been there for twenty years and know which document to ask for before taking on a matter, which clause causes trouble, which client needs a second phone call. Almost none of it is written down. When that person retires or leaves for a competitor, it all goes with them.

That know-how is what sets a seasoned business apart from one that has just opened its doors. It has always been passed on in the same way, from one mind to another, and always slowly. Bots change that: the firm's experience can be written into procedures that a machine applies to every assignment. I propose calling this codifying the craft, and I think it is one of the fundamental changes that artificial intelligence brings to business.

From mind to mind

First came the master and the apprentice. The craft was learned by watching and repeating for years. Guilds organised and passed on the knowledge of the craft through apprenticeship; that knowledge also travelled with itinerant journeymen.[1]

Then came schools, manuals and working methods. Repetitive tasks could be written down. Judgement stayed where it was. Ikujiro Nonaka described this by distinguishing between explicit knowledge, which fits in a manual, and tacit knowledge, which is acquired through practice and is hard to explain.[2]

Computing changed less than it seems. Programs helped people and automated whatever could be reduced to precise rules. The knowledge of the owner, the marketing director or the law firm partner cannot be reduced that way, and it continued to be passed on by word of mouth, matter by matter, to whoever worked alongside them.

Franchising is the most serious attempt to package the craft for sale. In European competition law, the know-how transferred under these agreements is defined as practical information, not patented, resulting from the supplier's experience and testing, which is secret, substantial and identified.[3] Even so, any franchisee knows that the manual does not teach everything and that the rest is learned by opening the shop every morning.

From mind to machine

A bot understands instructions written in everyday language. There is no need to translate judgement into precise rules; you just say it as you would to a trainee. Today's systems allow you to give them files of instructions and examples that they consult when that kind of work comes up.[4]

The implications go beyond those of a good manual:

• Speed. An apprentice takes years. A rule approved today is applied tomorrow.
• Consistency. It is applied to every assignment, even on a Friday afternoon.
• Scale. Ten assignments at once read the same manual.
• Copying. Duplicating what has been learned costs as much as copying a file. Using the model costs money and the manual needs maintaining; copying it costs practically nothing.

You could call it a digital brain, with one difference worth keeping in mind. The model does not change. What has been learned is in a file that the business can read, correct and take elsewhere.

Little by little

The word codifying rings a bell for a lawyer. Codification put into writing and brought order to a body of law scattered across practices, customs and separate statutes.

Something similar happens with a company's craft, on a small scale and without a date of enactment. In the previous article, a lawyer discovered that the sale and purchase agreement prepared by her bots did not protect a buyer who depended on a mortgage. She corrected the agreement and also left a rule in the manual: before drafting, ask about financing. Every human correction turned into a rule is another article in the firm's code.

Nobody sits down on a Monday to write down everything they know; it would not work. The code takes shape matter by matter, through real mistakes, and each rule is approved by a person responsible for it. That is why this cannot be put in place all at once. It has to be cultivated.

Why it is the most valuable asset

Models are rented. The newly opened law firm and the one that has been around for thirty years hire the same one, at the same price. When the tools are the same, the difference comes back to what each knows how to do. The veteran has thousands of matters to draw rules from. The newcomer has to make the mistakes first.

My impression is that many businesses already have artificial intelligence and do not quite know what to do with it. They ask it for isolated tasks and get generic answers, the same answers anyone would get. What is missing is architecture: an orderly way of bringing what the business and its people know into the machine.

Those people are the source. Experience that is not written down still belongs to the person who holds it in their head, and leaves with them. Experience that is codified stays with the firm. That changes the position of the person contributing it, and this needs saying: the partner or director who puts their judgement into the manual moves from doing the work to deciding how it is done, and their judgement reaches far more matters. A sensible business will recognise and pay for that, because without them the manual is empty.

A trade secret

An asset needs protecting, all the more so if it can be copied at almost no cost. In Spain, Law 1/2019 on Trade Secrets expressly includes organisational knowledge and protects it if it meets three conditions: it must be secret, have business value precisely because it is secret, and its holder must have taken reasonable steps to keep it that way.[5] The third is the one most often missing. A manual accessible to the entire workforce, copied to personal accounts or uploaded to a service without safeguards is unlikely to pass the test.

Here, reasonable steps means limiting who can read and amend the manual, recording changes and agreeing confidentiality obligations with employees and collaborators. The AI provider must be required not to use those files to train its models and to allow them to be exported in a readable format.[6][7]

The law has two further implications. A trade secret is transferable: it can be assigned or licensed, so the codified craft becomes part of the price when the business is sold. And protection does not extend to the experience and skills that a worker honestly acquires over the course of their career.[5] What the partner carries in their head can go with them. The file cannot.

What does not fit in the manual

Not everything can be written down. Judgement when faced with a case nobody has seen before still lies with people, and that judgement produces the next rules.

Rules age. The law changes, the bank's or land registry's practices change, and an unreviewed manual applies old mistakes with great discipline. Someone has to be tasked with maintaining it.

And the manual contains rules, not case files. The rule says “ask about the mortgage”; it does not say who the buyer was or how much they paid in advance. The craft belongs to the business. The client's data still belongs to the client, and in a law firm it is covered by professional secrecy.

The person everyone asks will still be needed. For centuries, what they knew could only pass to whoever sat beside them. Now it can stay with the firm, be applied to every assignment and grow with each correction. For a business with years of experience in its craft, that is the work worth starting: writing down what it knows, rule by rule, and keeping it safe.

Sources

Consulted: 13 November 2026. Numbering for this revision. Draft: citations [1], [2], [3], [5], [6] and [7] await checking against the original text.

[1] S. R. Epstein, “Craft Guilds, Apprenticeship, and Technological Change in Preindustrial Europe”, The Journal of Economic History, vol. 58, no. 3, September 1998, pp. 684–713. DOI: 10.1017/S0022050700021124.

[2] Ikujiro Nonaka, “The Knowledge-Creating Company”, Harvard Business Review, vol. 69, no. 6, November–December 1991, pp. 96–104; reprinted as HBR Classic, vol. 85, no. 7–8, July–August 2007, pp. 162–171, reference R0707N.

[3] Commission Regulation (EU) 2022/720 of 10 May 2022 on the application of Article 101(3) of the Treaty on the Functioning of the European Union to categories of vertical agreements and concerted practices, Article 1(1)(j). Spanish text in the BOE. For its application to franchises: European Commission, Guidelines on vertical restraints, 2022/C 248/01, paragraphs 85–87 and 165–168. EUR-Lex.

[4] Anthropic, Equipping agents for the real world with Agent Skills, 16 October 2025, updated 18 December 2025. Official source.

[5] Spanish Law 1/2019 of 20 February on Trade Secrets, BOE no. 45, 21-02-2019, Articles 1 and 4. Consolidated text. Definition and requirements of a trade secret, limitation concerning a worker's experience and skills, and transferability. Transposes Directive (EU) 2016/943. Whether a particular manual meets the requirements depends on the case.

[6] CCBE, Guide on the use of generative AI by lawyers, 2 October 2025, section 4.1, pp. 16–18, especially pp. 17–18. Professional guide. Risks of data reuse for training, confidentiality safeguards and examination of the provider's terms.

[7] CCBE, Guidelines on the use of cloud computing by Bars and lawyers, 27 February 2025, section II.6, pp. 10–12, especially p. 12. Contract terms and data recovery. Safeguards concerning the purposes for which data may be used and recovery in a readable format when the service ends.