Why Legal Work Starts with Documents: Files, Versions and Evidence

Strip away the courtrooms, the meetings, the negotiations, the calls with clients, and what remains is documents.

A lawyer's day is shaped by texts. Reading them. Comparing them. Annotating them. Drafting them. Sending them. Filing them. Retrieving them six months later when no one remembers exactly which version was signed.

This is not a complaint. It is a description. And it is the single most underestimated fact about legal practice.

If we accept that legal work is, at its core, document work, then a serious question follows. What kind of system actually supports that work? Not a chatbot. Not a search engine. Not a folder on a shared drive. Something else.

This article makes the case for taking documents seriously, and shows why doing so reshapes the entire conversation about legal technology and legal AI.

What actually circulates in a legal matter

Pick any matter, in any practice area, at any decent size, and look at what circulates around it. The list is remarkably consistent.

In a litigation file, you find the original contract, the correspondence that preceded the dispute, the mise en demeure, the technical reports or invoices supporting the claim, drafts of conclusions, the opposing party's filings, judgments and procedural decisions, calendar items for hearings, notes from client meetings, expert reports, and at the end, the final ruling.

In a corporate matter, you find term sheets, drafts of the SPA, due diligence reports, board resolutions, minutes of general meetings, regulatory filings, opinions, comfort letters, signed signature pages, and a closing memorandum.

In an in-house request, you find the original ask (often a one-line email or a chat message), the contract or policy attached, the prior versions, the comparable internal precedents, the response from external counsel if any, the memo for management, and the decision.

Different matters, different practice areas. Same underlying reality. Documents move. People act on them. Decisions get made from them. Disputes get won or lost based on what they say and what they do not say.

Why legal documents are not ordinary files

This is the point where most software companies stop reading carefully, and it is where the misunderstanding begins.

A legal document is not a Word file with words in it. It is a structured artefact with properties that ordinary documents do not have.

Versioning matters legally. The contract signed last Tuesday is not the same legal object as the draft from the Tuesday before. The difference between the two is sometimes a comma, and that comma sometimes decides a case.

Annexes and references matter substantively. A contract that references "Annex 3" without Annex 3 is incomplete in a way that no spell-checker will catch. A pleading that cites a decision without the decision attached is, in many courts, a problem.

Signatures change everything. An unsigned draft and a signed final are, legally, different objects. The first is a position. The second is a commitment.

Language matters. A Moroccan contract whose French body says one thing and whose Arabic translation says another is not a translation problem. It is a litigation waiting to happen.

Probative force is unequal. A scanned PDF is not the same as a signed original. An email is not the same as a registered letter. A WhatsApp screenshot is not the same as a procès-verbal.

These distinctions are obvious to any practising lawyer. They are invisible to most file systems. That gap, between what the lawyer knows and what the software understands, is the source of an enormous amount of daily friction.

The hidden cost of fragmentation

Now look at where these documents actually live in most legal teams today, in Morocco and elsewhere.

The original contract is in an email attachment from three years ago. The signed version is on a shared drive, in a folder named after the project, with no clear convention. The drafts are in three different versions on three different machines. The correspondence is in two inboxes, one Outlook, one WhatsApp. The expert reports were sent by Wetransfer and have probably expired. The judgment is in a paper file in a cabinet, scanned at low resolution into a PDF that no one has indexed.

This is not a critique of any individual team. It is the reality of how legal work has accumulated, year after year, in firms and in-house departments that never had the time to redesign their information backbone.

The cost is real, and it shows up in five places.

Time. Senior lawyers spend hours each week looking for documents they know exist.

Risk. Decisions get made from the wrong version. Or from a memory of the version. Or from a junior's quick summary that nobody verified.

Knowledge loss. When a senior associate leaves, half of their knowledge leaves with them, because it was implicit in the folder structure of their personal computer.

Onboarding pain. New collaborators take months to become productive, not because they lack legal skill, but because they cannot find anything.

AI failure. And, as a consequence of all of the above, any attempt to layer AI on top of this mess produces unreliable results. Because the AI is working on documents that are themselves disorganised, duplicated, mislabeled, and dispersed.

From the document to the structured matter

The shift that changes everything is conceptually simple, and operationally hard. It is the move from isolated documents to structured matters.

A matter is not a folder. A matter is a legal object with properties:

• a client and a counterparty,
• a practice area and a jurisdiction,
• a status (open, on hold, in litigation, closed),
• a team and an owner,
• a list of related parties,
• a list of associated documents, each with its own version history and role,
• a timeline of events (received, drafted, signed, filed, ruled),
• a list of tasks and deadlines,
• a list of approvals,
• a trail of who did what and when.

When a document lives inside a matter rather than next to a matter, several things change at once. You can ask better questions ("show me all signed contracts with this counterparty in the last three years"). You can compute things automatically ("the limitation period on this claim expires in nineteen days"). You can preserve institutional memory ("here is what the firm did the last time this exact issue came up"). And, importantly, you can deploy AI in a way that has a chance of being correct, because the AI knows where it is, who is involved, what the prior decisions were, and which documents are authoritative.

This is what is meant when serious legal AI platforms talk about working at the level of the matter, not the level of the prompt. Harvey, Legora, and the workflow-specific approaches behind Claude Legal Skills all share this premise. The unit of work is not the chat. The unit of work is the legal matter.

Why this matters for legal AI

If legal work is document work, and legal documents are structured artefacts inside structured matters, then a legal AI that ignores this reality is solving the wrong problem.

A chatbot that answers general questions about contract law is useful for a student. It is mostly useless for a Moroccan in-house counsel who needs to know whether this specific clause, in this specific contract, signed under these specific conditions, with this specific counterparty, in this specific regulatory environment, is consistent with the firm's prior practice on similar matters.

That last sentence is what legal AI actually has to handle. And it cannot handle it without the document layer.

This is why the next article in this Academy is about context engineering. Once documents are structured into matters, the question becomes how much of that structure the AI can safely see, retrieve, and use. Context engineering is the discipline of answering that question well.

In Morocco and francophone practice

Three Moroccan realities make the document-first view especially relevant.

Bilingual production. Moroccan legal teams routinely handle French and Arabic on the same matter. A contract in French with Arabic annexes. A procès-verbal in Arabic referencing a French-language report. A pleading in Arabic citing a contract written in French. The document layer must handle both languages natively, not as an afterthought.

Volume and dispersion. Even mid-sized cabinets in Casablanca and Rabat accumulate thousands of documents across hundreds of matters over a few years. The same is true for in-house teams in banks, insurers, telecom operators, and industrial groups. Without a structured matter layer, this volume becomes paralysing rather than valuable.

Procedural rigour. Moroccan litigation practice, especially before commercial and administrative jurisdictions, depends on procedural precision. Deadlines, filings, hearings, original-versus-copy distinctions, registered notifications. Treating documents as objects with status, timestamps, and version history is not a luxury. It is how the procedure actually works.

A Legal OS that takes documents seriously is, for these reasons, naturally aligned with how Moroccan legal teams already think. It does not impose a foreign discipline. It formalises a discipline that already exists, mostly in the heads of the senior partners.

Practical example

A Casablanca cabinet handles a commercial dispute before the Tribunal de commerce. The client is a Moroccan industrial group. The counterparty is an international supplier. The matter has been open for fourteen months.

In a fragmented setup, the team works like this. The original contract sits in an email from 2022. The amendment is on the partner's laptop. The correspondence is split between two inboxes. The technical reports were sent by the client through Wetransfer and have to be re-requested. The mise en demeure is in a Word file named MED_v3_final_FINAL.docx. The opposing party's last pleading is in a PDF the junior received yesterday and has not yet shared. The next hearing is in nine days.

The associate who joined the firm two months ago needs to draft conclusions. She spends one full day reconstructing the file before she can write a single sentence. The partner spends one hour answering her questions about which version is authoritative.

In a structured setup, the same matter lives as a single object. Contract v1 (original), Amendment v1 (signed), correspondence (organised chronologically), technical reports (uploaded by the client through a secure channel and linked to the matter), the mise en demeure (one version, marked as sent, with proof of delivery), the opposing party's filings (uploaded as received, tagged), the hearing (on the matter timeline, with the deadline computed automatically). The associate opens the matter and starts drafting in twenty minutes. The partner reviews instead of explaining.

The legal work has not changed. The infrastructure underneath it has.

This example is illustrative. Any legal output must be reviewed by a qualified Moroccan legal professional before use.

What this changes for Burhan

Burhan is built on the conviction that legal AI cannot be designed without first taking documents and matters seriously.

In practice, that means the foundational layer of Burhan is not the chat interface. It is the matter, the document, the parties, the timeline, the tasks, the approvals, and the audit trail. Everything else, including AI-assisted drafting, AI-assisted research, AI-assisted review, sits on top of that layer.

This ordering matters. A Legal OS that starts from the chat and tries to retrofit structure underneath produces unreliable AI. A Legal OS that starts from the structure and adds AI on top produces AI that knows where it is.

The lawyer remains the author. The Legal OS keeps the file in order so the lawyer can focus on the law.

Key points

• Legal work, viewed honestly, is document work. Almost everything else flows from this.
• Legal documents are not ordinary files. They have versions, annexes, signatures, languages, and probative force.
• Fragmentation across email, drives, chat apps, and paper has a real cost: time, risk, knowledge loss, and AI failure.
• The shift from isolated documents to structured matters changes what is possible, including what AI can reliably do.
• For Moroccan legal teams, bilingual production, document volume, and procedural rigour make the document-first view especially relevant.