Rebuilding law firm knowledge management with Google Drive and Codex
Hello, this is Legal Agent.
Legal Agent works as a law firm built for the generative AI era, and how a firm stores its data has a direct effect on whether that AI can actually be used. A firm holds a great deal of information, contracts, counterparty revisions, Slack threads, books and guidelines, and in practice this is rarely gathered in one place: some sits in email, some in Slack, the latest contract version is in Google Drive, a template sits in someone's local folder, and a client's judgment standard exists only in the responsible lawyer's head.
People alone can still work this way, remembering things or asking whoever knows. Bringing AI genuinely into legal work is harder, because AI can only read what it can reach, with no automatic way to know a background that exists only in someone's memory or a precedent sitting only in an old matter folder. Building an AI Native law firm takes more than using a strong model.
The first requirement is centralizing information, which means more than putting files in Google Drive: it means organizing what a firm actually uses, by matter, client and contract type, in a form AI can read. Knowledge management, in my view, means building a state where a person and AI can later go back and look at the same context, and once that is in place, how legal work gets done changes considerably.
Scattered information keeps AI from doing real work
Reading the same contract can lead to different conclusions depending on the background: whether a services agreement is a pilot for a new business line or a renewal, the balance of power with the counterparty, and any past friction with them. A major client where closing quickly matters may call for negotiating a few important points rather than returning many small edits; heavy dependence on a counterparty's performance may call for scrutinizing liability limits and information handling closely. None of that comes from the contract text alone; it requires the matter's background, the client's business, past positions, and the exchange with the requester, so handing legal work to AI means giving it that surrounding context too.
With an ordinary AI chat tool, a person has to attach that context by hand every time, a real burden. If using AI still requires gathering and explaining material first, a substantial amount of work has already happened before AI does anything. Folder design is what changes that.
What it means to mount Google Drive locally
Legal Agent uses Google Drive, but not simply through a browser. We use the Google Drive desktop app to mount My Drive and shared drives as local folders, so Drive files look, from the computer's point of view, like ordinary local files. That matters because Codex can then treat Drive material as a local folder rather than something a person must select and hand over, which feels far more natural in practice than a browser-based connector search.
With Drive mounted locally, Codex can look across folders on its own: find a similar past matter, a template close to a contract type, or past exchanges with a counterparty. Instead of a person attaching files, Codex goes to the folder itself and can surface candidates before the person even starts looking. Every firm runs into a similar matter with a forgotten folder name or an unclear final version; combining a locally mounted Drive with Codex hands much of that search to AI, while the lawyer still decides which file and judgment to adopt.
The matter folder as the starting point
Legal Agent puts a lot of weight on matter folders. Knowledge management can bring to mind a polished database, and that has its place, but designing the matter folder properly comes first, because work happens matter by matter. When a folder holds material received from the client, files in progress, counterparty revisions, Slack and email exchanges, and reference past matters, it becomes far easier to use.
Putting the exchange itself into the folder makes an especially large difference. What was said when a request came in, this clause cannot move, we want a long-term relationship here, speed matters most this time, can matter more than the contract text, yet none of it is written in the contract. If that stays scattered across Slack and email, anyone besides the original handler loses the context later, and AI hits the same wall: it can only read the contract body and cannot understand why the request was made.
Putting Slack and email into the matter folder
Legal Agent saves client exchanges from Slack and email into the matter folder itself. Codex connects to Slack and Gmail and can run scripts automatically, so on a set schedule it pulls that content and files it into the right matter folder on Drive. This is not unlimited retention: scope, what gets saved, and confidentiality handling are decided in advance, and because client exchanges are involved, some parts need a person to check while others can be automated.
When a client sends a review request over Slack, the message and attachments go into the folder's exchange record, and additional background by email is filed the same way, so Codex can be asked to read the folder and summarize the request and any conditions. This reduces dependence on human memory, which matters when the person in charge changes or a paralegal joins partway through: with the exchange already saved, the handoff burden drops considerably.
Organizing books, statutes and guidelines
Reference material matters just as much as the matter folder. Simply saved as PDFs, a firm's books, statutes and practice guides can be hard for AI to use well; making them usable takes OCR, converting to text, and organizing by contract type or area of law. Legal Agent has been building a structured knowledge folder of statutes, official guidelines, in-house research memos and checklists by contract type, so reviewing a privacy policy means reading the Act on the Protection of Personal Information, the relevant guidelines and past memos in the same working environment, and AI can check the matter's own material against the reference material directly rather than offering only general commentary.
Skill folders turn knowledge into something that runs
Knowledge left as a memo tends not to get used. An NDA review checklist is useful, but someone still has to open it, apply it to the contract, and write comments into Word each time. Turning knowledge into a skill changes that: Legal Agent has built skills by task type, contract review, Word editing, legal due diligence, that set out which material to read, in what order, which checklist to use, what format comments should take, and how to treat track changes. What to check differs between the disclosing and receiving side of an NDA, or the client and vendor side of a services agreement, so Codex told to review a contract from the vendor's side works from that perspective rather than defaulting to generic commentary.
Knowledge management becomes how the work itself is designed
Knowledge management is becoming, for a law firm, the design of the work itself, not only something that serves people searching later for an old memo. With the matter folder in order, AI can read the matter; with exchanges saved, it can read the request's context; with templates in order, it can find the right base file; with statutes and guidelines converted to text, it can raise issues grounded in the source material. A firm with organized knowledge management brings AI into real work far more easily; a firm where information stays scattered gets much less out of AI regardless of the model, since AI ends up answering from fragments each time.
This also changes what outside counsel can offer. Without visibility into a company's business, past decisions and counterparty relationship, an outside lawyer's advice tends to stay a one-off answer that in-house legal has to re-read and re-apply to its own risk tolerance. With the matter folder, templates, client-specific standards and skills already centralized, and Codex finding and comparing that background, outside counsel can give advice closer to the resolution an in-house team would reach.
Closing
Legal Agent combines generative AI with lawyers experienced in corporate legal work to support contract review, fundraising and M&A. If you are dealing with a legal question, please feel free to reach out.