At half past ten in the evening, the office lights are still on.
A lawyer is reviewing a third contract. The first two pages look fine. Page seven introduces a limitation of liability. On page eleven, a breach clause is ambiguous. Meanwhile, the client asks in a message, “Is this clause risky?”
Perhaps only a dozen passages need significant professional judgment. Finding them still requires reading the entire document.
This is familiar in law firms and corporate legal departments. Contract review, legal research, case organization, first drafts, and recurring client questions all require expertise. They also include considerable repetitive, structured work.
That work is not easy to hand to ordinary automation, because legal work rarely follows a simple “If A, then B.”
It requires understanding language, context, relationships between clauses, and the points at which a conclusion should not be made.
That is where agents may help: by taking on some of the reading, searching, and organization that must happen before a lawyer makes a judgment.
Where should a lawyer spend time when reviewing a contract?
Imagine corporate counsel reviewing three procurement contracts a day.
The traditional process involves checking payment terms, breach provisions, intellectual property, confidentiality, dispute resolution, and liability caps, then documenting concerns.
The valuable work is judgment. Is the liability limitation reasonable? Does the indemnity exceed what the company can accept? Does an obligation conflict with how the business actually operates?
Locating clauses, comparing them with a standard template, and identifying missing sections are suitable candidates for an initial AI-assisted pass.
In ZGI, a company can place its contract templates, review rules, and standard clauses in a knowledge base, then use a workflow to screen an uploaded contract. The model identifies the contract type and relevant clauses. The knowledge base supplies company standards. The workflow produces a structured account of passages to review, quoted source text, and items requiring confirmation.
High-risk judgments remain with the lawyer.
This is different from a contract summary. A summary says what the contract contains. An agent participating in the process should help show which passages deserve attention first.
That may be a more practical role for legal AI.
Conceptual illustration: agents assist with retrieval, comparison, and drafting; lawyers retain professional judgment.
The risk in legal knowledge retrieval is an answer that only looks real
Hallucination is unavoidable as a concern when using language models in legal work.
A model may confidently describe a nonexistent case or mix an outdated rule with a newer version.
A legal knowledge base therefore requires more than uploading PDFs. Sources and boundaries matter.
An internal legal knowledge agent in ZGI can be instructed to prioritize the organization's knowledge base, legislation materials, and case collection, and to cite its sources. When the material does not support an answer, it should say so rather than continue generating a plausible-sounding response.
Time-sensitive legal questions still require lawyers to verify current law, judicial interpretations, and authoritative databases.
We see AI as a first-pass assistant for legal work, not the issuer of a final legal opinion.
It can find, organize, and compare material. A person remains responsible for deciding whether that material supports the final advice.
Many recurring client questions are highly repetitive
An employment law practice may repeatedly be asked about ending employment during probation, calculating severance, identifying overtime pay, or responding to a refused reassignment.
For an experienced lawyer, the difficulty may lie less in the question than in answering it twenty times a day. Receptionists, assistants, and account managers keep interrupting for confirmation of routine information.
An internal knowledge agent can help by drawing on legislation, practice guides, FAQs, standard replies, and training material.
For straightforward legal information, it can retrieve sources and prepare an explanation. Questions involving litigation strategy, responsibility, or insufficient facts should clearly be referred to a lawyer.
This lets lawyers spend less time repeating explanations and more time on questions that cannot be reduced to a template.
Case research should not always begin from scratch
Legal research consumes time partly because many retrieved materials are eventually discarded. A search may produce dozens of cases, with only a few useful ones.
When a team has lawfully obtained case material that it is permitted to use for internal retrieval, knowledge search and models can assist with semantic retrieval and preliminary organization by cause of action, disputed issue, court level, and reasoning.
Lawyers then read the original sources.
AI is well suited to narrowing the reading set. That is different from telling a lawyer how to argue a case based on a retrieved judgment.
It can reduce unproductive reading without assuming responsibility for legal judgment.
Why source access and self-hosting matter
In legal work, the data itself can create risk.
Contracts contain customer names, prices, and commercial arrangements. Case files may contain personal information, trade secrets, and litigation strategy.
The deployment question changes when an agent moves from writing ordinary copy to processing contracts and case materials.
Where are the files stored? Who can access them? Which model is used? Are execution records retained? Can the system run inside the organization's environment?
This is an important reason ZGI emphasizes available source code and self-hosting.
As an internal Agent Runtime, ZGI can connect an organization's models, knowledge bases, databases, and business systems. Law firms and legal departments with specific security requirements can configure deployment and access boundaries around their own standards.
The value is not simply saving a software fee. It is having greater ability to understand, deploy, and extend a system that participates in sensitive work.
Skills can preserve a firm's own methods
A specialist law firm's most valuable asset may be its accumulated methods rather than a general legal model.
Which fields should be checked first in an employment contract? Which terms in a financing agreement always need business-team confirmation? Which ten questions should be asked when taking on a particular type of case?
This knowledge often lives in experienced lawyers' heads, Word documents, Notion pages, and internal training.
Clear, reusable parts can be organized as Skills and become lasting capabilities.
An employment contract screening Skill can define the review order, fields of concern, and output format. A case organization Skill can specify how to extract a timeline, disputed issues, and evidentiary gaps. A client interview Skill can guide fact collection according to the type of matter.
Models will change. A firm's methods should not disappear with them.
Legal agents should not aim for full autonomy
Whether to accept a liability provision may be a commercial decision as well as a legal one. Whether to settle a case cannot be determined solely by a model's estimate of success.
We therefore do not recommend systems that automatically issue legal opinions.
A workflow should explicitly distinguish tasks that can proceed automatically from those that must pause for lawyer confirmation.
Initial screening can be automated, but final advice should not be sent to a client without review. Cases can be organized automatically, but their originals must be checked before citation. Documents can be drafted, but a lawyer must review them before submission.
Stronger agents make these boundaries more important.
Keep professional judgment with people
A lawyer's scarcest contribution is not knowing how to write a standard contract. It is understanding complex facts, assessing risk, and accepting responsibility for a judgment.
If AI handles preliminary retrieval, comparison, organization, classification, and drafting, it can leave more time for that work.
Legal teams need an assistant that knows which tasks it can start and where it must stop and wait for a lawyer.
ZGI aims to provide an open, self-hostable runtime in which firms and legal teams can organize their knowledge, Skills, and processes.
AI reduces repetitive work. Final professional judgment remains with people.
Top comments (0)