Codex for Legal Work: Differences from Claude Code and Use in Word Practice
Hello, this is Legal Agent.
Depending on the matter, Legal Agent delivers most work within one business day, sometimes the same day. While we utilize multiple tools to maintain efficiency, Codex has become particularly well suited to our daily legal practice. This article outlines how our firm applies it in day-to-day work.
Practical role and comparison with Claude Code
Codex is widely recognized as a tool for software engineering, commonly associated with writing code, fixing bugs, and running automated tests. While technical development remains its most familiar application, Codex also functions effectively as a core operational tool for law firm practice.
Claude Code has gained significant attention for business use, and its capabilities are impressive. In our direct practice, however, Codex currently provides a better fit for three distinct operational reasons.
First, the operational interface is straightforward. In our early experience with Claude, selecting among Chat, Cowork, and Code felt less immediate from a non-engineering perspective. That reflects a personal impression from that period, and product interfaces naturally evolve over time. With Codex, varied tasks such as reading local matter folders, editing Word documents, and performing legal research can be initiated from a single workspace. That unified access is practical for a legal practice where most team members are not software developers.
Second, GPT-5.5, the specific model used during the practice described here, showed strong capability in Japanese legal drafting. Claude Opus 4.7 is also an outstanding model; nevertheless, in our specific workflow, GPT-5.5 produced nuanced analysis and natural Japanese phrasing compared with earlier GPT generations, while Fast mode maintained an operating tempo comparable to Claude. This observation reflects personal, historical experience with these specific model versions rather than an assertion about current latest releases or universal product superiority.
Third, in our working experience during that period, applicable usage limits accommodated extended stretches of continuous drafting. Available limits inevitably vary with subscription tiers, service periods, and overall workloads, so this does not constitute a categorical claim that one platform consistently offers greater capacity.
Direct Word integration through custom MCP
Generative AI workflows typically operate through chat windows: a user pastes a clause, submits a query, reviews the output, and switches back to Word to implement revisions. That process is helpful but structurally incomplete. Receiving an answer does not conclude the assignment. An attorney must still determine which revisions to accept, record edits cleanly in track changes, and distinguish external counterparty comments from internal risk notes, all while accounting for commercial leverage and ongoing counterparty relationships. Codex proves valuable because it reaches beyond textual responses: it accesses local directories, identifies relevant files, creates working copies, edits and compares documents, and inspects textual variances throughout the process.
Scoped search across matter folders
Under earlier workflows, practitioners had to select individual files and upload them manually. Although seemingly minor, this repetitive step imposes additional work: an attorney must locate directories, recall relevant prior matters, and verify current revisions before an AI model ever sees the document. Codex removes much of that preliminary file retrieval. When instructed to locate a prior agreement with a specific counterparty, identify a comparable past transaction, or duplicate a template as today's working copy, Codex searches authorized directories within the defined scope of the assignment. This automates the upfront search across past agreements, standard templates, and previous diligence files.
In-place Word editing and revision tracking
To address document editing directly, we developed a custom Model Context Protocol (MCP) tool that allows Codex to operate Word while the document remains open. Legal contract review demands meticulous handling: changes must be minimal and exact, revision history must remain intact, and clause numbering or table formatting must not shift. Standard Codex configurations struggled with these constraints, as default operations tended to disrupt Word document structures. Current Word operations require our configured custom MCP rather than default out-of-the-box functionality, and this tool does not provide guaranteed perfect redlines. Instead, our MCP enables Codex to interact directly with Word, managing body text, tracked revisions, and marginal comments together within the file the attorney is actively reviewing.
This integration changed our review workflow. Because Codex modifies the open Word document directly instead of merely proposing text in a chat prompt, the attorney can observe changes in real time. The attorney can issue immediate adjustments, such as softening aggressive comment phrasing, and handle multiple related files concurrently. The tool ceases to operate like a standalone chatbot and functions as an integrated drafting environment that interacts with accessible directories while managing text, track changes, and comments simultaneously.
Automated comparison versions
Corporate legal practice regularly requires redline comparisons generated through Word's compare feature. Generating these manually involves selecting folders and picking file versions by hand each time. While manageable for a single agreement, running comparisons across ten documents becomes tedious. We implemented an additional MCP that allows Codex to determine the appropriate file versions and generate redline comparisons automatically upon request, saving substantial administrative time.
Practical legal applications
Contract review workflows
In a conventional review process, an attorney downloads a contract from email, searches for comparable prior matters, checks template folders, and evaluates prior comments alongside counterparty background before opening Word. The lawyer must then draft revisions, insert comments, and protect document formatting, knowing that a single misaligned clause number requires re-checking cross-references throughout the agreement. With Codex, a single structured instruction, such as reviewing a services agreement from the contractor's standpoint using past client correspondence, the firm's standard template, and the applicable review checklist, allows the system to search the folder, examine the contract, evaluate discrepancies, and assemble a review framework with tracked edits and initial confirmation items. The attorney still examines every change at the conclusion, but spends less time locating files or comparing clauses, directing attention toward business context and negotiation posture.
Legal due diligence and consultations
Due diligence requires reviewing extensive documentation, including corporate registries, shareholder registers, and investment agreements. Traditional efficiency tools often require uploading files individually, which proves burdensome in active data rooms subject to ongoing additions. Scoped directory integration significantly reduces the need to attach documents manually.
At the same time, practitioners must remain mindful of data handling. Reading local files can still transmit relevant content to the model provider for processing; avoiding manual uploads does not eliminate external data transmission. A law firm must carefully verify permitted data categories, vendor service terms, and model training parameters. Teams must also confirm data retention rules and client confidentiality obligations. Once configured with proper safeguards, Codex can inspect folders containing disclosed materials, compile document inventories, flag missing items, analyze core provisions, and draft preliminary question lists and report sections. Legal Agent also created an MCP for Excel, enabling teams to manage follow-up due diligence question lists through Codex.
Similar efficiencies apply to legal consultations. When a business team inquires whether a marketing campaign or proposed contract terms are viable, an attorney must review past advice and verify whether similar matters were escalated to management. Directing Codex to retrieve comparable internal consultations and assemble an initial response improves both turnaround and consistency. Furthermore, Legal Agent has converted in-house reference treatises, statutory texts, and regulatory guidelines into searchable digital formats, strictly within lawful copyright permissions. Conducting research directly within Codex avoids the fragmented workflow of consulting isolated document-search tools.
Operational shift and the role of attorney judgment
These capabilities do not render attorneys unnecessary. In practice, the more extensively Codex is deployed, the more critical attorney judgment becomes. AI can retrieve precedent files, align templates, and draft suggested comments; however, counsel must assess legal risks, advise on negotiation boundaries, and identify issues for escalation. The client retains responsibility for its business decisions. Codex serves not as a substitute for final professional decisions, but as an operational assistant that prepares materials so counsel can focus on substantive judgment.
In traditional practice, junior attorneys search for precedent and prepare initial drafts, which senior attorneys subsequently review, providing genuine training value. Yet where matters require repetitive manual searching and mechanical document comparison, that workflow can evolve. Codex can locate precedent materials, propose candidate revisions, and prepare working drafts, while counsel assesses whether candidate files are suitable and establishes strategic direction. This model does not merely add headcount; it establishes an organizational structure designed around AI from the outset, reflecting our approach to an AI-native law firm.
At Legal Agent, attorneys and AI agents collaborate across contract review, ongoing legal consultations, and corporate transactions, with the objective of maintaining an AI-native law firm that provides tangible value to client operations.