Beyond the NYC Bar’s AI Framework: What Comes Next for AI and Transactional Lawyers

What the NYC Bar’s new AI framework proposes

The paper argues that AI can assist legal work in emerging companies and venture capital, but cannot substitute for professional legal judgment. It maps AI use onto the existing ABA Model Rules of Professional Conduct, and it applies those duties to specific transactional document types.

At the center of the framework are five factors for deciding whether AI should draft or review a given document:

• Standardization. Is the document based on a standard model or form?
• Complexity. How complex is the substantive content?
• Tailoring. Is the document bespoke or interchangeable across matters?
• Experience. Does the firm have a repeatable history with the document type?
• Level of negotiation. How much back-and-forth is expected before the document is finalized?

The paper then applies the framework to specific documents in the venture capital and emerging companies space, rating each as high, moderate, or low for AI suitability. Basic C-corporation formation documents, board minutes, and non-disclosure agreements come out as highly suitable for AI-generation. Employment offer letters come out moderate. LLC formation documents (including operating agreements), shareholder agreements, and merger and acquisition agreements come out low.

Where the NYC Bar’s framework works

The five factors are a good rubric. They accurately capture the variables that are helpful in determining whether a document is a good candidate for automation.

Standardization is the right first question. Documents that follow a standard form are more suitable for AI drafting than documents that don’t. The paper’s rating of basic C-corp formation documents as highly suitable is straightforward. Articles of incorporation with the Secretary of State are close to a UCC filing: the same document over and over, with the names changing. The primary risk driver is clerical accuracy and jurisdictional rules, and AI grounded in the right jurisdictional context handles both well.

The paper treats the risk drivers document by document. For each document type, the paper identifies what specifically could go wrong. Clerical errors, missed compliance requirements, mistaken tax treatment, or over-simplification of a negotiated point. This is more useful than a blanket judgment about whether AI is safe. It gives lawyers something concrete to check against.

The framework anchors to existing ethical duties. The paper doesn’t invent new rules. It applies existing Model Rules of Professional Conduct—including Comment 8 to Rule 1.1, which covers the duty to keep abreast of the benefits and risks of relevant technology—to AI use. That’s the right foundation. The obligations already exist. What’s needed is guidance on how they apply to AI in transactional legal workflows.

Where the framework is too narrow

The paper treats AI primarily as a document generator. That’s a familiar frame, and it’s one most bar guidance has used so far. But it’s a subset of what AI does in a modern legal workflow, and the ratings in the paper’s document-by-document table reflect that narrower view. Three of the ratings deserve reconsideration.

LLC formation documents can be rated higher. Consider LLC formation documents, rated low for AI suitability. The paper suggests AI is fine for a starting point and some clause suggestions, but not much beyond that. That rating assumes AI is working in a vacuum, with no client context.

That’s not how legal AI tools like Clio Work actually function. LLC operating agreements have real variation, but the variation is rules-based: manager-managed or member-managed, tax treatment (partnership, S-corp, C-corp), voting rights, distribution rules, buyout terms. Law firms have been automating these choices through document assembly for years. AI is document assembly with better context awareness. If a system already knows the client’s tax goals, the intended management structure, and the equity split, drafting the agreement becomes a straightforward extension of the intake process.

Employment offer letters carry more context than the moderate rating suggests. The paper rates these moderate for AI suitability, with the primary risk driver as compliance with employment laws. That’s understated. Employment offer letters carry significant context, such as prior inventions clauses, cross-jurisdictional hiring, non-compete enforceability, whether the hire is a leadership position or a front-desk role, what stock and benefits are on the table. A golden parachute for an incoming executive is a very different document from an offer letter for an entry-level sales rep. Without the client context to distinguish, AI drafts of employment letters can be a bigger risk than AI drafts of LLC operating agreements. If anything, the rankings for these two document types should be flipped: LLC formation rated higher (given proper client context) and employment offer letters rated lower.

M&A drafting deserves more credit once the negotiation is settled. The paper rates M&A agreements low because the transaction structure and negotiated risk allocation require human judgment. That’s true of the negotiation, but it’s less true of the drafting once the negotiation is settled. If the transaction structure has been worked out in a letter of intent, and the risk allocation has been decided in principle, the actual drafting is the mechanical part. That’s exactly the work AI is good at. Reserving M&A for humans conflates the negotiation stage with the drafting stage.

The workflow question the framework leaves open

The five-factor framework judges AI one document at a time. But AI in legal practice rarely works that way. A lawyer’s matter builds from intake through drafting through review, and AI is most useful when it can move with that context. A better framework would look at where AI fits in the whole matter, not just where it fits in one document.

A workflow-based system stores that context and gives the AI access to it. So the LLC agreement isn’t being drafted from scratch by a model that knows nothing about the client. It’s being drafted from the client’s tax preferences, management structure, planned capitalization, and negotiated deal points, all already in the record. Drafting becomes a natural extension of the work the lawyer has already done.

The paper’s framework doesn’t account for this because it treats each document as an isolated task. Intake, case management, document generation, and review are connected steps in the same matter, and a workflow-aware AI system is aware of all of them. Clio Work is one implementation of that principle. It ties together intake, matter context, and AI drafting so that a document is generated with the full record of the matter instead of from a blank prompt.

This is why the debate over which documents are “AI-suitable” and which aren’t can land in the wrong place. The suitability of AI drafting depends less on the type of document than on the context the AI has access to. Give an AI the client’s tax planning conversation, and drafting an LLC operating agreement is straightforward. Give it a blank prompt and a form template, and even a C-corp filing can go wrong.

The call for national AI rules meets the state-by-state reality

The second part of the paper argues for a national framework governing AI use in legal practice. That’s the harder problem, and the paper is right that the current state of guidance is fragmented. State bars, courts, and individual judges have each issued their own guidance, standing orders, or ethics opinions, and the pieces don’t line up. That fragmentation makes it hard for lawyers to know what’s expected.

The structural reality is that the United States doesn’t have a national legal regulator. The ABA writes Model Rules. States decide whether to adopt them. There is no federal bar to enforce a uniform AI standard, and there’s no realistic path to one. Any national framework would need to work through state bars, state legislatures, or state supreme courts, which is the same path the current guidance is already taking. The paper’s regulatory instinct is right, but there’s a more actionable target within reach.

Better guidance on existing duties is the first step. Federal Rule of Civil Procedure 11 (for court filings) and the duty of candor to the tribunal already require lawyers to verify what they file. What’s missing is guidance on how those duties translate into an AI workflow: What does verification look like when a first draft is AI-generated? What does lawyer supervision require when the work is being done by a tool rather than a junior lawyer? The paper takes some steps in this direction, but state bars, courts, and organizations like the National Center for State Courts or the Uniform Law Commission are the right venues for making the guidance operational.

Record-keeping standards are the practical fix. Bar associations already regulate lawyers’ record-keeping in specific areas (trust accounting is the obvious example). A similar approach would work for AI. Rather than regulating the tools, regulate what lawyers need to keep as a record of AI use: the prompts, the intermediate outputs, the final document, and which templates or firm-approved workflows were used. Clio Work, for instance, keeps a full audit trail of each AI action taken on a matter. That kind of record makes AI use defensible under scrutiny, whether the scrutiny comes from a bar complaint, a malpractice claim, or a court’s standing order. Standards for that record could come from state bars without requiring a national framework.

Celebrate successful AI use as much as caution about misuse. The paper spends most of its regulatory analysis on what could go wrong. One paragraph in the report is titled “AI as a Driver of Access and Innovation,” and it’s immediately followed by a return to risk. The Model Rules already frame this as a two-sided obligation. Comment 8 to Rule 1.1 asks lawyers to understand the risks and the benefits of technology. Bar guidance is strongest when it does both. Bar associations that only publish warnings without also celebrating successful use may be chilling innovation more than they realize.

What bar associations should do next

Bar associations sit at the closest point to where AI actually gets used, and they have the authority to move without waiting for the ABA or federal action. Five recommendations follow from the analysis above.

• Publish workflow-aware guidance, not just document-type guidance. The suitability of AI depends on the surrounding workflow. Guidance should reflect that.
• Set record-keeping standards for AI use. Prompts, intermediate outputs, and template use should be preserved as part of the matter record. That single change makes AI use defensible and mitigates most of the misuse risk that regulators worry about.
• Celebrate successful AI use. Bar leadership that promotes firms doing something new and useful with AI gives lawyers permission to explore. Silence, or a stream of warnings without counterexamples, chills the profession.
• Designate an innovation officer or AI officer at the state level. Court reporter shortages, legal deserts, and access-to-justice gaps are real problems where AI could help. Someone at the bar leadership level needs to own the affirmative case.
• Use existing rules as the foundation. The ABA Model Rules already cover most of what lawyers need to know. What’s missing is application to specific AI workflows, and that can come through state bar opinions and practice management advisor guidance rather than a new national framework.

Moving AI guidance forward

The NYC Bar’s paper opens a conversation the profession has needed for a while. The five-factor framework is a useful rubric, and the committee is right to map AI use onto existing Model Rules of Professional Conduct.

The framework needs more in two directions. The first is workflow. AI does its most useful work when it can see the whole matter, not one document at a time, and the ratings for LLC formation, employment offer letters, and M&A drafting shift once client context is in the picture. The second is regulation. A national AI framework for lawyers likely isn’t coming through federal channels, but state bars, courts, and practice management advisors can do the work that a national framework was meant to do.

Where the paper ends, the state-level work begins. If bar associations set the standard, technology providers will build to it.