AI Workflows · Workflow playbook · Updated June 2026

AI for the Solo and Small Law Firm: The Claude First Stack One Lawyer Can Run

Most legal AI coverage assumes a Harvey or CoCounsel budget and an IT department to stand it up. That is not your firm. You are one lawyer wearing every hat, and the question is not which six figure platform to buy. It is which stack you can run today, by yourself, without breaking a single bar duty. Here it is.

The short version: A solo or small law firm does not need an enterprise legal AI platform to get most of the leverage. A general purpose model like Claude, on an approved business or enterprise plan that does not train on your inputs, can run the four jobs that eat a solo's week: intake triage, research triage, first draft drafting, and client communications. The leverage comes from compressing the reading and the first draft, not from automating judgment. The catch that no vendor deck mentions is that a one person shop has no second reviewer and no IT department. You are the supervision, the verification, and the data security policy, all at once. So the stack is only as safe as your discipline: classify what can go into a model, verify every output against the source, and never put a privileged or client confidential matter into a consumer grade tool.

Key takeaways

You do not need a six figure platform. The Claude first stack a solo can run covers intake triage, research triage, drafting, and client comms on a general model you control, for a fraction of the cost and lock in of an enterprise legal suite.
The leverage is your hours back, honestly framed. AI compresses the reading and the first draft so the billable hour goes to judgment, not toil. We will not invent a hours saved number for you. We will show you how to do the math on your own work.
You are the supervision. A solo has no associate to check and no partner to escalate to. The bar's competence, confidentiality, and supervision duties still apply, and with no team to delegate to, every one of them lands on you.
Confidentiality with no IT department is a posture, not a product. Use only an approved deployment that contractually will not train on your inputs, classify before you paste, and when in doubt the matter stays out.

The professional's real problem

The legal AI conversation is written for the wrong reader. Open almost any piece on AI for law firms and it assumes a deployment committee, a pilot program, a budget line for a platform that costs more per year than a solo's entire overhead, and an IT team to integrate it with a document management system you do not have. That coverage is real, and for an AmLaw firm it is useful. It is also a world away from your desk.

Your reality is different in kind, not just in scale. You are the rainmaker, the intake coordinator, the legal researcher, the drafter, the proofreader, the billing clerk, and the person who answers the phone. There is no associate to hand the first read to and no partner to escalate the close call to. Every hour you spend on work that does not require a law license is an hour you cannot bill, cannot spend on the work that does, or cannot get back from your evening. The economics of a solo practice are brutally simple: your time is the entire inventory, and most of it leaks into tasks that are necessary but not lawyering.

That is exactly the gap a general purpose model fills, and it is why the enterprise framing misses you. You do not need a platform that promises to transform your firm. You need a fast, capable first reader and first drafter that you can switch on this afternoon, on a plan you can afford, that gives you back the hours the work was quietly stealing. The leverage is not in buying more software. It is in running a disciplined method on a tool you already control.

For a solo, AI is not a transformation project. It is the associate you could never afford and the research clerk you never hired, on the condition that you remain the lawyer who checks their work.

The solo Claude first stack

Here is the whole stack in one view: the four jobs a solo can hand to a general model today, what the AI actually does in each, and what stays entirely yours. The right hand column is the point. Nothing in it is delegable, because you are the only lawyer in the building.

The solo Claude first stack: task, what AI does, what you still own
Task	What AI does	What you still own
Intake triage	Turns a messy inquiry into a structured summary: issue, jurisdiction, timeline, parties, and the facts you would need to assess the matter. Drafts intake follow up questions.	The conflicts check, the engagement decision, and the legal assessment. AI never decides whether to take a client or whether a conflict exists.
Research triage	Frames the issue, suggests where to look, explains a doctrine in plain language, and produces a starting outline of arguments to pursue.	The actual authority. Every case, statute, and rule is read and confirmed in a real legal database. AI output is a lead, never a citation.
Drafting	Produces a first draft of a letter, a motion shell, a clause, or a client memo in your voice, fast, from your facts and your direction.	The legal substance, the strategy, and the final word. You rewrite, verify, and sign. The draft is a starting point, not a filing.
Client comms	Drafts status updates, explains a legal concept for a lay client, and turns your bullet points into a clear, warm email or letter.	Every word that goes to a client, and the confidentiality of what you put in. You edit for accuracy and tone, and you keep privileged facts out of the prompt.

Read the stack as a division of labor, not an automation. In every row, AI carries the reading and the first draft, and you carry the judgment, the verification, and the responsibility. That split is the whole method, and it is what keeps a one person firm both fast and defensible.

The leverage math, honestly

The reason this matters for a solo is arithmetic, and it is worth being precise rather than promotional about it. We are not going to hand you a measured statistic, because we do not have one for your practice and inventing one would be exactly the kind of slop this site exists to avoid. What we can do is show you the shape of the calculation so you can run it on your own work.

Think about a single recurring task. Suppose a particular client status letter, the kind you write a dozen times a month, takes you two hours from blank page to send: gathering the facts, drafting, polishing the tone, proofreading. Now suppose a model gets you a solid first draft from your bullet points in twenty minutes, and you spend forty minutes verifying, rewriting in your voice, and confirming nothing confidential leaked into the prompt. That is one hour instead of two on that task. This is an illustrative example, not a claimed result. Your real numbers depend on your work, your speed, and how disciplined your verification is.

The point of the illustration is the structure, not the figure. A solo's leverage does not come from one dramatic win. It comes from shaving the routine, repeated tasks, the intake summaries and status emails and first draft letters, and routing the recovered time into billable work or back into your life. Run the math on your own three most repeated tasks before you believe any vendor's headline number, including a number we might be tempted to give you. The honest version is the only version that survives contact with your actual calendar.

How to stand up the stack, step by step

Here is the rollout as a sequence a solo can actually execute, with no IT project and no committee. Do it in order. Each step builds the discipline the next one depends on.

Step 1: Pick an approved deployment and settle the data question first

Before you run a single client task, choose a plan whose terms contractually commit not to train on your inputs and meet your confidentiality obligations. Consumer and free tiers frequently do not. A business or enterprise plan of a major model usually does, but you read the terms yourself rather than assume. This is your one substitute for an IT department, so do it once, deliberately, and write down what you concluded. Our confidentiality guide for attorneys walks the exact questions to answer.

Step 2: Start with intake triage, the lowest risk win

Intake is the safest place to begin because the inputs are often a prospect's own description, and the output is a structure you check, not advice you send. Feed the model the inquiry and ask for a clean summary: issue, jurisdiction, timeline, parties, and the missing facts you need. Use it to draft your follow up questions. You still run the conflicts check and make the engagement call yourself, every time.

Step 3: Add research triage, never research itself

Use the model to frame an unfamiliar issue, explain a doctrine, and sketch the arguments worth pursuing. Treat all of it as orientation. Then go to a real legal research database and confirm every authority before it touches a brief. A general model can name a case that does not exist or misstate a holding with total confidence, so nothing it produces is a citation until you have read the source.

Step 4: Bring drafting in, as a first draft only

Hand the model your facts and your direction and ask for a first draft of the letter, memo, or motion shell. You are getting a fast starting point from a capable junior, not a finished product. Rewrite it in your voice, supply the legal substance and strategy, verify every factual and legal assertion, and sign only what you have made your own.

Step 5: Use it for client comms, with privilege in mind

Let the model turn your bullet points into a clear status update or translate a legal concept into plain language a client will understand. Edit every word for accuracy and tone before it sends. And keep privileged or sensitive client facts out of the prompt unless your deployment is approved for them. The goal is warmer, faster client communication, not a shortcut around confidentiality.

Honest usage notes

A few things become clear once a solo runs this on real matters rather than a demo.

Intake and client comms are where the stack pays off fastest and most safely. Structuring an inquiry and turning bullets into a clear client email are exactly the patient, low judgment tasks a model does well, and exactly the tasks that quietly drain a solo's day. Start there, build the verification habit on low stakes work, and the rest of the stack is easier to trust.

Research triage is the place to be most disciplined. The temptation is to let a fluent answer stand in for the authority, and that is the trap that puts lawyers in front of disciplinary boards. The model is a fine way to get oriented and a terrible substitute for opening the case. The line is simple: orientation yes, citation never, until you have read the source.

Drafting saves the most time on the most repeated documents. The first status letter the model drafts is unremarkable. The hundredth, once you have tuned how you brief it, is where the hours compound. For the broader stance a senior lawyer brings to all of this, our piece on the senior lawyer AI operating model goes deeper, and this stack is one room in the larger map of how law firms run on AI.

How this differs from running any solo business with AI

If you have read our general guide on running a solo practice with AI, you already know the universal one person playbook: automate the admin, draft the comms, and reclaim the hours. A law firm is a special case of that, and the differences are not cosmetic. Three things make it legal work rather than generic solo work.

First, intake is not just lead capture. It carries a conflicts check and a confidentiality obligation from the first conversation, and AI does not get to make either call. Second, the billable hour is the unit you are optimizing, which changes the math: the goal is not vague productivity but converting non billable toil into billable judgment. Third, and most important, you carry bar duties no consultant or freelancer does. Competence, confidentiality, and the duty to supervise the work apply whether your assistant is a human associate or a model, and in a solo shop there is no one else to carry them. That is why this page exists separately from the general one: the method is similar, the duties are not.

You are the supervision, the security, and the second reviewer

This is the part of the stack a solo cannot skip, because there is no one else to skip it for you. In a big firm these duties are spread across people. In your firm they are all yours.

No second reviewer means you verify everything

A larger firm runs AI output past an associate and a partner. You have neither. Under the bar's competence duty, captured in Model Rule 1.1 and its technology guidance, you must understand the benefits and risks of the tools you use, and that includes understanding that a model can be confidently wrong. Every legal assertion, every citation, every fact in a draft is verified by you against the source before it leaves your desk. An unverified AI output is a hypothesis, not work product.

No IT department means you are the data policy

The duty of confidentiality in Model Rule 1.6 does not get smaller because your firm is smaller. With no IT team to set guardrails, you are the guardrail. Use only an approved deployment that contractually will not train on your inputs, classify every matter before you paste, and keep privileged and client confidential material out of any consumer grade tool. When you are unsure whether a deployment is safe, the matter stays out until you have confirmed it. Anthropic and other major providers publish distinct policies for their consumer versus business and enterprise plans, and the difference is exactly the difference between a tool you may use on client work and one you may not.

The supervision duty still applies to a tool

Model Rules 5.1 and 5.3 govern supervising lawyers and nonlawyer assistance. The spirit reaches a model you direct: you remain responsible for the work the tool helps produce, exactly as you would for a paralegal's draft. There is no one above you to catch a miss and no one below you to blame. The duty to be right is yours and is not delegable.

How we built this method

This playbook reflects hands on use of leading general purpose models on the kinds of work a solo and small firm actually carry: intake summaries, issue framing, first draft letters and memos, and client communications. The four part stack is a practitioner workflow, not a product, and not a survey. The leverage math in this guide is explicitly illustrative: the two hours to one hour example is a worked illustration of how to calculate your own savings, not a measured statistic or a claimed result for any firm. The Leveraged Years just launched and we do not publish invented numbers, testimonials, or client results we do not have. Where we describe what AI is good and bad at, we mean what holds up in repeated practical use as of June 2026, on work that contained no real client confidential information. AI capabilities and vendor data policies change quickly, so we date this guide and refresh it. None of this is legal advice, and none of it changes your professional duties of competence, confidentiality, candor, and supervision. Confirm any approach against your jurisdiction's rules and your own obligations before using it on a live matter.

What this means for your week

You do not need a platform you cannot afford or an IT team you do not have to get most of the leverage AI offers a law firm. You need four jobs handed to a capable model you control, intake triage, research triage, drafting, and client comms, and the discipline to verify everything and keep confidential matters out of the wrong tools. Run the leverage math on your own most repeated tasks, not on a vendor's headline. The hours you reclaim from the toil go back into the billable judgment only you can provide, or back into your evenings.

That is the whole premise of how we train lawyers to work with AI: not faster, sloppier output, but the same standard of work reached with far less of the toil, with the bar duties built into the method rather than bolted on after. The Leveraged Attorney course is built to install this stack, the prompts, and the guardrails as a system a solo can defend to a client and a disciplinary board alike.

Part of TLY's AI Workflows → workflow playbooks for senior professionals.

Frequently asked questions

Can a solo lawyer use AI without an expensive legal platform?

Yes, and for most solo work a general purpose model is the better starting point than a six figure legal suite. On an approved business or enterprise plan that does not train on your inputs, a model like Claude can run intake triage, research triage, first draft drafting, and client communications, which is the bulk of where a solo's hours leak. The enterprise legal platforms are built for firms with deployment teams and large matter volumes. A solo gets most of the leverage from a disciplined method on a tool you control, at a fraction of the cost and with no IT project. The trade is that you supply the supervision and verification a bigger firm spreads across people.

How much time does AI actually save a small law firm?

Honestly, it depends on your work, and you should run the math yourself rather than trust a headline number, including any number a vendor offers. The way to estimate it is to take a task you repeat often, a status letter or an intake summary, time how long it takes you start to finish, then time the version where a model produces the first draft and you verify and rewrite. As an illustration only, a two hour task can become a one hour task once the drafting is handed off and you spend the saved time verifying. That is an example of the calculation, not a measured result. The real savings concentrate in your most repeated routine tasks, not in any single dramatic win.

What are the bar duties when a solo uses AI?

The same duties that apply to all your work, with no team to share them. Competence under Model Rule 1.1 means understanding the benefits and risks of the tools you use, including that a model can be confidently wrong, so you verify its output. Confidentiality under Model Rule 1.6 means keeping client information out of any tool not approved to protect it. The supervision duties in Model Rules 5.1 and 5.3 reach the work a model helps produce: you remain responsible for it as you would for a paralegal's draft. In a solo shop there is no second reviewer and no IT department, so every one of these duties lands on you personally. Confirm the specifics against your own jurisdiction's rules.

Is it safe to put client information into AI as a solo practitioner?

Only into an approved deployment, and only after you have classified the matter. Never paste privileged or client confidential material into a free or consumer grade tool whose terms may use your inputs to train models or do not contractually protect the data. Use a business or enterprise plan that contractually commits not to train on your inputs and meets your confidentiality obligations, or redact sensitive terms first, or keep that matter off the tool entirely. With no IT department to set the policy, you are the policy, so settle the deployment question once, deliberately, before any client matter touches a model, and when you are unsure, keep the matter out.

How is this different from running any solo business with AI?

The method rhymes, but a law firm carries duties a general solo business does not. Intake is not just lead capture: it triggers a conflicts check and a confidentiality obligation from the first contact, and AI does not get to make either call. The billable hour is the thing you are optimizing, so the goal is converting non billable toil into billable judgment rather than vague productivity. And the bar's competence, confidentiality, and supervision duties apply to every output, with no associate or partner to share them. Our general guide to running a solo practice with AI covers the universal playbook; this page covers what makes it legal work.

Build the stack, not just the opinion

Knowing the four jobs is the start. Running them every time, with the verification and confidentiality discipline built in and the bar duties carried by you rather than ignored, is the skill that compounds into a practice that is both faster and defensible. We teach the full stack, the prompts, and the guardrails as one repeatable system a solo lawyer can stand behind.

Start with Leveraged Attorney: the AI stack and guardrails built for solo and small firm lawyers Join The Leverage Club for $49 and get the intake, research, drafting, and client comms prompt templates Not sure where to start? Take the 2-minute course finder

Sources: Anthropic Claude consumer versus business and enterprise commercial data usage policies (Anthropic, 2026); ABA Model Rules of Professional Conduct on competence (1.1, including technology competence guidance), confidentiality (1.6), and supervision of lawyers and nonlawyer assistance (5.1 and 5.3); TLY hands on use of leading general purpose models on intake summaries, issue framing, first draft letters and memos, and client communications containing no real client confidential information (June 2026). The leverage math in this guide is illustrative, not a measured result. Capabilities and vendor policies as published as of June 2026 and subject to change. This guide is not legal advice.