If you practice at a small or midsize firm, you have probably noticed two things happening at once. Every legal publication is breathlessly covering the latest AI product launch, and almost none of that coverage is written with your firm in mind. Harvey’s enterprise pricing, BigLaw adoption statistics, and AmLaw 100 case studies make for interesting reading, but they do not answer the question that actually matters to you: what should a solo practitioner or a six-person firm (or practice group) be doing about AI today?
Start With an Honest Definition
Artificial intelligence, in the form that matters for your practice, is a prediction engine trained on an enormous amount of text — case law, statutes, treatises, and countless other documents. Its core skill is predicting, given what came before, what should come next. That is a very different thing from thinking, and the distinction matters enormously for how you use it.
A useful way to frame it for yourself: imagine an extraordinarily well-read associate who has reviewed millions of documents but has never practiced law, never met a client, and bears no accountability for what they produce. They can generate a polished, confident memo in seconds. They can also be completely wrong without any indication that anything is amiss.
That combination — fluency without judgment — is the single most important thing to understand before you adopt any AI tool.
The Risk That Should Shape Every Decision You Make
The fluency-without-judgment problem has a name: hallucination. It is when an AI model states something false with the same confidence and polish as something true. It is not a bug. It is the model doing exactly what it was built to do — producing the most statistically likely response — when it has no actual way to verify what is correct.
This is not a hypothetical risk. Attorneys have been sanctioned for submitting briefs containing fabricated case citations that an AI tool generated with complete confidence, including realistic docket numbers and plausible-sounding holdings. The lesson is not “don’t use AI.” The lesson is “never rely on AI output you have not independently verified” — particularly for citations, statutory references, and anything that will be filed or sent to a client without your direct review.
Where Small Firms Should Actually Start
The biggest mistake small firm attorneys make is either avoiding AI entirely out of caution, or diving into the most advanced and expensive tools without understanding what is actually useful at their scale. Neither approach serves you well.
The better approach is to start with reactive AI — tools that respond to a specific prompt you control, where you remain in charge of every step. You ask, the tool produces output, you verify and act. This is the lowest-risk entry point and the one your competence and supervision obligations are easiest to satisfy with.
From there, the highest-value use cases for a small or midsize firm tend to cluster around a few categories:
Research and summarization. Digesting case law, statutes, and secondary sources into structured summaries in minutes rather than hours. For anything you intend to cite, use a legal research platform with database grounding rather than a general AI tool.
First drafts. Contracts, demand letters, client memos, and routine correspondence. The discipline that matters here is treating every draft as a starting point, never a finished product.
Document review. Lease abstracts, due diligence checklists, and discovery summaries — high-volume work where AI assistance produces measurable time savings, provided someone is still reviewing the output.
Billing narratives. An underappreciated use case. Converting time entry notes into polished, client-ready narratives saves real time on a task most attorneys dislike and clients scrutinize closely.
Client communication. Plain-English explanations of complex legal concepts, intake summaries, and status updates — places where AI’s fluency is a genuine asset rather than a liability, because the stakes of an imprecise turn of phrase are lower than the stakes of an imprecise legal citation.
The Tool Question: What Actually Fits a Small Firm Budget
The legal AI product landscape has exploded, but most of the coverage is written for firms with budgets your firm does not have. Harvey’s enterprise pricing starts in the hundreds of thousands of dollars annually with seat minimums that make it irrational for a fifty-attorney firm, let alone a six-attorney one.
The practical decision tree for a small or midsize firm looks different:
If your firm already subscribes to Westlaw, your most natural entry point is Thomson Reuters’ CoCounsel, which grounds its output in Westlaw’s database rather than the open internet. If you already use Lexis, Lexis+ with Protégé offers the parallel option, with the added benefit of letting you toggle between different underlying AI models.
If your work is primarily transactional or contract-focused, Spellbook is worth evaluating — it runs inside Microsoft Word, which means no new interface to learn, and its pricing is far more accessible than enterprise platforms built for large firms.
And do not overlook the general tools you may already have access to. Claude, ChatGPT, and similar platforms are powerful, flexible, and — when used with appropriate protocols — entirely usable for non-client-specific drafting, research, and internal efficiency work. They carry the highest confidentiality risk of any category, which brings us to the issue that should concern you most.
The Confidentiality Problem Courts Are Now Taking Seriously
Here is something that should change how you think about which AI tools you use: courts have begun ruling that uploading confidential client material to public AI tools can waive attorney-client privilege, treating the upload as functionally equivalent to publishing the material on the internet. The waiver occurs at the moment of upload, regardless of whether you intended it or believed the information would remain private. Federal courts in the United States have issued similar rulings, in one case declining to permit an open AI tool for e-discovery work and ordering that only closed, secure tools be used instead.
This is not a future risk. It is current law, and it should inform a concrete policy at your firm today.
The practical framework is this. There is a meaningful difference between public tools, which accept inputs from anyone and may use those inputs to train future models, and enterprise or legal-specific tools, which operate within contracted, secure environments with explicit data protections. Before you enter any client-identifying information into an AI tool, ask three questions: where does my input go, is it used to train the underlying model, and what contractual protections actually apply. If you cannot answer those questions with confidence, the safest assumption is that the information is not protected.
For most small firms, the practical rule is straightforward: use general AI tools freely for non-client-specific work — research on legal concepts, drafting templates, internal efficiency tasks — and reserve client-specific, confidential work for enterprise-grade or legal-specific platforms with explicit data protection terms.
Your Professional Responsibility Obligations Have Not Changed — They Have Just Gotten More Specific
ABA Formal Opinion 512 and the underlying Model Rules did not create new categories of obligation for AI use. They applied existing duties — competence, confidentiality, and supervision — to a new context.
Competence under Rule 1.1 means you must understand an AI tool well enough to recognize when its output is wrong. You do not need to understand the underlying architecture. You do need to know that the tool can hallucinate, that it has a training cutoff and may not reflect recent legal developments, and that fluent output is not the same as accurate output.
Supervision under Rule 5.3 means AI-generated work product is treated exactly like work product from a paralegal or associate. You are responsible for what leaves your office regardless of how it was produced. A will that looks polished but uses an outdated exemption amount is not a time-saver. It is a malpractice exposure with your name on it.
Confidentiality under Rule 1.6, as discussed above, requires active attention to where your inputs go and what protections actually apply — not just an assumption that a vendor’s promises are sufficient.
A Realistic Starting Point
If you are running a small or midsize practice and feeling behind, here is a sequence that does not require a large budget or a steep learning curve.
Start by picking one repetitive task — a research memo format, a billing narrative process, a standard letter you draft often — and build a single, well-structured prompt for it. Give the AI a role, a clear task, the relevant facts, and the format you want. Test it on real matters. Refine it until the output requires minimal editing.
Once you have a handful of templates working reliably, evaluate whether a legal-specific platform makes sense given your practice volume and existing subscriptions. Build a one-page internal policy covering what tools are approved, what client information can and cannot be entered into them, and what verification is required before anything goes out the door.
None of this requires an enterprise contract or a dedicated legal-ops team. It requires a deliberate, modest first step — and the discipline to verify everything before it reaches a client or a court.
The Bottom Line
AI is not a BigLaw phenomenon that will eventually trickle down to smaller firms. It is already accessible, already affordable, and already capable of producing real efficiency gains for solo and small firm practitioners — provided you understand what it actually is, where it can fail, and what your professional obligations require before you rely on it. The firms that will benefit most are not the ones with the biggest technology budgets. They are the ones who start deliberately, verify consistently, and treat every AI output the way they would treat work from a new associate: useful, often excellent, and never beyond question.
This post is intended for general informational purposes and does not constitute legal advice. Attorneys should consult their jurisdiction’s specific bar guidance and ethics opinions regarding AI use in legal practice.