First in a three-part series on putting AI to work in a law firm.
If you feel two things about AI at the same time, you are in good company. There is the pull that says this could take real work off your plate and the push that says one wrong move could put your license at risk. Both are reasonable. The problem is that almost everything written for legal practices picks one and ignores the other. You get breathless promises on one side and cautionary tales on the other, and neither helps you decide what to actually do on Monday morning.
That gap is why this series exists. Over three articles, the goal is to give you a practical, vendor-neutral view of AI in a law practice, the kind you would get from someone who has put these tools to work inside firms rather than someone selling a tool.
This first article covers what the technology actually is and what the current field of tools looks like, so the vocabulary stops being intimidating. The second looks at where AI is already earning its keep inside small firms. The third gets into how to start safely, including the data privacy and confidentiality questions that matter most to you and your clients. The State Bar of Michigan has been investing in this area, including its 2025 report, Transforming the Legal Landscape in the Age of AI,1 and the resources being built into the practice management center. This series is meant to sit alongside that work.
Let’s start with the thing everyone talks about and few people explain.
WHAT THESE TOOLS ARE ACTUALLY DOING
AI is a broad category of technology, not a single tool. It includes rule-based automation that follows a fixed set of steps, machine learning that finds patterns in data, and the large language models most people picture when they hear the word “AI” today — the kind behind tools like ChatGPT and Claude. This series touches on several of these categories, including the automation and workflow tools discussed in the second article, but this article focuses specifically on large language models, since that is where most of the current attention, and most of the confusion, sits. Strip away the branding, and the core idea behind a language model is simple.
A language model predicts text. You give it words, and it calculates what words should come next, one after another, based on patterns it learned from an enormous amount of writing. Do that well enough, and you get paragraphs that read as if a person wrote them.
That one fact explains both the promise and the risk. The promise is that a tool this fluent can draft, summarize, understand, and reorganize language faster than any human. The risk is that a model predicting likely text can produce something that reads perfectly and yet is false. In the legal world, this shows up as invented case citations — correct in format, plausible in name, and entirely fictional. This is commonly called a hallucination.
The part most coverage gets wrong, and the reason the “AI always makes things up” fear is overblown, is that not all tools hallucinate at the same rate, because not all tools are built the same way. A raw, general-purpose chatbot — without access to verified sources — has nothing to check itself against, so it guesses. Purpose-built legal tools work differently. They ground their answers in real, verified sources, retrieving actual authority before they generate anything, and they are trained specifically on legal material. A 2024 Stanford study comparing leading legal research tools found meaningfully lower hallucination rates in platforms built this way than in a general-purpose model with no legal grounding.2
The widely cited case of Mata v. Avianca,3 where two New York attorneys were sanctioned for filing a brief full of fake, AI-generated citations, was not simply a story about AI being useless. It was a story about attorneys who failed to verify the authority they cited, a nondelegable professional duty that does not change no matter what tool produced the first draft. The real takeaway is not just about picking the right tool for the job. It is that no AI tool, however sophisticated, eliminates a lawyer’s obligation to independently verify that tool’s output, its citations, and any legal statements it produces, before any of it goes out under your name.
THE FIELD OF TOOLS TODAY
When people say they are “looking at AI,” they are usually looking at one small corner of a much larger field. It helps to see the whole map, because the right choice depends entirely on the job in front of you.
| Category |
Examples |
What it’s for |
General-purpose
AI assistants |
ChatGPT (OpenAI),
Claude (Anthropic),
Gemini (Google),
Copilot (Microsoft) |
Broad, flexible
tools for drafting,
summarizing, and
general thinking.
Not built for legal
work; use care with
confidential facts. |
AI-native legal
platforms |
Harvey, Legora,
CoCounsel,
Spellbook, Robin AI,
Vincent AI (vLex),
CaseMarkLegal |
Built from the
ground up for legal
work and
grounded in real
legal sources.
Where hallucination
rates are
lowest. |
AI features inside
tools you already
use |
Clio, MyCase,
Smokeball, PracticePanther,
Filevine,
Westlaw, Lexis |
AI layered onto
practice management
and research
software you
already log into.
The gentlest
on-ramp for most
firms. |
Custom AI
workflow builds |
Pivot180 (the firm
behind this series)
and other firm-specific
automation
builders |
Not a product you
buy off the shelf. A
custom workflow
built around how
your firm operates,
for processes too
specific for any
existing tool to fit. |
You do not need to evaluate all four today. You need to know they exist and that AI is not one thing to accept or reject. It is a field of choices, and knowing the map is most of the battle. This series stays vendor-neutral by design, and that includes the tools built by our own firm: The right fit for your practice depends on your specific needs, budget, and workflows, so evaluate the options that make sense for you rather than defaulting to any single vendor, ours included. Several of the tools already sitting on your desktop, Clio, Filevine, Smokeball, and PracticePanther among them, are vendors the State Bar of Michigan has arranged member pricing or discounted terms with,4 worth confirming before you go shopping elsewhere. That benefit is typically preferred pricing on the vendor’s own product, not the State Bar providing the AI features themselves as part of your membership, so confirm exactly what’s included before you budget around it.
THE RULES ALREADY POINT THE WAY
You are not operating without guidance here. Your duty of competence under Michigan Rule of Professional Conduct 1.1 already extends to the technology you use, which means understanding the benefits and risks of these tools is part of the job. Your duty of confidentiality under MRPC 1.6 already governs where client information is allowed to go. The American Bar Association issued formal guidance on generative AI in 2024,5 and the State Bar of Michigan’s own 2025 report is worth an afternoon of your time. None of them say avoid AI. They say use it with the same care and supervision you bring to everything else that carries your name.
Because these obligations can raise considerations beyond what’s covered here, it is worth consulting your firm’s ethics counsel or your state bar’s ethics hotline to determine your specific obligations before putting a new tool into practice.
THE MINDSET THAT MAKES IT WORK
Here’s the shift that separates firms getting real value from AI from firms just poking at a chatbot. Stop thinking of it as software you switch on, and start thinking of it the way you’d think about bringing on a new hire.
A brand-new employee with no training, no standard process to follow, and no feedback loop makes mistakes. Give that same employee a clear procedure, examples of what good work looks like, and consistent correction when they miss, and within weeks they’re reliable. AI works the same way, whether it’s a general-purpose assistant or a legal-native platform. The tool that earns its keep is the one given the same structure you’d give a person doing the job: a written standard operating procedure, examples of “done right,” and someone reviewing its output and correcting the pattern, not just the one mistake.
The advantage is what that AI “hire” costs you. It works for a fraction of a salaried employee. It doesn’t take sick days or vacation, and it doesn’t have an off week. It can draft briefs in Spanish or Arabic as easily as in English. And it doesn’t get tired reviewing document 40 of 2,000.
None of this replaces judgment, and none of it replaces your license. But it means the return on training an AI system properly shows up faster than the return on training a person, and that changes the math on whether it’s worth doing.
NO LICENSE, NO EXCUSE
You can hold two ideas at once. AI has no license and no stake in the outcome, which means you are always the one accountable for what goes out the door under your name. And AI, given the same structure you’d give any new team member, becomes dramatically more dependable than a tool you just open and start typing into. The firms pulling ahead are writing the SOP, training the assistant, and reviewing its work, exactly as they would for a person, and getting their evenings back because of it.
In the next article, we’ll get specific about where that value starts by looking at common use cases where you can put AI to work for your firm.