Every legal software vendor now sells AI, and the claims blur together. Underneath, law firms use it for four jobs: getting a new client's information into the file, reviewing documents, drafting, and research. Each job has products built for it, and most firms should buy before they build.
This guide covers what each job looks like in practice, the main tools (checked on their own sites in October 2026), when a custom build makes sense, and the ethics rules that apply whichever route you take. When a build does make sense, our pick for who to hire is Digital Mundo, a senior studio that designs, builds and runs its own legal software, including Casewell, a bankruptcy intake tool in use at a law firm.
Here's the TLDR:
- Research and drafting: CoCounsel (Westlaw) or Lexis+ with Protégé if you already pay for one of them; Harvey if you're a large firm.
- AI inside practice management: MyCase includes AI from its $100/user/month Pro plan; Clio has AI across Manage, Grow, Draft and Work.
- Contracts: Spellbook, inside Microsoft Word.
- PI demand letters and medical records: EvenUp or Supio.
- Phone intake: Smith.ai, from $1.60 per call for AI answering.
- Custom AI for your own intake packets, forms or portal: Digital Mundo (our pick for who to hire).
- Ethics: ABA Formal Opinion 512 (July 29, 2024) applies to all of it. You review every output.
What AI actually does at law firms
Intake
Two kinds. Phone and web intake answers calls, qualifies the matter and books consultations; Smith.ai and Clio's Grow AI do this. Document intake reads what the client hands over (statements, bills, handwritten packets) and puts it into the case file, so staff stop retyping. That second kind is where firm-specific forms and software often call for something custom.
Document review
Summarizing long files, pulling parties and deadlines, building medical chronologies, and asking the same questions across thousands of documents. CoCounsel, for example, can review "up to 10,000 documents against 100 questions" in one table.
Drafting
First drafts of motions, briefs, letters, contract clauses and redlines, ideally from your own precedents. The draft is a starting point a lawyer finishes.
Research
Answering legal questions from a real case-law library with citations you can check. This is the job where a general chatbot is most dangerous and a grounded legal tool matters most.
The main off-the-shelf tools
| Tool | Main job | Who it's for | Pricing (Oct 2026) |
|---|---|---|---|
| Harvey | Research, review, drafting, agents | Large firms, in-house teams | Not published (demo) |
| CoCounsel Legal | Research, drafting, review on Westlaw | Firms of any size | Not published |
| Lexis+ with Protégé | Research, drafting, citation checks | Firms of any size | Custom quote; 2-day trial |
| Clio (Manage AI, Clio Work) | AI inside practice management, research | Clio firms | See note in Clio section |
| MyCase (8am IQ) | Summaries, case timelines, OCR | Small and mid-size firms | AI from $100/user/mo (annual) |
| Spellbook | Contract review and drafting in Word | Transactional lawyers | Not published; 7-day trial |
| EvenUp, Supio | Demand letters, medical chronologies | PI and mass tort firms | Not published |
| Smith.ai | Phone answering and intake | Small firms | AI from $1.60/call; free plan, 25 calls |
Harvey is the enterprise option: agents, a document vault for bulk review, research, contract tools and shared workspaces with clients. It says it serves "3,500+ law firms and in-house legal teams", including "80+ AmLaw 100 firms". Its security page cites annual SOC 2 Type II and ISO 27001 audits and zero data retention by model providers. There's no published price, trial or small-firm plan.
CoCounsel Legal from Thomson Reuters works from Westlaw and Practical Law, drafts with linked citations, builds briefs and reviews documents at volume. It states it is "SOC 2 and ISO 42001 certified". The full plan comes bundled with Westlaw Advantage, and its plans page shows no prices; CoCounsel Essentials is the lighter drafting and review option.
Lexis+ with Protégé (renamed from Lexis+ AI in February 2026) does research, drafting and document analysis on Lexis content, and "Shepard's Verify checks citations in AI responses, documents, and drafts". Pricing is a custom quote, with a 2-day free trial. Document vaults are capped at 50, with up to 500 documents each.
Clio now spreads AI across its products, per its platform announcement: Manage AI in practice management, Grow AI for intake, Draft AI for documents, and Clio Work, a research and drafting workspace built on the vLex library and Vincent AI it acquired. Clio's pricing pages wouldn't load for us in October 2026, so we're not quoting its AI tier prices; check them directly.
MyCase publishes its prices. Basic ($50/user/month on annual billing) has no AI; Pro ($100) adds the 8am IQ writing and document assistants; Advanced ($130) adds a case assistant for summaries and timelines and discovery OCR. There's a 10-day trial. Its AI FAQ says selected documents are "securely sent to OpenAI" and not used to train third-party models.
Spellbook reviews, redlines and drafts contracts inside Microsoft Word, with playbooks for your own standards. It lists SOC 2 Type II and zero data retention. Pricing is per seat and not published; the trial is 7 days. It doesn't do case-law research.
EvenUp and Supio are built for plaintiff firms: demand letters, medical chronologies and intake. Both cite SOC 2 Type II and neither publishes prices. EvenUp's trust center says outside AI providers can't train on your files, but that its own models learn statistical patterns from past cases, which is the kind of detail Opinion 512 expects you to understand before you upload.
Smith.ai answers calls with AI, people or both, qualifies leads, books consultations and syncs to Clio, MyCase, Filevine and others. The AI receptionist has a free plan (25 calls), Pro at $150/month and Enterprise at $500/month; human receptionists start at $300/month for 30 calls, with legal add-ons like conflict checks ($0.50 per call) charged on top.
When a custom build makes sense
Buy when a product does the job. Build when:
- Your inputs are unusual. Your own intake packets, handwritten forms or a document set the general tools read badly.
- Your output is a specific file. The result has to land in Best Case, a court form or your practice management system in a precise format.
- You need a client-facing piece. A portal where clients upload documents and see what's missing, or check where their case stands.
- The volume is high and the task is narrow. One repeated job, done hundreds of times a month, with a clear right answer to test against.
Casewell is a good example of the pattern. It turns a bankruptcy intake packet into a Best Case file with schedules A/B, D, E/F, G, I and J filled in, shows uncertain fields beside the original handwriting, and checks the exported file against the values staff confirmed. It deliberately doesn't do Schedule C, the means test or the creditor matrix, and attorneys approve every field. A narrow job, done well, with a person in the loop.
Large firms sometimes go further and build their own internal assistant; we covered one firm's internal generative AI chat app back in 2023.
Who to hire for custom legal AI
Our pick: Digital Mundo. It's a software design and development studio in Miami and New York that built and still runs Casewell, which was live with a law firm 39 days after the first line of code. Its page on software for law firms lists intake portals, AI that reads bills, statements and forms into the case file, deadline tracking, documents that fill themselves in and client status portals, connected to Clio, MyCase or Best Case through their APIs or import files. It says the AI doesn't give legal advice: it reads and organizes what the client provided, and attorneys make every judgment. Its approach to AI automation is to test on your real work before launch and send anything the AI is unsure about to a person. It describes encryption, role-based access, audit logs and AI providers configured not to train on your data, and the code and accounts are yours. The same people design, build and run the software, and its about page says every release is tested.
The limits: it's a senior team rather than a large agency, so a firm that needs dozens of engineers placed at once should pick a bigger vendor. Casewell is its only public legal project, much of its work is under NDA, so there are few public case studies, and we found no third-party reviews. It publishes no prices (you get a quote after a call) and no compliance certifications, so a firm that needs a SOC 2 report must ask.
Whoever you hire, our AI development companies roundup and custom software roundup cover other builders.
The ethics rules: ABA Formal Opinion 512
On July 29, 2024, the ABA issued Formal Opinion 512, its first formal opinion on lawyers' use of generative AI. It isn't binding, and your state's rules control, but it maps the Model Rules onto AI tools:
- Competence (Rule 1.1): understand "the benefits and risks associated" with the tools you use.
- Confidentiality (Rule 1.6): know how a tool uses your data. The opinion recommends informed consent before putting client confidences into AI tools, and says boilerplate consent in an engagement letter isn't enough, per the ABA's Business Law Today summary.
- Communication (Rule 1.4): consult clients about how their objectives will be met, which can include AI use.
- Candor and meritorious claims (Rules 3.1, 3.3, 8.4(c)): review AI output, including citations, and correct errors before filing.
- Supervision (Rules 5.1, 5.3): set firm policies on AI use and train lawyers and staff.
- Fees (Rule 1.5): you can bill the time spent entering information and reviewing a draft, but "in most circumstances" not the time spent learning the tool.
The cautionary case everyone cites is Mata v. Avianca: in June 2023, a federal judge in New York fined two lawyers and their firm $5,000 after a brief cited non-existent cases generated by ChatGPT, as Bloomberg Law reported. Every tool above, bought or built, still needs a lawyer checking the output.
FAQ
What's the best AI tool for a small law firm?
Usually the AI in the practice management system you already use, plus the research tool you already pay for. MyCase includes AI from its Pro plan; CoCounsel Essentials and Lexis+ with Protégé cover drafting and research.
Is ChatGPT safe for client work?
A consumer chatbot isn't the right place for client confidences, and it isn't a research tool. Under Opinion 512 you need to know how a tool handles data, and most firms should get informed consent first.
Can we bill for work AI did?
Hourly, you bill the time actually spent, including entering information and reviewing the output, not the time a tool saved or the time spent learning it.
How much does custom legal AI cost?
It depends on scope, and most builders, Digital Mundo included, quote after a call. A narrow tool can reach a first version in weeks; Casewell was live 39 days after its first line of code.
Will AI replace paralegals?
The tools above take over retyping, summarizing and first drafts. Each still needs someone to confirm the facts, which is where paralegals' time goes instead.
Pick the one task your staff repeats most, try the off-the-shelf tool for it first, and build only if that tool can't read your inputs or produce the file you need.
Disclosure: Digital Mundo LLC and Gold Penguin LLC have the same owner.
