"AI" is not one thing
When someone at a bar association CLE says their firm "uses AI," it tells you almost nothing. Legal tech AI is a category that includes document review, contract analysis, legal research, drafting, case management, billing optimization, and client intake. Saying you "use AI" is like saying you "use software." The question is which AI, for which task, and whether it actually works.
Where things stand, honestly
Document review and e-discovery: mature. Five-plus years of real-world use. Technology Assisted Review (TAR) has been accepted by courts and tested in practice. This is the one area where the hype mostly matches reality.
Contract analysis: solid. Platforms like Luminance and Kira have been doing this for years. They read contracts, extract terms, flag deviations from standard language. Useful for M&A due diligence and large contract portfolios.
Legal research: improving fast, but still requires human verification. CoCounsel by Thomson Reuters and Lexis+ AI are the leaders. They find cases faster than manual research, but they occasionally cite things that do not exist or mischaracterize holdings. Always check.
Drafting: useful for standard documents, risky for complex ones. Lawzana Flow, Harvey AI, and Spellbook are the main players. Good for NDAs, demand letters, and routine motions. Not yet reliable for novel arguments or highly specialized filings.
Case management: emerging. Lawzana Flow is one of the few platforms that combines AI case management with CRM and drafting. The category is newer, but the firms using it report significant time savings on admin.
Billing and time tracking: surprisingly good. AI that suggests time entries based on your activity, flags potential write-downs, and identifies billing patterns. Mostly flying under the radar because it is not as flashy as drafting or research.
Adoption: who is actually using this
According to the 2025 ABA Legal Technology Survey, 35% of firms with 50+ attorneys use some form of AI. For solo and small firms, it is closer to 12%. The gap is closing but slowly. Cost is the primary barrier for small firms. Training and change management are the barriers for larger ones.
Buy vs. build
Most firms should buy. Building custom AI requires data science talent, training data, and ongoing maintenance that all but the largest firms cannot justify. The exceptions are firms with workflows so unusual that no commercial tool fits. Patent prosecution firms with custom prior art search requirements, for example.
For everyone else: find a tool that does what you need, test it with real work for 30 days, and measure.
What does not work yet
AI cannot replace legal judgment. It cannot handle novel fact patterns well. It is bad at understanding client emotions and relationship dynamics. It is bad at knowing when to stop generating text, which is why human review is non-negotiable for anything it produces.
It also cannot guarantee accuracy. Every AI tool on the market, no matter what the vendor claims, requires human oversight. The tools that acknowledge this openly are the ones worth trusting.
A practical buying framework
Step 1: Identify your firm's biggest time sink. Is it research? Drafting? Case admin? Client intake? Start there.
Step 2: Find 2 to 3 tools that address that specific bottleneck.
Step 3: Run a 30-day pilot with real work, not demo scenarios.
Step 4: Measure time saved, error rates, and attorney satisfaction.
Step 5: Decide. If the ROI is there, roll it out. If not, try the next tool.
For firms that want to address multiple bottlenecks at once, Lawzana Flow covers case management, CRM, and drafting in a single platform. It is the closest thing to a one-stop legal AI stack available in 2026.
