Half Baked AI Isn't Cutting It.

A conversation between Josh Schmerling, co-founder of LawPro.ai, and Nathan Walter, CEO and co-founder of Briefpoint, on what's actually working in legal AI right now, and what isn't.

There's no shortage of legal tech right now. According to Legal Tech Hub, the number of legal tech companies has gone from around 50 a few years ago to more than 1,300 today. That's a lot of pitches landing in your inbox. It's also a lot of platforms promising to do everything, from intake to resolution, for a single (often eye-watering) price.

So we wanted to cut through it. Josh Schmerling sat down with Nathan Walter, founder of discovery automation platform Briefpoint, to talk honestly about where the legal AI market actually is in 2026. The full conversation is worth a watch, but here are the takeaways our team kept coming back to.

1. The "do-it-all" platform is losing the argument

Adoption is no longer the story. In the survey LawPro.ai ran with Morgan & Morgan across more than 300 personal injury firms, 92% of firms reported at least some engagement with AI. Only 8% said they weren't using it at all.

What's changed is the type of AI firms are choosing. Both Josh and Nathan landed in the same place: horizontal "all-in-one" platforms tend to do many things at 60%, and that gap matters. As Nathan put it, when a tool only solves a problem partially, it often creates new ones. You end up wrangling formats, learning prompts, and forcing a triangle solution into the square hole of your actual practice.

Point solutions go deeper. They commit to one workflow, discovery, medical record review, demand drafting, and they keep getting better at exactly that. They're also typically less expensive, because they have to be. They're competing with what you can already get out of ChatGPT or Claude for a couple hundred dollars a month.

"I love LawPro.ai because I feel it doesn't try to be everything to everyone. It does what it does, and it does it well." — Lee Rudin, Attorney

2. ChatGPT is the new baseline, so vendors need a real moat

Nathan made a sharp point: with ChatGPT, Claude, and Gemini now table stakes, every legal tech vendor needs to answer one question, what do you offer that an out-of-the-box LLM can't?

He outlined four moats worth looking for:

• Legal repository data. Court records, statutes, case law at depth (Westlaw, Lexis, vLex).
• Institutional data. A firm's accumulated experience and judgment.
• Large repository data. Reliable performance across thousands or millions of pages, where general LLMs degrade.
• Systems of record. A single, living place where work product lives and stays current.

Josh shared a test he runs at conferences. He uploaded a small set of fictitious medical records to ChatGPT and asked for a complete list of dates, treatments, and diagnoses. ChatGPT returned about half of what LawPro.ai pulled from the same records. The reason isn't that ChatGPT is bad, t's that it isn't trained for this specific task at scale. LawPro.ai processes millions of pages every day with a model purpose-built to extract event types from medical records. That's the moat.

There's also a hard ethical line here. General-purpose LLMs are typically not HIPAA-compliant and don't carry a BAA. Pasting client medical records into them can put bar licenses, work product privilege, and client confidentiality at risk. Recent cases have already started to reflect that.

3. The verification rule

If there's one quote from the webinar to tape to your monitor, it's this one from Nathan:

"Any tool that you use for AI must take you less time to verify those results than it would have taken you to just do the thing yourself the first time."

That's the entire test. And it's why "show your work" matters more than feature lists. BriefPoint shows the source documents behind every RFP response and lets you click "verify" to see what the AI grouped, what it included, and what it explicitly excluded. LawPro.ai links every fact in a chronology, summary, or report back to its source page, and runs patent-pending hallucination prevention before any output reaches the user.

This is what 69% of PI firms said matters most when they told the Future of Legal Tech 2026 report that accuracy is the #1 barrier to deeper AI adoption. Speed is expected. Confidence is the differentiator.

4. The 8% has a problem, and it's not a tech problem

Josh asked the question a lot of firms are quietly asking: who are the 8% still sitting it out, and what are they thinking?

Nathan started from sympathy. Successful litigators built their practices around very specific workflows. Their staff is trained to it. Their cases run on it. Asking that machine to change is not a small ask.

But the existential pressure isn't coming from AI tools. It's coming from AI-native firms. Y Combinator has stopped funding new legal tech companies that sell to attorneys, and started funding AI companies that compete with attorneys. Combine that with private equity money flowing into law firms via MSO structures, and the calculus changes. As Nathan said, choosing not to use AI is no longer a long-term strategy.

5. Implementation isn't a tech problem either

The firms that get value from AI aren't the ones with the biggest budgets. They're the ones that get three things right:

Communicate the vision. Staff who think AI is here to take their job will not adopt it. Tell them, clearly, what their job becomes, usually one with more client contact, more strategy, and less staring at a 600-page PDF. Josh's own firm reframed paralegal work as a higher-skill role focused on clients, and watched adoption follow.

Pick one specific use case and start there. Don't roll out a horizontal platform across the firm and hope. Pick a real pain point, medical record review, discovery responses, demand drafts, and use the tool every time that situation comes up.

Get the vendor on the call. Nathan recommends having the vendor work through your first real case with you, on a screen share. That's how you build best practices early, instead of inventing them under deadline pressure.

"Prior to LawPro, my paralegals would spend days or even weeks creating a chronology for a simple motor vehicle accident. With LawPro, it can be done in less than an hour." — Zirkin & Schmerling Law

6. What to ask a vendor before you sign

Nathan's checklist is a good one, and one we'd add to our own:

• Security. SOC 2 Type 2 certified (not "pending"). Zero data retention. A signed BAA. A real data use policy.
• Support. A human who answers when something breaks. Because something will break, and the vendor knows how to fix it faster than you do.
• Training. Unlimited onboarding for your team — and for every new hire after them. Adoption dies when the second wave of staff doesn't know how to use the tool.

A practical bonus: stress-test your own briefs

Late in the conversation, Nathan offered a tactic worth stealing. Instead of asking AI to grade your own brief, it'll be sycophantic, and tell you it's great, paste in your brief and tell the AI it's opposing counsel's. Ask for the strongest arguments against it. You get a real critique. And as a bonus, you see roughly what opposing counsel will see if they're using the same tools.

Josh added a LawPro.ai variation customers already use: drop in the defense expert's report and have the platform generate cross-examination questions targeting what the expert didn't include. Customers have reported real shifts in case value from exactly this.

The bottom line

The conversation kept circling back to one idea. AI in 2026 is not magic, and it's not optional. It's a partner, one that has to earn its place by solving a specific pain point, showing its work, and protecting your clients' data.

And if you'd like to see how LawPro.ai handles your medical records, every page of them, book a demo.