The Claude Test: How to Know If an AI-Powered Legal Tech Product Is Actually Worth Buying

Walk the expo floor at any legal tech conference right now and count the booths with “AI” in the name or the tagline. At ABA TECHSHOW 2026, it was most of them. AI intake. AI timekeeping. AI drafting. AI for everything.

Two years ago, that density would have felt exciting. Today it feels like something else—not cynicism exactly, but a kind of wariness that wasn’t there before. The lawyers walking those aisles have changed. They’ve spent a year using Claude, Copilot, ChatGPT. They’ve learned what AI does easily. And that experience has quietly shifted what “AI-powered” means as a selling point.

It used to open doors. Now it raises a question: compared to what?

The Test

Here’s the question. Open Claude. Describe the problem the product claims to solve. See how far you get.

If you can replicate most of what a product does in a single conversation—for free, right now, without a subscription—that product isn’t solving your problem. It’s charging you for a prompt.

That’s the Claude Test. It’s not a knock on AI. It’s not a reason to cancel your entire tech stack. It’s a filter—a first question to ask before you buy, and an honest question to ask about what you’re already paying for. It separates products built on real foundations from products that borrowed AI’s credibility without building anything underneath it.

Why This Test Didn’t Exist Two Years Ago

Building an AI-powered legal product used to be technically hard. That difficulty created a natural filter. If a vendor had shipped something, they’d at least solved an engineering problem. That filter is gone.

Today anyone can build a wrapper. The cost of shipping a product that summarizes documents, drafts intake emails, or generates form language is close to zero. The tools to build it are the same tools your firm already uses. There is a whole movement—vibe coding—built around the idea that you can describe what you want and have working software in an afternoon.

When building gets that easy, the products with real foundations and the products with nothing underneath them start to look identical from the outside. Same landing page. Same demo. Same pitch. The Claude Test is how you tell them apart—because the ones with nothing underneath will fail it immediately.

What Actually Passes

The products that survive the Claude Test share something in common: their value isn’t in the output they produce. It’s in what they’ve built, accumulated, or become over time. There are six categories worth knowing.

Problems that were already hard. If solving something required years of work before AI existed, AI didn’t make it easy. It just made everything else easy by comparison. The dominant document management platforms in legal—NetDocuments, iManage—aren’t valuable because they store files. They’re valuable because they’ve spent decades building the permissioning architecture, compliance infrastructure, security certifications, and migration tooling that enterprise legal work requires. Those things took a long time to build and they’d take a long time to replace. AI didn’t change that. It just made everyone else’s moat look shallower.

Networks and relationships. You can’t prompt a network into existence. ABC Legal has over 8,000 process servers nationwide—that’s not a feature, it’s a logistics operation built over decades. BriefPoint has live production integrations with Clio, 8am MyCase, and Smokeball. LawToolBox has spent years maintaining court rules databases that power deadline calculations across jurisdictions. These things exist because of relationships and commitments made long before any of them had an AI layer. The AI layer, when it came, was just a new surface on a real foundation.

Accountability. When something goes wrong, who answers for it? A firm that uses Claude to draft a discovery response owns all of that risk—every word, every citation, every omission. A product that maintains jurisdiction-specific compliance rules, carries SOC 2 certification, and produces an auditable record of what it did has shifted some of that accountability away from the firm. That transfer of responsibility is a real thing of value. It is not something you can ship in version one, and it is not something a general-purpose AI tool provides. Products that carry accountability—genuinely, contractually, operationally—are providing something structural that the Claude Test cannot replicate.

Workflow position. There is a difference between a product that touches your workflow and one that is your workflow. The products that become load-bearing—that your billing connects through, that your DMS indexes against, that your conflict checks pull from—are not replaceable by a better prompt. Removing them is a project, not a decision. That depth is not lock-in as a flaw. It’s integration as a feature. The products that have earned that position did it by being reliable over years, not by shipping a clever AI feature. The products that haven’t earned it yet are always one good Claude conversation away from being unnecessary.

Trust and the human layer. As AI commoditizes outputs, the person or firm that vouches for the outcome gains value. Ruby Receptionists has spent over two decades building a reputation for answering phones with genuine human warmth and legal market knowledge. That is not a feature that can be replicated at zero cost. The same is true for the consultant who has been inside your systems for eight years and knows where every integration lives and why every decision was made. (Cough, cough, Affinity Consulting). Earned credibility and human judgment are not version-one features. They accumulate. The products and people who have accumulated them are providing something AI cannot undercut.

Scale and operational consistency. Doing something once with AI is easy. Doing it the same way ten thousand times, with a complete audit trail, across a two-hundred-attorney firm, without a human reviewing each instance—that is a product. The tools that operationalize AI rather than simply exposing it are providing something a general-purpose model cannot: governance, repeatability, and organizational trust. The difference between “we use Claude for this” and “we have a system for this” is the difference between a capability and an operation. Enterprise-grade platforms that wrap AI in real process infrastructure are building on that gap.

What This Means for Your Buying Decisions

The Claude Test isn’t only for evaluating what you’re about to purchase. Run your current stack through it.

Not as a reason to cancel everything—workflow position and deep integrations may still justify the cost of a product even if its AI features would fail the test. But as a lens for honest conversations with yourself and your vendors. What am I actually getting here? If the AI layer disappeared tomorrow, would I still need this? What is load-bearing and what is just convenient?

The question worth sitting with: if you rebuilt your tech stack from scratch today, which products would you pay for without hesitation? Those are your foundations. The ones you’d hesitate on—that’s the test doing its job.

Can Your Firm Pass the Claude Test for Your Clients?

The same test applies to your practice. Which parts of what you do would a client struggle to replicate with a good AI system and an afternoon? Which parts of your work are loops dressed up as expertise—pattern-matching that AI is coming for first?

The firms that can answer that honestly are the ones positioned to make good technology decisions. And to build something that lasts.