The Tool You're Looking For Isn't Built
ABA TECHSHOW 2026. McCormick Place, Chicago. Notes from the other side of the expo hall.
Jordan Furlong is telling a room full of lawyers that the age of AI marks the end of law as we know it.
I’m sitting in the back, and I’m thinking about a man in the Palm Beach County jail who hasn’t seen his evidence yet.
Furlong is talking about how lawyers will no longer be hired for what they know, but for how they judge, guide, and lead. He’s describing the post-AI attorney as an architect, a guardian, a counselor whose value comes from the irreducibly human act of standing next to someone and saying, I believe you have something worth hearing. And he’s right. I think he’s exactly right. But the client I’m thinking about doesn’t need a post-AI lawyer. He needs a pre-AI one — the kind who can get evidence reviewed before the next calendar call.
There’s a gap in this room that nobody is naming. The gap between the future of law and the present tense of criminal defense.
By Thursday afternoon I’m on the expo floor and I’m learning something about myself that I didn’t expect to learn at a technology conference. I’m learning where I fit — and where I don’t.
People see me. They scan my badge. They start their pitch. And then I say the words “public defender’s office” and the conversation shifts. One vendor — a practice management platform, cloud-native, AI-embedded, award-winning — visibly recalibrated when he realized I wasn’t a buying decision. I don’t blame him. He has a product built for a market, and I’m not in that market.
But I came here looking for something, and I haven’t found it yet.
What the expo floor is solving is real. The death of the billable hour. The optimization of client intake. The automation of time entry so that attorneys can capture more revenue without thinking about it. These are genuine problems for the people who have them. I respect the engineering. I just can’t use it.
I don’t think in billable hours. I don’t think in hours at all. I think in jail visits. Calendar calls. Evidence dumps that arrive weeks late and contain the suppression issues and civil rights violations buried in body camera footage that no one has watched. I think about evidence that starts at a sheriff’s deputy’s body-worn camera, gets uploaded to an Axon docking bay when the officer ends their shift, sits there, gets pulled by the State Attorney’s Office, sits there longer, gets sent back to us through the discovery portal, and then we send it right back onto Axon’s platform to review. A round trip. And the client lost time in custody while the footage sat at every stop, and the issues — the Fourth Amendment issues, the Brady material, the inconsistencies between what the officer wrote and what the camera recorded — could have been flagged by a model running locally on a workstation in our office the day the evidence arrived.
Nobody at this conference is building that. Nobody’s booth says We process body camera evidence for public defenders at the speed the Sixth Amendment requires.
I don’t want to overstate this. The conference isn’t hostile. It’s just oriented toward a different gravitational center. The sessions are good. The Barcus and Joy presentation on AI security threats names the exact problem I’ve been solving for months: the risk that AI tools which train on user inputs will leak confidential data back out. They frame it through the NIST AI 600-1 framework and ABA Formal Opinion 512. They’re right about the threat. But they don’t mention the alternative — that you can run inference locally, behind a network perimeter you control, on hardware in your own office, and the threat model they’re describing simply doesn’t apply. Not because you’ve mitigated the risk. Because the risk doesn’t exist in that architecture.
I sit with this for a while. The entire conference is organized around a question — how do we adopt cloud AI safely? — that assumes a premise I’ve already moved past. Not because I’m smarter. Because my constraints forced me there. Attorney-client privilege in criminal defense isn’t a compliance checkbox. It’s a constitutional obligation. You don’t get to say “our vendor promises they don’t train on your data” and call it due diligence when the client’s liberty is on the line. You build it yourself, on hardware you control, with network rules you wrote, and you verify.
The people I connect with aren’t always on the main stage. They’re in the hallways between sessions. And they’re the reason I came.
There’s a former judge from Texas who’s building his own data pipeline to train models to predict judicial rulings. There’s an immigration attorney who needs to batch-search a government database by A-number and country of origin for dozens of clients, and I sit down with Claude Code and build her a scraping tool in about twenty minutes that does what she’s been doing by hand for hours. There’s a conversation at lunch with sitting judges who are working on AI judicial working groups, and when I describe what I’m building, they lean in. Not because it’s impressive. Because it’s relevant to what they’re trying to figure out from the bench.
These are my people. I didn’t know that before this week. The ones who build things because the thing they need doesn’t exist. The ones who see a workflow problem and reach for code instead of a vendor contract. The ones who are in the practice, in the cases, in the system, and building from the inside because nobody is coming from the outside to build it for them.
The real value of this conference is that it put us in the same building. We found each other between sessions, over coffee, in the hallway after a talk, and realized we had more to share with each other than we expected.
And then it’s Friday afternoon and Nilay Patel takes the stage. The editor-in-chief of The Verge. He practiced law before he became a journalist, and he brings both frames to the question. He’s talking about how law firms are adapting to AI, what successful implementation looks like, what the ethical and practical implications are.
And somewhere in the middle of it, something lands.
The tool I’m looking for isn’t on the expo floor. It’s not in the startup pitch competition. It’s not on anyone’s product roadmap. The tool I’m looking for is the one I’m already building. Because it has to be built by someone who practices criminal defense, who processes evidence from Axon, who visits clients in custody, who argues motions that WestLaw’s AI confidently validates but no judge will hear. It has to be built by someone for whom the training data isn’t a commercial corpus — it’s the actual work of representing people the system is designed to process.
I have a workflow where I generate search strings for WestLaw myself, download the cases, unzip them, discuss the merits with a local model, draft the brief, tag it to the evidence, and argue it in court with a confidence that doesn’t come from a product’s marketing page. It comes from having touched every step. From having read every case myself. From having built the pipeline that connects the research to the evidence to the argument. That is a different kind of confidence than the one you buy. It’s earned.
And the attorneys in my office — the ones who develop instincts over thousands of cases about what matters in a deposition, what a witness’s hesitation means on camera, what a particular judge responds to — those instincts are heuristics. They can be articulated, tested, refined, and used to train models that continue watching and reviewing and flagging, so that the next attorney on the next case starts from a higher baseline. Not replacing judgment. Sharpening it. The attorney teaches the model and the model extends the attorney’s reach.
Saturday morning. The last day. Three ABA presidents sit on a stage — the current president, the immediate past president, and the president-elect — and they’re talking about the rule of law, the independence of the judiciary, and the resilience of the profession.
They talk about diversity. They talk about safeguarding public trust. They talk about what it means to be a “legal superhero” — not through slogans, but through integrity, adaptability, and decisive leadership. These are good words. I believe they mean them.
And then they talk about AI, and emerging technologies, and how the profession must embrace innovation.
And I want to ask: Which AI?
Because the AI on the expo floor downstairs — the AI in the Diamond and Platinum sponsor booths, the AI in the startup alley, the AI in the keynote slides — that AI is trained on commercial legal corpora. It is optimized for the workflows of firms that bill by the hour, that serve clients who can pay, that operate in practice areas where the market provides. It is built to make the profitable parts of law more profitable. And when it normalizes legal reasoning toward that center, when it generates briefs that sound like BigLaw briefs and research memos that read like corporate research memos and intake forms designed for clients who have bank accounts and email addresses, it does not diversify the profession. It homogenizes it.
The diversity the presidents are talking about — diversity of thought, of perspective, of experience in the legal system — is not going to come from that AI. It’s not going to come from any AI trained on data that doesn’t include the experience of representing someone who can’t afford a lawyer, in a system that would prefer they didn’t have one.
If you want AI that reflects the full range of legal practice, someone has to build it from inside the practices that the market ignores.
I’m on the plane home. The laptop is open. The evidence is still sitting on the Axon server.
I know what to do. I just need to do it.