Biglaw firms have all the resources in the world to tailor their AI solutions, and, in some cases, build their own. Professional tech staff can keep on top of the ever-changing landscape. Whenever multiple tools provide comparably secure and accurate results… the firm can just buy both and let the lawyers vote with their clicks.
The rest of the legal industry is still trying to get a handle on what a “model” means and wondering how they’ll ever figure it out without saddling themselves with an ongoing commitment to a system they’ve never been able to really try out. The comparison layer for legal AI is, at the moment, almost entirely competitor content marketing. The closest thing the industry has to a neutral scorecard is the Vals Legal AI Report, which is useful, but also required vendors to volunteer for evaluation.
Given that environment, imagine a three-lawyer plaintiff-side shop in Peoria trying to figure out if it even needs AI, let alone what AI vendor(s) to adopt.
Rachel See felt this pain listening to a labor law panel, realizing a gulf had developed between management-side lawyers able to set their staff to stay on top of this tech and plaintiff-side lawyers too busy keeping the lights on to figure out how to prompt. That’s when the lawyer who built eDiscovery infrastructure at the EEOC and the NLRB — using federal agency budgets to go toe-to-toe with firms boasting mega IT departments — and then spent a few years as senior counsel at Seyfarth and advising Fortune 100 companies on AI risk, started work on what would become Bells Up AI, which officially launched in July. Having seen both sides of the resource gap, she set to work building Bells Up.
Quite literally building it. She coded the minimum viable product herself before fleshing out her whole team.
Bells Up runs $99 a month for one lawyer, $94 a seat for three or more, with a free tier. But for that price, users don’t have to lock themselves into a model and can begin exploring across eight provider families — Claude, GPT, Gemini, Grok, Llama, DeepSeek, Mistral, Qwen — routed through enterprise APIs with all the necessary privacy notifications and contractual no-training agreements, on one invoice. And unlike some very deep-pocketed legal AI providers, it’s an invoice that a small firm can actually afford.
That’s not exactly fair… small firms might be able to afford to pay higher prices, but they worry about shelling out those prices for a product that might lock them into the wrong solution.
Clio’s data has roughly 71% of solos and 75% of small firms already using AI, with almost half of them on generic consumer platforms and more than half operating without any policy at all. The interest is there, but actually putting money behind a professional-grade solution tends to be where lawyers suffer executive action paralysis.
Bells Up counters with, “why choose?” The freedom to use multiple models is baked into the design. There’s a “Panel of Experts” mode to run one prompt through up to six models, with a synthesis layer that flags where they agree and where they split. The complaint analysis workflow sends a filing through multiple analytical orientations and produces strengths, weaknesses, a disagreement matrix, and a memo starting point. An AI model can do weird stuff. The consensus of multiple models will converge around the right answer. The wisdom of AI crowds, if you will.
And outliers aren’t necessarily wrong. A model that gets a burr in its algorithm over an issue can prompt the human to dig deeper into a good idea.
People still need to be the real thinkers. As the company’s homepage says:
Bells Up gives you safe, private access to the most powerful AI tools available. It doesn’t file anything on your behalf. It doesn’t replace your professional judgment. And it doesn’t guarantee that AI outputs are accurate. Verification is part of the work, just as it’s always been.
See explained that the baseline for a lawyer in 2026 simply cannot be “learn to write code.” Some tech-savvy lawyers will get into vibecoding and — if they have the right helping hands — that can deliver a great product. But for many lawyers, they just need something that gives them a working AI tool that’s not going to accidentally send all their client confidences to become the next training data set.
That seems right to me. There may only be a couple of providers building the AI guts, but meeting professional users where they are takes a village. Some will write code, some will hire consultants to write code for them, some will be true-believers in their preferred model, and some will just need an affordable and flexible user experience.
Joe Patrice is a senior editor at Above the Law and co-host of Thinking Like A Lawyer. Feel free to email any tips, questions, or comments. Follow him on Twitter or Bluesky if you’re interested in law, politics, and a healthy dose of college sports news.
