When I wrote last week about BARBRI\u2019s acquisition of Lega, I had not yet had the chance to speak with Lega founder Christian Lang, who was traveling. So I closed that post with what was admittedly conjecture on my part \u2013 that the large-language-model governance product that defined Lega when it launched in 2023 was [\u2026]<\/p>\n
When I wrote last week<\/a> about BARBRI\u2019s acquisition of Lega, I had not yet had the chance to speak with Lega founder Christian Lang, who was traveling. So I closed that post with what was admittedly conjecture on my part \u2013 that the large-language-model governance product that defined Lega when it launched in 2023<\/a> was now going away, in favor of the AI-training business that the company had developed more recently.<\/p>\n When I finally caught up with Lang this week, I learned that conjecture was only half right. While the governance product is not going away, exactly, it is taking on a new role, no longer focused on testing and building AI, but rather on learning and developing fluency in AI.<\/p>\n \u201cLega the platform is such a tremendous teaching tool and platform to support the workshops, to support the hackathons, to support AI innovation labs,\u201d Lang told me. \u201cIt is the perfect infrastructure for that. So we\u2019re using it.\u201d<\/p>\n So while the company is keeping the technology, it is changing the job it does. \u201cWe\u2019re not going to go out and try to sell the platform as the place where you build production AI solutions for your frontline teams anymore,\u201d he said. \u201cIt\u2019s going to be about: this is where you learn and test and try and develop the AI fluency, kind of further up the value chain.\u201d<\/p>\n Lang drew a distinction between the two architectural halves of Lega. One is the \u201clab,\u201d the sandbox where users bring resources together, compare models and test prompts. The other is the \u201cgateway,\u201d the layer through which traffic flows and where the platform builds audit trails and handles compliance scanning, observability and analytics.<\/p>\n In the product\u2019s original incarnation focused on firm governance, the gateway existed largely to enforce outside-counsel guidelines and keep a record. Now, under BARBRI, the same technology becomes a measurement instrument.<\/p>\n \u201cIt\u2019s actually the analytics we\u2019re doing through the gateway \u2013 that\u2019s exactly the feedback loop you need when you\u2019re trying to test people and understand how well they\u2019re doing and where are they on that fluency arc,\u201d he said. \u201cThere\u2019s probably not any piece of the platform that isn\u2019t integral to what we\u2019re doing at BARBRI, but it will be positioned slightly differently.\u201d<\/p>\n Having said that, Lang added that the platform itself is not sacred and that the technology could evolve over the longer term. \u201cIf we end up two years from now running everything on top of Claude Cowork as the platform, great, that\u2019s what we\u2019ll do.\u201d But for now, he said, Lega\u2019s own infrastructure remains the better fit for what the workshops require.<\/p>\n As I said in my story last week about the acquisition, although Lega launched in 2023 with a focus on LLM governance, it had evolved to increasingly focus on helping legal professionals develop AI fluency through practical, experiential learning.<\/p>\n The experiential side of Lega \u2013 the part, no doubt, that most interested BARBRI \u2013 predates the deal by about 18 months. Lang said the company ran its first hands-on AI workshop at the Legal AI Conference in New York roughly a year and a half ago, initially \u201cjust for fun,\u201d on the theory that there was no reason to limit experiential learning to the firms that licensed the platform.<\/p>\n From there it expanded to innovation teams, partner retreats, innovation summits, law schools, and some corporate legal departments.<\/p>\n \u201cThese are basically almost like high-end cooking classes for AI,\u201d Lang said, where Lega defines a realized use case, builds the materials for it, and then puts participants in a room, in teams, on machines.<\/p>\n He described the workshops as a three-part journey. Participants start in the model-comparison sandbox \u2013 \u201cpick their weapon,\u201d as he put it, choosing the model they want to build on.<\/p>\n They then work through what it means to build a context picture for a model, which Lang said is where the black box gets opened up: how retrieval augmentation actually works, what dials exist, what happens when you call tools and MCP servers, and where that content lands in the prompt.<\/p>\n In the third phase, each team builds a working prototype of an app or agent by the end of the day.<\/p>\n That progression, Lang argued, is what moves people past surface-level \u201cAI literacy\u201d and into what he calls fluency \u2013 the point at which users can reason analogically about what the technology can do in their own work.<\/p>\nRepurposing the Sandbox<\/strong><\/h3>\n
\u2018High-End Cooking Classes for AI\u2019<\/strong><\/h3>\n
Stop \u2018Defending the Past\u2019<\/strong><\/h3>\n