Five stories. One idea underneath all of them.

Last week the model under your legal AI turned out to be rented and revocable. This week, the obvious next thing said out loud: if the model is a commodity, the product is the harness around it. A Legal Nodes study ran the same Claude Opus 4.8 three ways and got three different results at very different costs. The capital agrees — it’s funding the back office (Jupus’s AI secretary) and the training layer (BARBRI buys Lega), not a smarter brief. Adoption is already ahead of the slides — a third of professionals on shadow AI — and a managing partner argues the juniors come out ahead, not erased. The model isn’t the moat. The harness is.

Same model, three setups, three results — the harness decides

A Legal Nodes study ran the same Claude Opus 4.8 three ways — plain chat, Cowork with a legal plugin, and an open-source harness — across 40 data-protection and DORA tasks. Same model, materially different answers. The scaffold around it, not the weights inside it, set the quality.

Cost moved even more than quality. The open-source harness ran roughly 60% cheaper than Claude chat at only slightly lower quality. The takeaway lands hard for any firm shopping by model name: you are not buying a brain, you are buying the workflow, retrieval and tooling wrapped around it.

[ Artificial Lawyer · The Legal AI Scaffold Changes Everything — Claude Study ]

Europe funds the AI secretary, not the AI lawyer

Cologne-based Jupus raised a €13M ($14.8M) Series A led by Semapa Next, with NRW.BANK, Acton Capital and High-Tech Gründerfonds joining. The product is unglamorous on purpose: an AI service that answers calls, runs intake and drafts routine documents for small and mid-sized firms.

Note where the capital points. Not at a smarter brief-writer for the partner, but at the back office — the calls, the intake, the paperwork that never made it into a vendor demo. The harness firms actually need first is the one that clears the desk.

[ Tech.eu · Jupus Raises €13M to Power the Next Generation of AI-Driven Law Firms ]

BARBRI buys Lega to sell the skill, not just the tool

Legal-education company BARBRI acquired Lega, the gen-AI adoption startup founded by Christian Lang, who joins as Head of Innovation. Terms were undisclosed. The plan: experiential AI training — workshops, simulations, hands-on labs — for students, firms and in-house teams.

The tell is that adoption is now a product worth buying outright. The tools exist; what firms lack is people who can drive them well. If the harness is the product, the skill to operate it is the next thing someone monetizes.

[ Artificial Lawyer · BARBRI Buys Lega — Christian Lang Interview ]

A third of professionals are on shadow AI while firms stall

A Thomson Reuters survey of 1,816 professionals found 33% using unapproved ‘shadow’ AI tools, and 25% saying they want to leave within two years over the inadequate AI at their firm. Meanwhile, nearly half of senior leaders expected no real talent pressure from AI for at least three years.

That gap is the story. The people doing the work have already adopted — off the books, on whatever works — while leadership plans for a slower timeline that has already passed. Govern the harness your people are using, or they’ll keep using one you can’t see.

[ Artificial Lawyer · AI Adoption Often Slow, Chaotic, TR Survey Finds ]

The contrarian case: AI gives juniors better work, not no work

Against the now-standard worry that automation guts the junior pipeline, Gary Wingens, chair of Lowenstein Sandler, makes the opposite case. Cheaper, faster legal work expands demand: matters that were never economic become worth doing, which needs more juniors, not fewer.

His frame is AI as co-pilot — it clears the grunt work so an associate reaches higher-value work sooner and learns faster than prior cohorts could. Whether he’s right depends entirely on the harness: a tool that drafts for the junior trains them; one that drafts instead of them does not.

[ Bloomberg Law · AI Won’t Replace Junior Lawyers. It Will Give Them Better Work ]

For two years the legal-AI argument was about which model is smartest. This week it quietly became the wrong question. The same model, in three different harnesses, is three different products at three different prices — so the work that matters is the work around the model: the workflow it runs in, the back office it clears, the people trained to drive it, and the juniors it either teaches or replaces. The model is rented. The harness is yours to build. – The Editors▌

Exhibit AI reports on the AI industry. We do not provide legal advice. Sponsorships are disclosed and never shape coverage. Issue N°010 published WED 24 JUN 2026.