Law://WhatsNext

🎙️ Peter Duffy is back for our quarterly deep dive into the biggest stories from his ever-popular Legal Tech Trends newsletter. It's been quite the quarter: a $500 million headline from Kirkland, frontier model companies falling over each other to claim legal, and the US government switching off a frontier model for the entire world. As Peter puts it: "What a bizarre world that we live in."
---
What we dive into:
Kirkland's $500m statement of intent — The announcement that rocked the market, and the nuance beneath it: Palantir (not an internal team) is helping building a platform, third-party procurement continues, and what counts as "investment" when partner hours bill at thousands of dollars? Either way, the proprietary layer marks that fascinating shift from service provider to potential product provider.
The frontier land grab for legal — Ex-Ironclad CEO Jason Boehmig joins OpenAI to lead its legal vertical, Freshfields signs a multi-year deal with Anthropic, and Perplexity launches Computer for Counsel. Peter shares his three-lens test for reading these partnerships — and why the real battle might be for the interface lawyers work through every day.
The fortnight the US switched off Fable — AI's first true geopolitical moment. A jailbreak, an export-control order, and the awkward detail that other models could reportedly do the same thing. We nerd out on the eerily prescient Europe 2031 scenario published the day before, and what it all means for sovereignty and model dependency.
Revolut Partners and the AI-audited panel — Out goes the static law firm panel; in comes a performance-based roster with quarterly reviews and internal AI tools to run RFPs and scrutinise advice and invoices.
---
Listen if: You want a hype-free tour of the quarter that was.
Rate, subscribe, comment, and share if you enjoyed this chat with Peter!
---
For more conversations at the intersection of law and technology, head to https://lawwhatsnext.substack.com/.
---
Show notes / links:
Peter's Legal Tech Trends newsletter: https://legaltechtrends.substack.com/
Europe 2031: https://europe2031.ai/
Legal Hops Amsterdam — Peter's quarterly meet-up (boat party included): https://www.linkedin.com/groups/12922521/

🎙️ This week we sit down with Will Lilley who has built something genuinely strange and uniquely clever: the Vibe Justice System, a working governance layer for AI-assisted work, complete with courts, a constitution, a Gazette, and 150+ recorded judgments that cite one another like real case law.
It began with a familiar frustration. AI agents make endless micro-decisions as they work, then forget them the moment a session ends.  Will calls this "drift," and his answer was to give the agent a "definition of correct" that is explicit, durable and self-enforcing.  The crucial distinction, as he explains, is that this isn't a bigger or better CLAUDE.md. A rules file tells an agent what to do, but the agent forgets to read it, and you could never write down every rule your workflow needs in the first place.
---
Will's system is principle-based, like common law: instead of front-loading every instruction, it lets the agent reason against principles and generatively build its own rule system — recording decisions as judgments, citing precedent, and aligning over time around what "correct" means for the project.
For our audience, Will frames the potential utility of the system as being capable of applying across three buckets of work: project work (any repo or directory: code, contracts, a litigation timeline, a due-diligence checklist), linear workflows that move through deterministic steps, and non-linear work, with litigation as the apex case where you only know the next move from the current one. 
---
Connect with Will Lilley — Product Operations at Clara | LinkedIn
Vibe Justice System — the full repo, including the Gazette, Acts and judgments: github.com/wlilley93/vibe-justice-system
---
If you enjoyed this conversation, please share it with someone who'd get something out of it — and if you have a moment, tell us what resonated, what didn't, and rate the show (it genuinely helps us reach more listeners and land great guests).
---
For more conversations at the intersection of law and technology, head to⁠⁠⁠⁠⁠⁠⁠⁠ https://lawwhatsnext.substack.com/⁠⁠⁠⁠⁠⁠⁠⁠.

67 agents, 155,000 lines of code, and not really a product — why one of the most insightful legal AI experiences right now is one you can see inside.
---
🎙️ This week we sit down with Antti Innanen — Finnish lawyer, law firm founder, legal designer and professor of practice — the mind behind Lavern, and one of the most prolific builders in legal tech, shipping open-source experiments at rap-album pace.
Lavern is the headline act — 67 specialist legal agents that debate, verify, and escalate, released free under Apache 2.0 not as a product but as an artifact you're invited to pull apart. As Antti puts it, it's the first agentic legal tool where you can actually pop the hood and look inside.
---
This is a conversation about building in the open. Antti walks us through Lavern live — the intake modes, the 67-agent roster (Easter eggs included), the verification loops — and uses it as a way into bigger questions: why context is the whole game in legal AI, what it costs to run agents at the frontier, why he's closing a beloved legal-design business even as its ideas finally become scalable, and why "f***ing around" with AI might be the most responsible thing a lawyer can do right now.
---
Connect with Antti Innanen — Lawyer, legal designer & builder | Lavern (GitHub repo, 100% open) | Find him on LinkedIn
---
If you enjoyed this conversation, please share it with someone who'd get something out of it — and if you have a moment, tell us what resonated, what didn't, and rate the show (it genuinely helps us reach more listeners and land great guests).
One thing we'd love to hear: if you've forked or remixed Lavern — or taken it somewhere Antti didn't expect (access to justice, strategy, anything) tell us. We'd love to see it!
---

🎙️ This week we sit down with Shawn Curran, CEO of Jylo — for a conversation that starts where a lot of legal teams and law firms are quietly stuck: Now that almost anyone can build software, should they?
Shawn is unusually well placed to answer. He spent the best part of two decades as a technologist inside private practice — McGrigors, Latham & Watkins, Freshfields, then head of legal tech and later director at Travers Smith, before spinning Jylo out of the firm two and a half years ago. 
---
He's watched the industry crawl from paper files 📄 to systems of record 🗄️ to systems of productivity ⚙️ and now to what he calls "systems of intelligence" 🧠. So when he says the barrier to building software is "literally on the floor" 📉, it lands with some authority. And he's thrilled about it. The open-source wave of lawyer builders democratising access to their products — Will's MikeOSS, Antti Innanen's Lavern — is, to Shawn, a long-overdue correction. Now the person who actually understands the work describes it in plain English and watches it take shape.
---
This discussion plays out against a legal technology market that refuses to sit still. One week it's a new open-source release; the next, Kirkland & Ellis commits half a billion dollars to its own platform; the week after, a frontier lab ships a legal plugin. Shawn admits he loses a faintly ridiculous amount of time to "is this Jylo for free?" emails. 
Which brings us to Shawn’s stand out observation from our conversation - we talk endlessly about ROI; Shawn would rather we talked about ROT (return on token). If a firm spends thirty grand vibe-coding something nobody ever opens, it hardly matters that it would have cost two hundred grand in engineers; it's still money set on fire. 
"Return on token. What's the return on token? 'Rot.' … There's a lot of rot out there." 
Shawn warns that as token prices climb rather than fall, the honeymoon phase of trying everything is going to give way to harder questions about what's actually worth the spend. 
---
Connect with Shawn Curran — CEO of Jylo 
---
If you enjoyed this conversation, please do share it with a colleague or community wrestling with the same questions — and if you have a moment, tell us what resonated, what didn't, and rate the show. It genuinely helps us grow the audience and land great guests.
---
For more conversations at the intersection of law and technology, head to⁠⁠⁠⁠⁠⁠⁠ https://lawwhatsnext.substack.com/⁠⁠⁠⁠⁠⁠⁠.

🎙️ This week we sit down with Stephanie Dominy — General Counsel and Head of Ops at Tessl.  Stephanie has practised law for nearly three decades, but at Tessl she hires across the whole organisation, including against one of its operating principles: "Be AI Native."
What begins as "how do you spot an AI-native engineer?" becomes "how do you hire an AI-native lawyer?" — and then a deeper question still: if AI can already draft, research and build, what is the irreducibly human part of legal work?
Stephanie's answer is taste and judgment. She explains why Tessl keeps iterating on how it hires, most recently reverse-engineering its interviews (inspired by Sierra's AI-native interview framework). From there we touch on cognitive surrender, the commoditisation of knowledge, legal training pathways, and her observation that those with humanities backgrounds are often better equipped to interrogate AI than purely data-driven minds.
---
Tom is equally sceptical of organisational efforts to adopt this technology. He points to Claude's Constitution — the document Anthropic uses to steer a probabilistic model through competing demands via a clear order of precedence — and argues it isn't so different from a legal constitution, and is a surprisingly useful mirror for any company trying to go AI-native. (It's also humbling, he suggests, to work out where you sit in the "model stack.")  His argument: every organisation is made up of teams with competing priorities, so be deliberate about where to go all-in. His example — the overzealous sales team optimising for deals while legal supercharges risk prevention. Today those incentives collide only occasionally, and human leaders resolve them. But as each team is amplified by AI, the collisions get faster, and their resolution is quietly outsourced to the model — piling on competing instructions that degrade its performance. 
---
Tom and Stephanie are taking this one to the stage — Wednesday 3 June 2026, at Crafty Fest, Regent's University London. Crafty Fest is Europe's largest festival for the in-house legal community, and their session picks up exactly where this episode leaves off.
---
Connect with Stephanie Dominy, General Counsel & Head of Ops at Tessl.

And the open question Stephanie left us with: if you've built something that actually works for testing taste and judgment — an interview question, a practical test, anything — we want to hear it. Bring it to Crafty Fest, or drop it in the comments 👇
---
For more conversations at the intersection of law and technology, head to⁠⁠⁠⁠⁠⁠ https://lawwhatsnext.substack.com/⁠⁠⁠⁠⁠⁠.