Missed it, or want to go back over it? This is the recap of Lorna Khemraz's session on building a legal system your team can rely on. The hiring analogy that makes agents legible, the four questions that tell you what to automate, what supervision actually looks like once agents are live, and a conversation with Alistair Maiden on how GCs decide. The full recording and the slide deck are both here.
Lorna opened with a poll: how far would you trust an agent right now with one routine workflow? The room's answer was conditional and specific. If exceptions are escalated and someone owns the playbook, yes. She re-ran the same poll at the end. The answer held.
That condition is the whole session. Her argument is that legal already knows how to do this, because building a team and building an agentic system take the same discipline: you assess capability, you give context, and you build a structure that lets you trust the work at scale. What legal tends to skip is the third one. Teams get absorbed in procurement and then under-invest in what happens after something is deployed, which is exactly where the value is won or lost.
Every slide from the session: the tech stack shift, the four questions, the alignment challenges, the supervision cockpit and the playbook dashboard. Useful for briefing your own team.
Lorna's talk, the conversation with Alistair Maiden, and the audience Q&A. If you would rather skim, the written recap below covers the substance.
The mismatch Lorna started from: in most legal departments, highly trained and highly paid people spend most of their capacity on repetitive, rules-based work.
First-pass reviews of counterparty paper so sales can move. NDAs so partnership conversations do not stall. DPAs. Vendor contracts for procurement. Managing the legal inbox. Drafting from templates and precedent. None of it needs novel judgement every time, and judgement is what lawyers are hired for. Until recently there was no alternative, because every route to legal throughput ran through a lawyer.
Her framing of the gap cites Everlaw, which reports that 64% of in-house teams expect AI to reduce their reliance on outside counsel, while only 7% report total matter cost actually falling. The difference is between assisted work and delegated work. Prompting your way through a review is faster, but you are still doing it, and supplying the context every time. In an agentic setup the business emails legal as it always has, an agent triages against a routing matrix, the work happens overnight, and you sign in to a supervision queue and apply about twenty minutes of judgement.
Legal has been building teams forever. You know how to hire someone, brief them, supervise them, and work out over time how much you can trust them with. Lorna's argument is that an agentic system takes exactly the same discipline, and the three questions she asks when hiring are the three that matter for agents.
The third one is the afterthought. Legal operations gets absorbed in the procurement stage and forgets what happens after deployment. What matters then is consistent performance at scale, work that can be audited so you can see exactly what happened and why, and a system robust enough to hold up under scrutiny and real volume.
Lorna flagged this as the one thing to take away. Four questions, asked of any task that lands on your desk.
If you can explain it easily to a human, you can explain it to an agent.
Things that tend to pass all four: drafting full agreements from a pack of documents on your own templates, handling counterparty objections on first-party paper, reviewing third-party paper, answering the policy questions that recur, and triaging the shared legal inbox.
Writing playbooks is nobody's favourite task. Lorna's point is that most of the material already exists, scattered across work your team has already done.
Past redlines and escalation threads carry it. Comparing real redlines against the written playbook shows what is actually happening in practice, which tells you either that the playbook needs updating to match, or that positions have drifted and everyone should be making the same counter-proposal again. That is how a deployment starts: read the redlines, capture the micro-decisions, close the gaps and conflicts, then test the agent hard against your standards before it goes live. The most useful learning arrives after that, once real users are involved.
Work arrives in the legal inbox as it always has. An agent triages it against a routing matrix agreed with you, and sends it to the right agent, or to the right human if that is what the rule says. Completed work moves into a supervision queue, and what leaves carries a full audit trail: fileable, queryable, and showing who approved what.
The supervision cockpit is the side panel. You review escalations first, meaning high-risk items and anything your playbook said to loop a named person in on, then the items that genuinely need your judgement. Everything else the agent already had the context to handle. Cards map to plays in the playbook, so you can click through, sense-check the reasoning, and approve or edit. It is the same motion as reviewing a junior's first pass.
Because every micro-decision is captured as you supervise, the system produces data as a by-product: what is coming through the inbox, how much the agent handled alone, and, in the view Lorna called her favourite, how often a provision gets pushback and how often your fallback is accepted. That is the raw material for reducing negotiation friction and lightening the reviewer's load.
From a reactive function to a legal engine run like a system. The skills change accordingly: reading edit patterns to improve the playbook, managing agents, reading pushback patterns to inform negotiation positions, and using query flow to make resourcing and headcount decisions. Less execution, more auditing.
Alistair spent over a decade on legal transformation inside large corporate teams, founded SYKE and grew it into a global advisory business before its acquisition by Consilio, then led their global advisory practice. He joined Flank's leadership team this year and speaks to GCs every week.
What GCs are actually trying to solve. The same thing they always have: the gap between demand for legal work and the supply of people to meet it, with too much of the work repetitive and low value. It is a resourcing problem before it is a technology one.
What the board hears. Legal reads as a cost centre, so lower cost helps, but the sharper argument is velocity: matters resolved faster means the business sells and buys quicker. Every other function is already moving, so expectations of speed around legal are rising whether or not legal moves with them.
What earns trust. The upfront legal engineering, before anything is deployed. Flank writes it up as an agent design document that prescribes precisely how the agent operates. The customer signs it off, and both sides test against it. Alistair's current one for a procurement process runs to 27 pages. His advice on rollout: go department by department rather than moving a whole contracting process at once.
"There's as much refinement to the design document as there is to the agent's performance. It's normally us, the humans, that have got it wrong somewhere." The more comprehensive and explicit the instruction, the better the agent performs from the start.
Lorna closed with the practices she recommends once a system is running:
01. Keep a supervision matrix. Name who is accountable for each agent's performance, add a backup owner, be explicit about who owns playbook hygiene, and keep it current as people join and leave.
02. Treat codification as ongoing. Reverse engineer playbooks from whatever context exists, and add to them whenever the same question arrives twice.
03. Start small. Take the simplest workflow that passes all four questions, and prove the system there first.
Paul Lacey on putting agents into drafting workflows. A live walkthrough of decomposing a difficult template into a system, gathering deal facts the CRM never sees, and supervision that leaves a short list of gaps.
Share the full presentation with your team. Everything Lorna walked through, in one PDF.