Precedent London 2026 | Legora Agent demo — Transcript

Full transcript

1. 0:01[music]
2. 0:05[music]
3. 0:07>> The Legora agent is the most powerful AI
4. 0:09ever built for legal work.
5. 0:11The agent is remarkable on its own, but
6. 0:14it's inside of the AOS that it becomes a
7. 0:17legal powerhouse.
8. 0:19It is a generational leap from assistive
9. 0:22AI to agentic execution, redefining what
10. 0:26a lawyer can achieve.
11. 0:28In the last 5 years, we have gone from
12. 0:30rule-based software to simple prompting,
13. 0:33advanced workflows, and now agents that
14. 0:35can carry on complex work for hours at a
15. 0:39time.
16. 0:40The agent autonomously plans, executes,
17. 0:43reviews, and completes legal work
18. 0:46end-to-end with human oversight
19. 0:48integrated throughout the system.
20. 0:52Until now, every legal AI tool required
21. 0:55the lawyer to drive the work forward.
22. 0:58When you stopped, the work stopped.
23. 1:01The agent changes that.
24. 1:03It works with you and for you,
25. 1:06even when you're not there. While you're
26. 1:08with clients, with your kids, playing
27. 1:10golf with Ludvig Åberg, watching a great
28. 1:12movie movie with Jude Law, or even while
29. 1:16you're sleeping.
30. 1:18You set the objective, you approve the
31. 1:20plan, and it executes using any tool it
32. 1:23needs inside of the AOS.
33. 1:26And to make the agent your own, skills
34. 1:28are the custom instructions that ex-
35. 1:30tend what the agent can do.
36. 1:32You can customize exactly how the agent
37. 1:35executes work across teams and practice
38. 1:37areas.
39. 1:38In other words, you can explain a
40. 1:40process to the agent in plain English
41. 1:43like you would to a new team member, and
42. 1:45it will execute work exactly to your
43. 1:47standards. And turns out, lawyers are
44. 1:49the best prompt engineers in the world.
45. 1:53Skills allow legal teams to encode their
46. 1:55knowledge and processes into the AOS.
47. 1:58The agent proactively loads the right
48. 2:01skill based on the task, and we offer a
49. 2:03library of skills out of the box for
50. 2:05common legal work. You can then take
51. 2:07them and customize it to your style,
52. 2:09processes, and practice areas.
53. 2:11And our legal engineers are there to
54. 2:13help every step of the way.
55. 2:16Now,
56. 2:18let's run through some use cases,
57. 2:20so you can see the agent in action.
58. 2:24And this is a full live demo.
59. 2:28So, here I've loaded up a matter.
60. 2:32I have added a couple of files into this
61. 2:35exciting project precedent.
62. 2:38We have an entire unstructured data
63. 2:42room.
64. 2:43Boo.
65. 2:45Not the thing I wanted to see on a
66. 2:46Friday afternoon.
67. 2:48I'm going to ask Legora to reorganize
68. 2:51this entire folder,
69. 2:53and every matter starts the same. So,
70. 2:55let's sort the documents into the
71. 2:56subfolders by each custodian, and let's
72. 2:59fire it off. So, without me giving any
73. 3:01files to the Legora agent, it's able to
74. 3:04find all of the files and the relevant
75. 3:06context inside of this matter.
76. 3:09And transaction teams know this. Data
77. 3:11rooms with thousands of messy files,
78. 3:13wrong folders, wrong names, trainees
79. 3:16spending days manually renaming
80. 3:18documents for a closing bible one by one
81. 3:21by one.
82. 3:22The prompt here asks to sort all of
83. 3:24these documents into the relevant folder
84. 3:26by custodian. And a custodian is
85. 3:28basically a person or entity that a
86. 3:30document belongs to, like the target
87. 3:31company.
88. 3:32Now, you can see the agent starting to
89. 3:34read every document inside of this data
90. 3:37room that I've given it access to. So,
91. 3:39it's reading through, and it I haven't
92. 3:41told it what the custodians are, it's
93. 3:43figuring that out by itself.
94. 3:46So,
95. 3:48now we can see here that it started to
96. 3:49read out all of them. It has found
97. 3:51somebody that attendees wasn't um
98. 3:54wasn't clear in.
99. 3:56And before it comes back to me with any
100. 3:58finished work product, it is going to
101. 4:00produce us with a plan.
102. 4:03This plan is very important because when
103. 4:04you're sending off these agents to do
104. 4:06work at a longer period of time, you
105. 4:09really want to make sure that you set
106. 4:10the right intent with the model from the
107. 4:12start up.
108. 4:14So, now Legora seems to have everything
109. 4:15it needs and it's going to put together
110. 4:17a full plan that we can review before it
111. 4:20goes up to work.
112. 4:22One of the um
113. 4:23behaviors that I've started noticing in
114. 4:25our engineering team is that when they
115. 4:28leave their computers to go to lunch,
116. 4:30they never close the top all the way
117. 4:32down. They'd leave just a little bit of
118. 4:34a gap because somebody's working under
119. 4:37there.
120. 4:38Now, Legora has produced us with a plan
121. 4:41for how it's going to go about this
122. 4:42folder reorganization.
123. 4:45Instead of just reading the short
124. 4:46version, I want to see the full plan.
125. 4:48So, inside of its own editor, it's
126. 4:50created a plan for me here. Step one,
127. 4:53create subfolders. Step two, move and
128. 4:55rename.
129. 4:56And then it's basically renaming and
130. 4:57moving based on every custodian that it
131. 4:59found. And it has its own to-do list of
132. 5:02what it needs to do.
133. 5:04I'm going to close the plan and I'm
134. 5:06going to say approved.
135. 5:08If we want to, it's at that moment we
136. 5:10can change and influence the plan before
137. 5:12Legora goes off to work. And now I
138. 5:14actually want to open this
139. 5:17um Oh, there we go.
140. 5:19Started to create one folder for each of
141. 5:21the custodians. There's zero files in
142. 5:24each of them as you can see and all the
143. 5:25documents are lying very, very messy
144. 5:27here. Oh, now six files in Vidal.
145. 5:31There we go. Renamed very nicely so we
146. 5:34know what we're looking at. Oh, there's
147. 5:36something again. It's so refreshing just
148. 5:38to watch it go to work.
149. 5:42It is quite soothing. You can just You
150. 5:44can just sit there and maybe have a bit
151. 5:46of a water.
152. 5:49This is what it must be like to be a
153. 5:51partner, right?
154. 5:55I'm just joking.
155. 5:56So, [snorts] so what's happened here is
156. 5:59um this new agent, we've given it access
157. 6:01to every tool or or capability inside of
158. 6:05the platform. So,
159. 6:06everything that you can do as a human
160. 6:09user, now the agent can do.
161. 6:11And turns out the agent is actually a
162. 6:13lot better at navigating this UI and
163. 6:16knowing what to do than a lot of people.
164. 6:18It's a lot easier to onboard the agent
165. 6:19than to onboard all of you guys.
166. 6:22So, here we go. It's found all of the
167. 6:24folders. It's done with its thing, and
168. 6:26it even flagged a duplicate document for
169. 6:28me.
170. 6:29And now we're done.
171. 6:31This obviously extends to data rooms and
172. 6:34with our data site integration to
173. 6:36thousands and thousands and thousands of
174. 6:37documents and different document types,
175. 6:39etc. The cool thing about the agent is
176. 6:41that it's really flexible, and if it's
177. 6:42unclear, it's going to ask you for
178. 6:44clarifying questions.
179. 6:47So, let's move on to the next task here.
180. 6:49In this case, we have some intercompany
181. 6:51agreements, and I want help visualizing
182. 6:54them in a diagram. So, one of the other
183. 6:57cool things about the agent is that we
184. 6:58can give it access to a lot of new tools
185. 7:02that the previous assistant was too
186. 7:04stupid to use.
187. 7:07Uh the old assistant had a really hard
188. 7:09time understanding
189. 7:11when it should use a certain tool to
190. 7:13solve a certain task.
191. 7:15But basically, the new agent has just
192. 7:18received this very, very long list of
193. 7:20tools that it's allowed to use in its
194. 7:22planning layer, and you can add your own
195. 7:24custom tools there. It's where skills
196. 7:26come in and things like MCP servers. An
197. 7:29MCP server is just a cool word on
198. 7:31LinkedIn for APIs for LLMs.
199. 7:34So, our agent here is using a pre-baked
200. 7:38skill that we've given it access to,
201. 7:40which is
202. 7:41uh to help it
203. 7:42uh data. So, we're effectively asking it
204. 7:45to do a tax uh
205. 7:46transfer pricing here. And this
206. 7:48intercompany agreements uh I'm asking
207. 7:51Legora to show basically which company
208. 7:54owns which. Haley's like, "Yes, this
209. 7:56looks good." So, different companies and
210. 7:59who owns which and where is business
211. 8:01flowing between these uh between these
212. 8:04um organizations.
213. 8:05And then I'm asking it for for three
214. 8:07things. I'm asking for a corporate
215. 8:09structure. I'm asking for a diagram. I'm
216. 8:11asking for a risk heat map that scores
217. 8:14each of the agreements across pricing,
218. 8:15documentation, structural, and audit
219. 8:17risk. And then I'm also asking for a
220. 8:19radar chart across this. So, there we
221. 8:20go.
222. 8:22Now it's created this uh pretty graph,
223. 8:24which is a little bit hard to see in the
224. 8:25display, but I promise the the lines are
225. 8:27there, so you can see.
226. 8:28And it's showing me with flags because I
227. 8:30I asked for that. It helps me present
228. 8:32things to my stupid CEO.
229. 8:34Um now it sees here like what is going
230. 8:37where, right?
231. 8:38And this is the cool thing about agents
232. 8:40because
233. 8:41the legal space when we arrived was so
234. 8:44full of these point-based solutions that
235. 8:46had, you know, motivated their existence
236. 8:49by solving a very very very narrow
237. 8:50problem. Turns out, this agents can
238. 8:52solve a lot of these narrow narrow
239. 8:54narrow problems out of the box even. And
240. 8:57when we build all of that together under
241. 8:59a single roof inside of the AOS,
242. 9:03that is when it comes to life.
243. 9:05Uh risk heat map here, fully
244. 9:07interactive. And this is also something
245. 9:09that I noticed. Uh I only survived at
246. 9:12McKinsey, but they're really good at uh
247. 9:14packaging their work, so they can charge
248. 9:16a lot for it.
249. 9:17And this is something I think a lot of
250. 9:19you in the room can get better at.
251. 9:21Receiving an Excel file or a boring Word
252. 9:23document doesn't get me going. And so,
253. 9:26instead sharing something exciting like
254. 9:28this in the portal, now now we're
255. 9:30talking.
256. 9:31And here we also asked it to to complete
257. 9:33the radar chart.
258. 9:34So, this is all exciting.
259. 9:36But um listening to the previous panel,
260. 9:40there was a dream about being able to
261. 9:42complete end-to-end due diligence. Oof,
262. 9:45that has been so hard. We have worked on
263. 9:47it, I tell you, for 3 years.
264. 9:49But, I think we finally got it.
265. 9:51So,
266. 9:53here I've given Legora a a bit of a
267. 9:55longer prompt, actually.
268. 9:56And I've also given Legora access to uh
269. 9:59all of the um
270. 10:01the uh what do we have here? The RFI,
271. 10:03the LOI, the um due diligence review
272. 10:06guidelines, and the entire due diligence
273. 10:09report template. And the reason I've
274. 10:12loaded this into a video is because it
275. 10:14takes 40 minutes to run.
276. 10:16And I didn't get 40 minutes.
277. 10:18So,
278. 10:19this is a full legal due diligence
279. 10:21report. We're asking it to complete a
280. 10:23corporate, uh IP, commercial,
281. 10:25employment, real estate, data
282. 10:26protection, and tax, all in the same go.
283. 10:29And as you can see, you have all the
284. 10:31unstructured virtual data room
285. 10:34files over here, and we've fired it off.
286. 10:36Okay. So, Legora's going to start. It's
287. 10:38going to start to go read all the
288. 10:39documents. It's again going to create a
289. 10:41plan.
290. 10:43This is a bit of a longer plan, because
291. 10:44in this case it has to use sub-agents
292. 10:47for many of the different phases here,
293. 10:49like employment, real estate, data
294. 10:50protection, tax, litigation, blah blah
295. 10:52blah blah blah blah. It is going to
296. 10:53launch a sub-agent, because watching
297. 10:55these things complete sequentially would
298. 10:57mean that this would take
299. 10:596 hours, right? So, basically the main
300. 11:02agent launches many sub-agents that goes
301. 11:05off and does their little part of the
302. 11:06work, kind of in a law firm.
303. 11:08Uh
304. 11:09and then when they're done, they're
305. 11:10like, "Hey, I'm done."
306. 11:12And they go [snorts] back to the main
307. 11:14agent, and they receive that work
308. 11:15product. So, now we can see it's like
309. 11:17spinning very quickly, and it's because
310. 11:19the video is sped up a little bit.
311. 11:21But, here the Legora agent has decided
312. 11:23to create a tabular review for each of
313. 11:25the different parts. So, it's the
314. 11:26corporate, it's the IP, it's the
315. 11:28employment. It's going through them.
316. 11:29But, now this is the
317. 11:31we did this video way too fast I'm
318. 11:33realizing.
319. 11:34Here here's the really important part.
320. 11:36The agent is like unclear and it's like
321. 11:38oh my god I found a bunch of issues. I
322. 11:40need your help. I need your judgement. I
323. 11:43am not fit to make these decisions for
324. 11:45you so I am involving you.
325. 11:48That is so important. Here it's like
326. 11:50asking us what we want and we've added a
327. 11:52bunch of answers here.
328. 11:54And I feel like this is also how you you
329. 11:56stay engaged because you can't just send
330. 11:58off a task like this and then just
331. 12:00expect the results to come back at the
332. 12:01end of the sausage sausage factory and
333. 12:04be completely done yet.
334. 12:07And here again it's like a lot of
335. 12:09different tasks now that it's going
336. 12:10through
337. 12:12and I really do mean it when I said that
338. 12:13this took 40 minutes to run. But it's
339. 12:15created a due diligence report for us.
340. 12:17So I'm opening that in full mode and
341. 12:19here we have all of the edits into our
342. 12:23existing template using our firm
343. 12:27guidelines and language just as we
344. 12:29wanted it. And it's gone through every
345. 12:31single piece. Now I don't get to take
346. 12:33any credit for this but our engineers do
347. 12:35so I think that's worth a round of
348. 12:36applause.
349. 12:39>> [applause]
350. 12:41[music]