WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=Btk8wDUVs74 NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/Btk8wDUVs74.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:01.309 --> 00:00:03.309 [music] s2 00:00:14.480 --> 00:00:16.000 Um, hey everyone. s3 00:00:16.000 --> 00:00:17.760 Uh, we are super excited to be here. s4 00:00:17.760 --> 00:00:19.119 Thanks for having us. s5 00:00:19.119 --> 00:00:27.119 Uh, today we're going to talk about how we shift monday.com from a system of record uh, into a system of context. s6 00:00:27.119 --> 00:00:32.079 And honestly the title tell the whole story uh in just a single line. s7 00:00:32.079 --> 00:00:45.120 Uh for decades we build uh software that record what's happened um every task, every document, every message um every status update um just put into the record. s8 00:00:45.120 --> 00:00:48.640 What we want to talk today is like take it step further. s9 00:00:48.640 --> 00:00:53.840 Uh we want software that actually understand the connection between them. s10 00:00:53.840 --> 00:00:59.600 So I want to start from a simple question that each one of us ask himself every morning. s11 00:00:59.600 --> 00:01:02.480 Um what should I focus on right now? s12 00:01:02.480 --> 00:01:14.640 Um it sounds almost trivial but uh to be honest with yourself if you ask your agent whether Gemini GPT or even cloud um if you ever typed this question s13 00:01:14.640 --> 00:01:19.520 you probably got list of bullets um not related to each other. s14 00:01:19.520 --> 00:01:24.080 um list of like item dressed up like a a confident paragraph. s15 00:01:24.080 --> 00:01:30.000 Um but but it's not really um um connected to what you're working on. s16 00:01:30.000 --> 00:01:34.400 Actually, I tested last week and uh Cloud asked me to go to the gym. s17 00:01:34.400 --> 00:01:38.400 I don't know if it's a compliment or not, but uh this is what he suggested. s18 00:01:38.400 --> 00:01:47.439 Um and and what really make it so frustrating is like your assistant um have all this data. s19 00:01:47.439 --> 00:01:54.240 It has all the boards, the tasks, the emails, the Slack messages, everything that you ever touched if you connect it. s20 00:01:54.240 --> 00:01:57.600 Um, but it still can't really answer it. s21 00:01:57.600 --> 00:01:59.920 Um, it has all the data. s22 00:01:59.920 --> 00:02:04.000 Um, but it has zero understanding. s23 00:02:04.000 --> 00:02:06.399 Um, so this is the real challenge we are facing. s24 00:02:06.399 --> 00:02:09.200 Um, this is the heart of the entire talk. s25 00:02:09.200 --> 00:02:12.560 The problem was never the missing of data, the retrieval. s26 00:02:12.560 --> 00:02:15.120 The problem is like the missing understanding. s27 00:02:15.120 --> 00:02:21.040 uh those are two totally different things and almost everyone mix between them. s28 00:02:21.040 --> 00:02:25.440 Understanding is the word that we're going to focus the entire the entire talk. s29 00:02:25.440 --> 00:02:30.160 Not context, not memory, not retrieval, understanding. s30 00:02:30.160 --> 00:02:32.560 So um quick about ourselves. s31 00:02:32.560 --> 00:02:33.599 My name is Tormer. s32 00:02:33.599 --> 00:02:34.480 My name is Omri. s33 00:02:34.480 --> 00:02:35.440 This is Tor. s34 00:02:35.440 --> 00:02:45.920 Um uh we both engineering manager at monday.com working on exactly uh the problem that we going to talk about a little a little bit context about where we coming from. s35 00:02:45.920 --> 00:02:48.800 Uh monday.com is a global software company. s36 00:02:48.800 --> 00:02:54.400 Uh we build a work platform uh used by hundreds of thousands of teams. s37 00:02:54.400 --> 00:03:00.879 Um and the part that really mattered for the talk is that Monday is where the work um lives. s38 00:03:00.879 --> 00:03:04.879 uh we help companies to doing the work not just like saving records. s39 00:03:04.879 --> 00:03:12.800 Every project, every task, every decision, every meeting, the notes, the action item, everything logs logged into the system. s40 00:03:12.800 --> 00:03:18.080 And our mission has always been to help the teams to achieve their business outcome. s41 00:03:18.080 --> 00:03:26.000 Whether you are a salesperson, so we want to help you like create an NSDR agent that call to your prospects and help you sell the product. s42 00:03:26.000 --> 00:03:32.080 And if you are like a finance team or marketing we want to help you uh with the research. s43 00:03:32.080 --> 00:03:40.959 So each one of the discipline we want you help you to doing the job and together we have like four big bats on the platform. s44 00:03:40.959 --> 00:03:44.400 We have like Monday sidekick we're going to talk about it mostly today. s45 00:03:44.400 --> 00:03:47.360 Uh Monday vibe if you want to build your own software. s46 00:03:47.360 --> 00:03:57.439 Monday agent if you want to create your own agent into the platform and like if you want some more deterministic flow we have like Monday workflows but today we're gonna focus on psychic s47 00:03:57.439 --> 00:04:09.360 um psychic is your intelligence AI personal assistant that's understand your work think and execute with you like a bird on your shoulder uh he know you he know your business s48 00:04:09.360 --> 00:04:20.959 um it help you with your work on every aspect uh he work the way you do uh with your tones Um um it's keep you totally in control but like s49 00:04:20.959 --> 00:04:24.800 putting all these bullets and uh building all these promise is really hard. s50 00:04:24.800 --> 00:04:30.080 Um so let's talk about why it's hard and there are three points for that. s51 00:04:30.080 --> 00:04:36.400 Um first like picture one assistant sitting on top of absolutely everything. s52 00:04:36.400 --> 00:04:40.240 You have your u your u slack messages. s53 00:04:40.240 --> 00:04:42.720 You have all the notes for your meetings. s54 00:04:42.720 --> 00:04:53.919 um absolutely everything and it's beautiful and overwhelming at the same time because as it stand like there's wall of records everywhere. s55 00:04:53.919 --> 00:04:59.600 Um so what why it's making so hard to connect between them. s56 00:04:59.600 --> 00:05:02.000 There are three reason for that. s57 00:05:02.000 --> 00:05:05.040 One is something that we call the agent gap. s58 00:05:05.040 --> 00:05:11.039 Your agent every agent is really sharp and doing task um when you when he know what to do. s59 00:05:11.039 --> 00:05:14.960 Um but he sometimes lost find them find the problem. s60 00:05:14.960 --> 00:05:22.800 Um if I ask my agent like please help me draft and and and reply to like the escalation from a customer he nailed it. s61 00:05:22.800 --> 00:05:24.080 He know how to do it. s62 00:05:24.080 --> 00:05:25.120 He take the context. s63 00:05:25.120 --> 00:05:38.720 I like doing the work but asking him what should I focus on first he guess because he doesn't understand what is my priority who am I it doesn't matter if you have a memory or something s64 00:05:38.720 --> 00:05:54.960 he still don't know what is the problem the second problem is like um we have the records but we don't have the meaning um a log never said what it mean let's take an analogy from our s65 00:05:54.960 --> 00:05:58.080 Let's say that you have one line of code in your GitHub. s66 00:05:58.080 --> 00:06:01.199 You look at a code, maybe there is a comment on top of it. s67 00:06:01.199 --> 00:06:02.000 Nice. s68 00:06:02.000 --> 00:06:06.319 But you don't really understand why someone wrote this line of code. s69 00:06:06.319 --> 00:06:08.639 Today, no none of us writing code. s70 00:06:08.639 --> 00:06:12.639 But somewhere in the past, someone wrote this code line of code. s71 00:06:12.639 --> 00:06:20.880 And if you really want to know, you can go to get blame and understand uh from the commit log what why why someone do it. s72 00:06:20.880 --> 00:06:26.080 But if you want to go farther, you can go to the PR and maybe read the description. s73 00:06:26.080 --> 00:06:38.080 And if you really want to go hard, you can go to your Monday board and see which PR connected to this item and understand that this line of code came because some customer complain about something. s74 00:06:38.080 --> 00:06:41.039 So this is what we're trying to build. s75 00:06:41.039 --> 00:06:51.600 And the third point that is a bit challenging as well is like um it's really hard to build ahead of time um at runtime um the meaning of things. s76 00:06:51.600 --> 00:06:53.680 Um it's really too late. s77 00:06:53.680 --> 00:06:59.919 Um you simply can't do something like that the moment someone asked the question. s78 00:06:59.919 --> 00:07:04.880 So understanding understanding of the context has to be ahead of time. s79 00:07:04.880 --> 00:07:10.240 Um you need to build it much before someone asks the question. s80 00:07:10.240 --> 00:07:13.680 Um and this is why we have built what we are building. s81 00:07:13.680 --> 00:07:16.000 We are building uh the Monday world model. s82 00:07:16.000 --> 00:07:18.479 This is what we called the Monday world model. s83 00:07:18.479 --> 00:07:28.160 Help you um understand why this matter um how to help you, who you are, when and and what's not to do. s84 00:07:28.160 --> 00:07:31.280 Um it's the context that follow your work. s85 00:07:31.280 --> 00:07:38.160 Um he understand who you are and it's simply not a bigger prompt. s86 00:07:38.160 --> 00:07:40.800 uh it's not a longer uh context window. s87 00:07:40.800 --> 00:07:45.840 It's a totally different um from from what we know until now. s88 00:07:45.840 --> 00:07:49.039 First before like Tor going to talk about how we build it. s89 00:07:49.039 --> 00:07:51.520 What it's not it's not a retrieval problem. s90 00:07:51.520 --> 00:08:01.120 The problem never been getting the data we don't we have all the data we have all the connection to what all the provider that we want to get all the MCPS. s91 00:08:01.120 --> 00:08:04.879 The problem is really to understand how it works. s92 00:08:04.879 --> 00:08:08.560 Understand how each one of these entity connected to each other. s93 00:08:08.560 --> 00:08:10.960 So go ahead. s94 00:08:10.960 --> 00:08:12.160 Thank you. s95 00:08:12.160 --> 00:08:13.759 Hi everybody. s96 00:08:13.759 --> 00:08:17.199 Um so what's the data model? s97 00:08:17.199 --> 00:08:28.400 We collect thousands of data points on the user every item status change their activity log messages and meetings and construct three things the agent can resone over. s98 00:08:28.400 --> 00:08:36.399 The first is how the user's work is structured, their key entities and relationships and connections between them. s99 00:08:36.399 --> 00:08:42.080 What depends on what, how a message in Slack connects to a task, who's blocking whom. s100 00:08:42.080 --> 00:08:54.399 The second is a current snapshot, live signals over those entities, what's overdue, what's critically urgent right now, which co-workers you've been actively working with, and why. s101 00:08:54.399 --> 00:08:57.360 The third is what we can learn about the user over time. s102 00:08:57.360 --> 00:09:04.720 There are decisions and outcomes, work patterns and cadences distilled into a durable profile. s103 00:09:05.040 --> 00:09:06.959 So how do we build that data model? s104 00:09:06.959 --> 00:09:11.040 Um we use two engines running on different time windows and schedules. s105 00:09:11.040 --> 00:09:20.720 A slow engine that runs on a long time window and learns the user and their work and a fast engine that reads what's happening right now and how it affects the user's work. s106 00:09:20.720 --> 00:09:24.959 One knows you and the other one knows your day. s107 00:09:24.959 --> 00:09:26.800 First the slow engine. s108 00:09:26.800 --> 00:09:42.640 It takes as context users activity over weeks and minds it for patterns and the type of persona the user is their routines their work rhythm who they collaborate with their main goals and current projects. s109 00:09:42.640 --> 00:09:48.880 Those patterns get distilled into a durable profile and every time a profile holds it's reinforced. s110 00:09:48.880 --> 00:09:55.600 This engine tries over time to learn who exactly the user is and how they work. s111 00:09:56.000 --> 00:09:58.160 The fast engine is the opposite. s112 00:09:58.160 --> 00:10:06.080 It takes as context a short recent window and recomputes a set of live signals over the user's current state. s113 00:10:06.080 --> 00:10:11.440 What server do, what's suddenly urgent, which co-workers you've been pulled in with. s114 00:10:11.440 --> 00:10:16.320 This engine tries to understand your day and updates frequently. s115 00:10:16.480 --> 00:10:19.680 Uh this split isn't something we invented. s116 00:10:19.680 --> 00:10:22.399 It's present in two totally different fields. s117 00:10:22.399 --> 00:10:26.800 In neuroscience, this split is referred to as complimentary learning systems. s118 00:10:26.800 --> 00:10:31.279 And in data processing architecture, it's referred to as a lambda architecture. s119 00:10:31.279 --> 00:10:36.720 We apply the same concepts to how we construct the agent's data model. s120 00:10:37.279 --> 00:10:39.440 Our brain uses the same split. s121 00:10:39.440 --> 00:10:50.399 Uh every important experience gets captured instantly by the hippocmpus and over time the new cortex distills those into durable lessons. s122 00:10:50.560 --> 00:10:54.640 In data infrastructure, the same there's the same split. s123 00:10:54.640 --> 00:11:07.040 A fast speed layer over a recent real-time window and a slow batch layer over the full history that gets recomputed and the two are merged into a single served view. s124 00:11:07.040 --> 00:11:13.920 Two different fields landed on the same idea and that's what we're trying to apply to our data model. s125 00:11:14.240 --> 00:11:16.720 So how does it all come together? s126 00:11:16.720 --> 00:11:19.760 We collect data from everywhere our users work. s127 00:11:19.760 --> 00:11:27.279 uh Monday, Slack, emails, calendar, and we turn it into data structures, signals, and patterns. s128 00:11:27.279 --> 00:11:32.480 Both engines premputee on top of that offline and ahead of time. s129 00:11:32.480 --> 00:11:42.880 And when a user engages with Sidekick, a thin slice of logic is recomputed for recent activity, and the entire context is served to the agent. s130 00:11:42.880 --> 00:11:48.720 Sidekick can then decide when and how to traverse and retrieve context from the data model itself. s131 00:11:48.720 --> 00:11:53.120 and it's primed to reason on top of it. s132 00:11:53.120 --> 00:11:57.040 And building it this way gives us two behaviors out of the box. s133 00:11:57.040 --> 00:11:59.440 The data model, it's resilient. s134 00:11:59.440 --> 00:12:03.839 Sources are isolated so a bad feed can't break the rest. s135 00:12:03.839 --> 00:12:13.680 And the thin layer of logic that runs at serve time verifies part of the context against live data while the rest falls back to the last verified context. s136 00:12:13.680 --> 00:12:18.079 So it degrades gracefully, but it doesn't fail. s137 00:12:18.079 --> 00:12:21.120 Second, it actually understands the urgency of facts. s138 00:12:21.120 --> 00:12:29.839 It's able to understand when and how it should be proactive and notify the user and when it should stay silent. s139 00:12:30.560 --> 00:12:33.200 And the crucial part is that it compounds. s140 00:12:33.200 --> 00:12:38.560 Every day the data is captured, the layers fill in and the profile sharpens. s141 00:12:38.560 --> 00:12:43.279 And adding a new data source is deliberately cheap and only contributes. s142 00:12:43.279 --> 00:12:45.440 So the surface only grows. s143 00:12:45.440 --> 00:12:47.760 The more it sees, the more it understands. s144 00:12:47.760 --> 00:12:51.040 And the more it understands, the more you can lean on it. s145 00:12:51.040 --> 00:12:52.480 And the data model is unique. s146 00:12:52.480 --> 00:12:55.920 It's unique to how you work. s147 00:12:56.240 --> 00:12:58.720 We're not pretending this solves everything. s148 00:12:58.720 --> 00:13:04.560 Uh the model itself is always trailing the actual live world. s149 00:13:04.560 --> 00:13:11.440 New users have no reliable data to reason from yet, and signals have our own biases built in. s150 00:13:11.440 --> 00:13:16.160 So the hardest part is actually telling the important parts from the noise. s151 00:13:16.160 --> 00:13:22.800 But this is an architectural design that we can enrich and test and improve over time. s152 00:13:22.800 --> 00:13:24.959 Thank you. s153 00:13:25.659 --> 00:13:27.659 [applause] s154 00:13:28.320 --> 00:13:32.560 So back uh back back to the original uh question that we had. s155 00:13:32.560 --> 00:13:34.240 What should I focus on right now? s156 00:13:34.240 --> 00:13:35.760 Remember the the question. s157 00:13:35.760 --> 00:13:38.079 So now Psyche can really answer that. s158 00:13:38.079 --> 00:13:39.120 Uh let's see how. s159 00:13:39.120 --> 00:13:50.800 So first like Thomas said we collect all the data point we called it breadcrumbs for example the board that I working on right now um all the emails from the past uh months s160 00:13:50.800 --> 00:14:03.440 um the transcript for the meeting all the action item that I got from those meeting and we processing them offline um we see a full picture of your calendar this is what help us to understand the pattern so if you have like bi-weekly s161 00:14:03.440 --> 00:14:17.600 with your VPs I understand that tomorrow you have another meetings and even slack messages from the last few days to understand if you say something in the Slack that's relevant that you didn't say in your daily uh standup with the team. s162 00:14:17.600 --> 00:14:23.040 We took all this data and we uh process them either fast or slow. s163 00:14:23.040 --> 00:14:26.639 Um the slow build some kind of like profile on you. s164 00:14:26.639 --> 00:14:30.480 So you can see it on the top like combin is an engineering manager. s165 00:14:30.480 --> 00:14:33.519 We I'm working on two projects psychic and notaker. s166 00:14:33.519 --> 00:14:34.800 I'm from Tel Aviv. s167 00:14:34.800 --> 00:14:40.000 And this is how many hours I have per day and like how I work every day. s168 00:14:40.000 --> 00:14:43.199 And the fast is something that's helped me to understand what is the action item. s169 00:14:43.199 --> 00:14:46.720 I have like three commitment that I promised to other people. s170 00:14:46.720 --> 00:14:50.399 I have to to reply to a VP that sending email and I didn't. s171 00:14:50.399 --> 00:14:56.079 Um so this is only from the past uh uh day or few days window. s172 00:14:56.079 --> 00:15:01.199 This is something that's relevant more for today and this is how psychic can answer those questions. s173 00:15:01.199 --> 00:15:08.160 So to summarize it, the bottleneck was never uh the capability where you can where you take all the data from. s174 00:15:08.160 --> 00:15:12.560 It was the understanding how you connect each one of the dots to each other. s175 00:15:12.560 --> 00:15:18.720 Um the most capable agent in the world whether it's like a cloud and Gemini, it doesn't understand you. s176 00:15:18.720 --> 00:15:20.560 He need to process it beforehand. s177 00:15:20.560 --> 00:15:22.639 This is what we're trying to solve at Psychic. s178 00:15:22.639 --> 00:15:24.880 Um this is the Monday word model. s179 00:15:24.880 --> 00:15:26.480 Uh this is what we're building. s180 00:15:26.480 --> 00:15:29.040 Um if you have any question, uh we out of time. s181 00:15:29.040 --> 00:15:31.279 We are here uh uh near the stage. s182 00:15:31.279 --> 00:15:33.120 Thank you very much for the time. s183 00:15:33.120 --> 00:15:34.880 Uh you can follow us on LinkedIn. s184 00:15:34.880 --> 00:15:37.120 We post things on on the area. s185 00:15:37.120 --> 00:15:39.680 Thank you very much. s186 00:15:39.935 --> 00:15:41.935 [applause]