데이터와 AI 혁명! 스노우플레이크 창업자가 밝히는 클라우드 미래 전략

[Applause] Yeah.

 So actually today we are pretty much honored to have you guys here.

 So you guys are the legendary entrepreneurs and it's really really amazing and inspiring still.

 So you guys were born in France, never studied in the states and you had your degree in the Paris and you guys moved to the Silicon Valley in your mid20s and spent a time more than 10 years in Oracle.

 So more than actually I came in the US in 96 and Terry 99.

 So 16 years at Oracle, 13 years.

 A lot of Korean entrepreneurs and engineers they are also you know interested in moving to the Silicon Valley.

 So what made you move to the states back then 1996? So at the time I was working for Bull which was a French IT company a little bit like IBM but for France and they they were their machine uh were running the Oracle database system and I was doing research at the time in databases.

 So I was hired by Bool, you know, to to take care of all their databases and I was sent to Oracle to port the Oracle database on top of of the Bool platform.

 And I love it so much in San Francisco because I I wanted to be a software, right? I was doing research in databases.

 I was I was not really interested by hardware.

 Uh uh so I decided to um to join Oracle because at the time you know in France it was not a great environment to do software and and software was not really recognized like it it is in the in the US.

 Um so so that's why yeah so what about your yeah so at the time at the time in I was working for IBM I was part of a European center of applied mathematics and and I wrote a very long proposal actually as part of that work to integrate data mining algorithm and text mining algorithm in database system and I ship that to the US to you know the research center that we had in in the US at Almaden and We had no reply, no answer, no reply.

 Nobody cared about what we were doing and I was so excited about it and a little bit frustrated about it.

 So this I did but if I wanted to do things I would have to be much much closer to the development and the engineering of of this database system otherwise it will never work.

 Yeah.

 Actually 1996 was right before boom.

 Yeah.

 Yeah.

 And and the funny thing is that so so I was working at Oracle and then and then my boss at the time in 99 told me we have a resume from a French guy and I called you know this French guy and I cannot understand anything about what he says because of his French accents.

 Yeah.

 Because of his accent was was so foir can you interview that person who was Terry? So I called Terry.

 Terry was in France and I interviewed Terry and we discovered that we did the same PhD in France and we went you know both to live in Germany.

 So we almost met in France and we we didn't met uh so we met the first time in 99 when he he came to interview at Oracle.

 So you guys are best friends from that time on.

 Yeah we were since then.

 Yeah.

 Interesting.

 Could you please tell me about in the Oracle because you guys spent more than a decade in Oracle.

 So and I I watched in the other interview of Benois said that you were a little bit disappointed that Oracle then Oracle was losing some the big data opportunity and in the cloud opportunity.

 So what exactly happened? Why did you leave Oracle? Yeah.

 So I mean I we loved I mean both of us we we love our time at Oracle.

 We were in the core database group uh building you know for the data warehouse but but yes we after you know 16 years we realized that Oracle was missing two important revolutions.

 One revolution was was big data, right? It was Adoop at the time and the ability to um do analytics on semiructured data on on machine generated data at pabyte scale, right? This is what Adoop was doing.

 But but we thought that Adoop was such a bad solution uh uh because it was slow.

 You know, you would ask a qu a question and you would only get the answer the next day.

 I mean I exaggerate a little bit but that that was and it was complicated to manage and and you know very difficult but at the same time you know um we also thought that that there is no reason to have a system just for semiructured data and a system for structured data like data warehousing.

 So we thought you know it should be only one thing that that can do both uh very well.

 That was one one idea I mean for snowflake and the other idea was the cloud uh analytical system are are very know spiky right you want to ask uh questions and you might use them for few hours or few minutes but then you know you don't use them or you know the load their load is is very up and down and when you want to do a lot of analytics on them you are going to consume a lot of of compute power.

 So the cloud was we we thought was a great place because you know in the cloud you you can provision resources and and and you know for the first time in the cloud you know software can drive hardware that that was magical to us because if if you think about the old days of Oracle system right someone would decide okay this software needs to be installed on that machine and that machine I'm going to configure it with that much memory that many you CPUs or the set of machines and the software will be confined in this environment and to survive potentially in the bad house is is like when you are born someone put you in a house and might be a great thing or not a great thing.

 So, so, so we thought that oh if software could decide what is good and depending on the demand of course is not a static decision and can you know ask for more hardware and and and expand to more hardware or contract right when it's not used that would be amazing yeah I clearly remember in 2012 I was also doing a kind of analytic startups and we had a hard time to figure out you know all the Hadoop and securing the spot instances from AWS yeah so Let me just proceed to the next question.

 So you guys founded your company in 2012 in the small apartment in Akamino San Mateo in your apartment and so and you launched your first product around in 2014.

 So you prepared a two years.

 So could you please to tell us you know what happened during those two years? I mean this two years started with a lot of discussion between Benoa and I in his apartment.

 the first thing we bought because we knew we knew we had to build a new data warehouse for the cloud but we didn't really know at the time how we would do be doing that.

 So so we spend a lot of time trying to distillate what building a system in the club meant what businessicity of a cloud meant for us.

 We spent maybe six months up to six months both of us in that apartment.

 We bought a whiteboard.

 We we bought a Okay.

 a printer to took us a little bit of time.

 Yeah.

 So, any interesting story during those two period? Yeah.

 The the great thing is that Terry was doing the cooking cuz I'm very bad at cooking.

 Uh so, and I was doing the dishes and we were working off my living room.

 So my wife was not that happy to have a big whiteboard and computers.

 We had two laptops.

 So we bought two laptops.

 We had a small server that we were sharing in it was in the closet.

 Uh and and as the was saying the the goal was was to determine what is the architecture that could leverage the cloud and and and interestingly we when we left Oracle we had no idea.

 We knew what that we wanted to build you know a system fully in the cloud as a managed solution.

 So we wanted simplicity and at the same time we wanted to you know unify you know structured data and semiructured data.

 So we knew what what we wanted to build but we didn't know how to do it.

 So it was a bit scary.

 we had to jump in in the water, so to speak, and and see if we could swim.

 And and we knew that if we didn't have like an amazing idea on how to do that, we would probably, you know, stop, you know, the startup before it started because we yeah, we we thought that the only way we can compete against the Oracle and Amazon and and all the big, you know, data vendors.

 The only way we we could compete is if we were building a product that was highly differentiated from its architecture.

 So so that so that was the challenge.

 Do you remember the the launch moment exact the launch moment of your product 2014? It was the type of the grand launching at once or it was a gradual launch.

 No, no, it was um it was uh so first it was 2015, right? in the the GA for the product was 2015 and we remember very very well the the most that that led to that to that GA uh because because at the time we had our new CEO Bob Bob Mia and Bob was actually driving the the weekly cadance monthly cadance of these different these different milestone that we had to hit before G and so we we named each of this milestone with you know runs ski runs.

 So we had uh you know we had rapids we had challenger we had hyonoros which was because we were skiing.

 Okay snowflake comes from skiing and our passion from skiing.

 So everything everything at snowflake is named after snow and something and there was a northstar lake tow you know ski resort and he had a run.

 So we were naming all this milestone uh uh as runs in in that in that particular you know ski resort.

 So we remember you know that that GA moment we had you know ready customers uh before in 2014 um big you know few customers that that's you know hardened the product but but we wanted our GA to be very solid right with security and and really call it no product that anyone could use.

 But the the last run the last runs to GA was called village run and it was called village run because this is the the last run you take before you go home and that was the name of the last milestone.

 So we remember this moment very very clearly and one more interesting thing is that know you guys never were the CEO of your own company.

 Yes, we always wanted to build a a product.

 We we we never wanted really to build a company.

 Uh so it's a little bit inverse you know a lot of founders wants they want to build a business uh and therefore they want to be the CEO we wanted to build a product and and for that we had to build a company and that was I I always say collateral damage no no choice so so we never wanted and desired to be the CEO and we found that our our special power if if you want were not about you know driving the business.

 Our special power were were to build software.

 So so we wanted to stay focused on the software and not you know the business side.

 So so we knew from day one that that we wanted to hire you know the best CEO and and and that's very important in our philosophy.

 I always say you know building a startup and and you know a few things about that yourself.

 Uh uh but but you know building a startup is a little bit like having kids right except except there's a big difference when you have you know kids you know the two parents are going to be there forever and you have only two while when you build a startup you you bring you know new talent new expertise and and they are you know these people are becoming really true parents of the kids.

 So the number of parents you you can think about it you know expands and the CEO of course is a huge responsibility but you know the sales and you know marketing we had to have so many talents to Snowflake and the magic of of of really building a company I think is is to add uh these talents and these people that you can trust you know like really you trust you know your spouse and that's that relationship that that you have with with other employees and colleagues uh of snowflake is is critical.

 So to me when people say wow you must be proud I I always think no was not only us yes we maybe started it but snowflake is so many more than just you know the the two of us and and that's that's very important I I remember that you know in the interview with the EO you referred to some you know the European and American analysis.

 So could you please tell a little bit about the to the audiences? Yeah, I I think you are you are referring to um yes the trust so in the US you know in Europe.

 So so it was an employee we we an employee from from Europe and he was asking a lot of question he was very suspicious about how much you know stock option he will get and what will be salary and how it was working.

 So he was asking a lot of question and at some point he said you know bona I'm asking all these question because I'm from Europe and in Europe when you interact with someone you always start at zero right zero confidence and then as you know that person you you build up you add you know confidence and then maybe you get to 100% right for your best friend and so in the US it is really the reverse you always start at 100% you don't know someone and you start to interact you assume the the best you assume the best intents and then potentially you know you you downgrade from there but but it was an interesting philosophy and I I you know to me I I I I very much like the US you know model is is like you when you don't know someone you assume and you you assume positive you assume you know you you give your trust you you might remove it at some point you know and and that would be bad but but you you you you entrust someone you know and that's that's what we did with snowflake as we expanded uh uh the rank of of of of employees working is is really uh okay so let's shift the gear to the next one data cloud so I'll ask you a question to Benoir so is a question about in the unique architecture of a snowflake the concept of a cloud no the elasticity and flexibility simplicity is all common and is quite natural in these days But it was not back then 12 13 years ago.

 So what was the know kind of you know aha moment of your architectural decision? Yeah.

 So the aha moment was you know working with terry in that department and and realizing that in a cloud fast is free and and we call it fast you know going two times faster cost exactly the same amount of money as as not going two time faster.

 So fast is really free the way it works right in the cloud if if you have a problem and you know how to paralyze.

 So so you have to see that these are analytical system that can put you know more compute resources to go faster right and to paralyze you know processing of data.

 Let's say you have one compute resource.

 If you take 100 root compute resources, 100 servers in parallel, you know, processing, you know, data, you can go 100 times faster.

 But the price because you go faster, you use this 100 compute resources 100 times less than if you had only one.

 So because you use this comput this 100 compute resources one or two times less even though it cost one or two times more because you have one of them at the end it evens out and it's the same price.

 So when you look at at compute in the cloud you can dramatically accelerate you know processing of data go way faster than you know you would go on premise and at the same time you don't pay more and and that's the the the compute model of snowflake we charge you know by the compute you use but also by the time you use this compute such that you know in snowflake we we wanted you know computing to go faster you know for the same price.

 It really only is function of of of the processing you want to do not how how much you know resources you use to do that processing and and that's the moment when we say oh my god you know in any system uh traditional database system when you grow that system you have to pay much more right um you know not not in snowflake and that's that's was the when we realized that with theory of course data systems are not architected to leverage that that compute on demand.

 They cannot you know in snowflake today it takes literally less than 1 second to allocate you know 100 you know servers and you can allocate as many pools of of independent you know servers and use them for whatever time and you know when you are done you shut it down.

 So you can have all these compute resources all working on the same data updating the same data when you know this these different uh workloads are are done you just shut it down uh and and that's that's really uh the magic of the cloud and the elasticity that that was kind of the hard moment the the other aspect is you mentioned cloud right you know sole is is is now a cloud we we we it's a true cloud in in the sense that that's not only we we want to do process processing of data uh but we have added AI so you have these different services like like the cloud which are based on this elasticity and the serverless so you pay only for the compute that you use and for the time you use these compute resources but also you can run applications and and that's you know very important for for builders and and and so it started with the data layer I guess but but it expanded to be a full-fledged cloud let let me ask a question to t this time So I'm actually that I was hearing a lot about the snowflake because I had a friend in iconic capital.

 So you know he always notified you know what's going on in snowflake and I'm kind of feeling that you know the at the beginning snowflake was a kind of you know the cloud based Hadoop plus NSA server but you know suddenly it transition to some data cloud and data more than that and then now you guys are transitioning to the know something combined with AI.

 So you know could you please explain the know your transitional moment? I mean it's not so much of a transition than it is an expansion was snowflake at the very beginning.

 We we had to build a data platform in the cloud.

 We started with analytic because we we are a little bit lazy and it's probably one of the easier problem to solve when you have plenty full of resources when you have the elasticity of a cloud and the resources of a cloud.

 But our ambition was even at the beginning our ambition was much more than that.

 Uh and so over time you know we introduce we introduce collaboration because the architecture of snowflake allows for collaboration between inside inside a company but outside of a company a little bit like Google doc you know where you can exchange no not document not you know a Google doc document or a spreadsheet but you can exchange between companies share pabyte of data.

 No.

 And and it was a little bit magical and it is magical today to see how companies are creating that network of data which is the the data mesh in a sense that mesh all these enterprise and they exchange data they create value.

 It's not so much that we had to switch and create new things each time.

 You know the platform expanded by its own nature.

 you know, it's a little bit like a fish out of water that lands into into a new place and grows legs and grows, you know, start to walk on the earth.

 Uh, so Snowflake evolved over time bringing strength from his DNA.

 We we build our we call it the AI data cloud and and we built it, you know, bottom up and and the first, you know, key capability of the cloud was data, right? data was so critical for us and all type of data at any you know scale.

 So it started with this data you know art if you think about it.

 Then we added you know AI and AI is is super critical because and and I maybe we'll talk some more about it but AI you know is is needs data but data needs AI and and in so somewhere the relationship between the two capabilities is very symbiotic.

 If you think about the cloud, the cloud you know is an operating system and is a modern operating system for applications to execute in.

 So first you know when I say snowflake is a cloud often time people say but you're not a cloud like Amazon or Azure right you you you don't have data centers.

 Yes we don't have data center we we we decided from day one that we will not build infrastructure is a little bit boring and and too difficult to do.

 So we are the cuckoo cloud in the sense that we are creating this cloud diagnostic layer and we ported you know snowflake on all the top of cloud providers Microsoft you know Azure you know AWS with Amazon AWS and and of course Google GCP and our this abstraction that that we created on on top of each of these cloud provider allows you know our application not to have any dependency with a particular cloud and actually data can flow from different cloud region and you can connect as the was saying data set and and and you know your large organization can use you know many clouds many different cloud region and cloud provider in a unified way right we have the a little bit the Linux uh of of the cloud operating system uh so you don't need to worry too much about which hardware you use right you are using Linux at the end so for the just for the audience's understanding so could you Please just get us some example of a pabyte of data.

 How big is it? It will not fit in that room.

 I think a pabyte is one trillion bytes.

 It's a lot of data.

 Um at the same time you have to see that snowflake ID compresses you know data sets.

 So when we say you have one pabytes of of data in in in a in a table let's say in snowflake in reality you have many more than one pabyte generally depending on if it is semiructured or structured data semi-structured data comprises a lot more so generally uncomprised will be at least 10x or 20x you know that that volume uh so you'll be really 20 pabyte of of raw data uh uh that that will comprise down to one pabyte I mean it's important the world of data is exploding today and there is a lot more data in front of us than there is behind us and so you really want to have a platform that can scale and and and and take advantage of that data going forward.

 You know each and every of our customer can have multiple pabyte of data and snowflake itself is on multiple exabyte of data.

 Yeah, I think our our biggest customer has more something like 50 pabyte or 100 pabyte just one customer.

 So so if you take the sum of all our customers is is is a lot.

 So okay so since the most of the audiences are very interested in AI things.

 So let's shift the gear to the AI.

 I think that you guys already know working with a lot of famous guys in the Silicon Valley, you know, Anthropic, OpenAI and Google.

 You're all working together with the the great companies that Dar said maybe a two weeks ago.

 Maybe we're going to reach it sometime Asia situation until 2027.

 He says what is your you know general take on the the recent advancement rapid development of AI models be before we go to LLM and and you know why why AI and LLM in particular are are critical for data platform why why why is it used why why is it even relevant right when you use chpt and and you use a data you know system you you don't see how how they relate right so so let's start just with ads.

 So, so when we started Snowflake, I explained how you know our vision was to consolidate structured data and semiructured data, right? But I didn't say anything about unstructured data.

 Unstructured data is any type of documents and and human generated data, right? Human generated data was a different type of data that you would never put in this data platform because of course you cannot process you could store the data, right? anyone can store files that was not a problem but you cannot you know do processing on that data do analytics for example on that data now with AI you know you can do that right you can take a document an image a multimodel you know document and and you can extract you know information of it you can structure that you know generally generate you know semiructure from extracting you know information you you can you know index you know these documents and you know search them in natural language for example.

 So all of a sudden uh data platform were relevant for unstructured data.

 So that's the the huge revolution for the and for the first time in history right unstructured data became part of the data you know data sets of of any large organization.

 Now, now it's game and and people don't really know how to use this data, how they can leverage this data that has a lot of of super useful information a little bit like at the time when we started snowflake a lot of companies said okay you know semi-short data right machine generated data what what am I going to do with my logs you know how to create logs and and this is complicated so so a lot of company took a long time to use you know to leverage this data how every company leverage you know machine generated data and I I strongly believe that unstructured data in in in one year or now actually every company will you know use you know unstructured data and as part of their you know data platform so so that that's why it's so you know critical for snowflake now if if what one aspect which is also super important for snowflake is governance right companies large organization they don't want to send their data to to anyone you know for to do any processing they want processing to come to the data and level zero of the processing is LLM right so that's why it's so critical for snowflake to run inside the snowflake security parimeter so inside our cloud run natively run all these LLMs so we are running we are the only cloud We have just announced you know this partnership with open AAI.

 We are the only cloud that can run open AAI you know entropic you know even deepseek lama you know mistrol and all these these great LLMs run them you know inside our data cloud such that you know our customers who want to use one of these you know large language model you know they don't need to send their data to a to an LLM service provider the LLM is coming to them but but of course it doesn't stop there is just you LLM is is a core building block as you said in in in your uh presentation is you you need to build you know agents or you need to build workflow with this LLM you you you need to use them and you needed to at the end build applications right this is why it's so important for snowflake also to run full-fledged application in our cloud which can leverage you know this level zero of LLMs and build amazing applications uh that can you know work on the data that our customers have whether it's it's unstructured data or or or even structured data.

 So, so that so that's the the the philosophy if you want of of our cloud.

 Yeah, I totally agree with you.

 So, so actually my my company is a wholesale and I do have a similar experience because we are 300,000 numbers of a small beauty brand and we sold in 4 million products but you know all the database the sage database was unstructured.

 So we then transformed it to the organized form.

 Now we can just ask LLM to you know make a query that really really complicated the substate query so that you know marketer directly you know talk to the data now that was the expression you used this quite many times talking to your data right yes how to interact directly with data and as I said we want to this convers conversion right you you you you talk to your data and and it it's not that you don't talk only to unstructured data you might talk also to your structured data when you talk to your structural data of course you are going to use the LLM to generate you know code SQL you know for example to extract you know the data which is relevance you know to this this interaction when it's unsocial data you might use you know vector search and it's a little bit different and the agent you know will direct to the appropriate you know source of of data so so creating this experience and it's not it's very important also to understand that you might extract ract, you know, data from unstructured documents and then join it with structured data, right? You know, structured data know your customers and so so so being able to do that in one platform is is critical.

 Don't you know it's very important to pick you know back to my cloud you know we are building a cloud if you are building if you are starting a startup and and you have to decide which cloud am I going to use you you want really these capabilities to be tightly integrated right the the the decision of what is your data platform what is your data cloud uh uh is a very important decision because you need all these capabilities you know very well integrity because you don't want to spend your time developing you know that's uh that's you know type of technology.

 So, so actually I have a one more more question about that actually you know even though we are just in talking about how great LLM is but I think that you know the the actual application on huge amount of data and talking to the parabytes of the data I think that you know it's not there yet I think so do you have any great example to show that you know how ready we are for paradite of data there is a lot of these application but one one thing which is important.

 I I'll go to to an example as you said it we are not there yet and you need to ground this LM interaction with with actually real system and and predictable system a deterministic system in order to to interact with these things because both the input and the output of the system needs to be very well structured in order to get anything out of the systems.

 An example of a company which is not pabyte or terabyte of data but which is quite interesting is is a small startup which is called Maxa.

ai AI which is it's an ERP system and they are you know large company have multiple ERPs this company is aggregating all this ERP data from multiple of the ERP into a single ERP model and and they are building business functions you know we have we have LM that are building actually application business application on top of this unified semantic model of a CRP data so I don't sure how large is their data but and and what what one thing which is important to understand you know this and we have many you know applications that are built in the snowflake cloud and and the reason why Maxa for example has has grown so quickly is because they are running in size so the the the model of snowflake snowflake is a little bit of an iPhone model in the sense that as an application provider like Maxa you can put your application on the the marketplace what we call it the Snowflake marketplace.

 You think of it as the app store.

 So you can source your application in the in the Snowflake app store and our customers can take your application and install them in their accounts.

 Okay.

 So so everything that this application has and this application might even include some data from the provider, right? All these things are now installed and running in the security parimeter of the customer who is installing this application, right? A little bit like your iPhone.

 When you install an app on your iPhone, it it can access, you know, local, you know, for example, your your your your calendar or your contacts and and but it's it's it's protected by the platform, right? So so so it's exactly the same model with with native apps like Maxa.

 So Maxa was able to touch you know all our large you know customers because our customers they know that when they install you know even though Maxa was a small startup not very well known they know that when they run you know their their their this application in their account it has all the security guarantees right the application cannot exfiltrate data for example it cannot you know and it can access local data that the customer has So the ERP data that Terry was mentioning is customer data right they don't want that data to leave it it's the most important you know financial data of a company no company wants to share that data with anyone so so having the expertise the application going to the data is such an important you know aspect uh uh and and that's what we we enable with with this you know iPhone you know application model so since you covered in the most of the the question we prepared for AI.

 So let me just get back to the general question of AI.

 So I asked you that you know what do you think about you know when do you think that we can achieve into AGI or things like that because I think that you guys have a lot of information will not come fast enough for me.

 Yeah, I don't know about a AGI.

 Um, you you know when you know I mean it's it's super interesting depending on who you talk and and we of course have a lot of interaction with we we have also our you know at at Snowflake we build also our LLMs and and we are very you know very um into that technology too and we have you know top top class you know researcher actually on on that subject.

 Um so so so they have a lot of insights.

 It's kind of interesting because because even you know people in that field don't really understand how it works.

 It's it's a lot of trial and error and and the way they they they trained you know this model and and we train our own LLMs which was named Arctic.

 Um um and it was fascinating to to see you know how they they how they they they work and how they think about it.

 I I think people who are who are in the field really think that that's it's it's it can be achieved right it's just you know compute power and and at the end you you can build a neural network which is you know maybe bigger than the human brain and and or as big as the human brain and but you will need multiple nuclear plant in order to power it because the human brain is so you know power efficient but but but then you have the cloning so you know the cloning is is is is aware of.

 So, so I I it's a little bit science fiction now.

 I I don't think we should be scared about it because often time AGI is is used a little bit as a way to to scare people like like okay if machines are becoming as as smart as human being you know what what would be the word and and it's true that it has a lot of you know open you know questions so so I would not think too much about that when you know how LLM works you know they are just it's a prediction algorithm right they you know based on on on their training and what they saw uh they they just predict you know the next world.

 So so it's not you know uh it's not intelligence like human intelligence I would say but but yeah you know in the past there was a chewing test.

 I think all the today are passing the chewing test.

 I think we need a roomba test where you look at your room and if your room eat your stocks that we've we have not achieved AGI yet.

 So that's the test you want to to the physical world is actually a lot more complicated than you know than the than the virtual world which is true.

 I mean when you see you know even if you had a brain um um you know as smart as a human brain if it cannot you know act on the physical world is is is not yet you know what makes us unique is is our ability also to have physical interaction.

 uh um so so so I think it will take some time you know there's a lot of robotic you know associated to it so this is going to be a fascinated fascinating fe the Roomba test I will be worried Siri had admit so many socks eaten by the Roomba yeah since we are running out of time right now so so let me just like you know back to the the last question because there are so many you know talented tech entrepreneurs and engineers in the audiences so could Could you please give some advice know for the the future tech entrepreneur and they want to follow your trail? We always say we never have advice.

 We we don't like advice.

 Don't take any advice.

 Uh follow follow your intuition and what you think is right.

 Now saying that my advice will be to trust others and and you know as I said you know when you want to scale your your company if you are building a company you know having that and and you know trusting the value of people right right at the end is not only talents that you bring but it's it's people that you you have to trust right as I said you are bringing a new parents and and this is more than just you know competency it's it's also you know values you know, sharing the same values and it could be, you know, values, company values, but it could be also product values, right? Snowflake, you know, simplicity, you know, making everything everything frictionless is is such an important value of our products that that's, you know, when we hire engineers, you know, this is an important value that that we have to share and and that we look at, right, when we interview people.

 So, so it's important to think about your values.

 What are your product values or or your company values you know what are your human values because at the end we work all together and and a functioning team often time you know share the same you know passion values and and and and so that's very important that's a lesson we learn and every time we compromise on that we always regretted it uh when we say oh this is it doesn't have the same value and product value but is such a great competency that that we should hire that person.

 We always regretted it.

 So don't hire someone just because of competency.

 Think about values.

 That would be one advice I would have.

 Then you know never never hesitate to be ambitious.

 Never hesitate to be to be against you know everyone else.

 You know if you have strong opinions you know go with your opinions.

 You know what when we started snowflake everyone told us is never going to work because the system that will work is adoop right adoop is going so either you are building a system on top of adoop or you are going to fail right because adoop is such a an amazing system and no and we say no we believe adoop is a really a bad system actually and you know adoop now is dead no one is using I mean no one is using adoop there are still people using adoop but but it's not you know everyone knows that that is not it was not a great system.

 So, so don't hesitate to voice, you know, what what you strongly believe in.

 You know, belief is important.

 Um, and and staying true to your belief is is very important.

 Don't, you know, go left, right, and you you have to have a nostal and and and and is very important.

 I think, you know, we talk about it simplicity.

 You believe in simplicity.

 Don't compromise on that.

 [Applause] any additional advice from your side? I don't know like Bena says we never give advice but when I think back of 2012 the revolution was in the air you know everybody was talking about big data about you know and I feel like today year last two years you feel that the revolution is in the air again there is a revolution the air revolution there is a wind of change which everybody feels it nobody knows where the boat is going to go but but you know there is revolution in the Yeah, you want to be curious about that revolution.

 You want to be, you know, optimist about that revolution.

 You want to be ignorant about that revolution a little bit to to rediscover things that you are doing.

 So I think this curiosity and this ability to feel the wind is is super important.

 Thank you.

 Thank you.

 Thank you.