100x Engineering Productivity with Garima Kapoor (MinIO)
Garima Kapoor, co-founder and co-CEO of MinIO, joins Greg Shove to explain how a 200-person, decade-old infrastructure company mandated AI use across every function — and what it’s learned about measuring the return.
MinIO builds the object storage and memory layer that enterprise AI runs on, founded on the bet that most enterprise data would live outside the public cloud. The company raised $100 million in a 2022 round it has never had to touch, is cash-flow positive, and has grown from 40 to over 200 people, with most of that hiring in go-to-market roles over the last two years.
In this episode, Garima and Greg discuss:
- Why MinIO mandated AI use company-wide — “either you use AI, or this isn’t the place for you” — and what happened when they rolled it out
- The claim that some engineers are 100x more productive with AI, and why that number isn’t uniform across the team
- Cutting time-to-market for complex backend infrastructure from roughly 12 months to about one month
- Managing an inference spend in the millions of dollars annually, and why MinIO is already holding executives accountable for token budgets, not just headcount budgets
- Why she thinks there’s no single “primary” AI model — using Claude and Codex for different workloads depending on the task
- Building internal AI tools via MCP servers to connect sales, calendar, and CRM data for faster onboarding and management
- Hiring and interviewing in an AI-saturated world — screening for curiosity and clarity of thinking over raw AI fluency
- Running MinIO as co-CEO, with a spouse among her co-founding team, and how she thinks about AI’s role in her own kids’ education
If you’re an executive trying to figure out how to measure AI’s return on investment — not just adopt the tools, but hold teams and budgets accountable to them — this conversation is a real-world look at how one profitable, fast-scaling company is doing it.
Transcript
think of models as reasoning engines.
NR:Now, you need to provide it the right kind of context to give you the right
NR:kind of reasoning at the end of the day.
NR:So again, it'll come down to how smart you make your agents, what kind of
NR:right context you give your agents to then feed these engines to provide you
NR:with the right kind of context, right?
NR:So we will see a world of multi-frontier models that are getting
NR:used even for same task, right?
NR:Because there are different workloads even when, one task is involved.
NR:greg_shove-f5d793f5-90bb-442f-9a2c-1-CFR: Welcome to Super Companies, my quest
NR:to understand the smartest and most valuable companies and how they're
NR:gonna get built, how they operate on the inside in the age of AI.
NR:I'm your host, Greg Schove, seven-time founder, CEO of Section AI, a company
NR:that I hope will also be a su- super company in the years ahead.
NR:And when I say I hope, what I mean is super companies don't just
NR:grow fast, they also make money.
NR:And so, uh, today's candidates for super company status have to turn into
NR:profitable companies for many, many quarters and eventually, of course, years.
NR:That's what super companies really do.
NR:Before we get to our interview today, here's on my mind.
NR:Empathy.
NR:Believe it or not, I have a little bit of empathy for AI CEOs, at least the frontier
NR:lab CEOs, guys like Sam Altman and Dario.
NR:They are getting a lot of conflicting messages, so to speak.
NR:One is beat China at all costs.
NR:That's what they hear from Washington.
NR:Of course, from investors who have bet hundreds of billions of
NR:dollars, it's beat each other.
NR:This will be a winner takes all, uh, uh, world or economy, so yes,
NR:be the number one or number two.
NR:There won't be much room for anyone else.
NR:Of course, for consumers it's like, make us really safe.
NR:Don't take our jobs.
NR:Don't kill us, but also plan my travel.
NR:Employees who've been working probably, you know, 16-hour days, six days a
NR:week, a lot of them for years inside these companies are walking around
NR:telling their CEOs, "I need an IPO.
NR:I, you know, I wanna buy a house and actually start a family."
NR:So listen, we know they're power driven.
NR:We know they're money driven.
NR:We, w- we know they want to become the most powerful and
NR:richest men in the world.
NR:But I, in this moment, I have some empathy for them in terms of there
NR:aren't a lot of rules right now.
NR:And, and as Scott Galloway, uh, my good friend and founder of Section said,
NR:uh, in our conversation on stage last week in New York, capitalism needs
NR:rules and guardrails, and everyone's better off, both the, uh, investor,
NR:the entrepreneurs, and, and, and, uh, company leaders and consumers.
NR:And we don't have that right now.
NR:A lot of com- competing agendas.
NR:And as a, a, a leader in your own organization, you're
NR:probably feeling the same.
NR:Are we supposed to go fast?
NR:Are we supposed to protect, uh, jobs and human dignity?
NR:Are we supposed to go fast or go safely and so on?
NR:And again, I think that's why if you're a leader you get paid the big bucks.
NR:You're gonna have to handle those competing objectives and tensions
NR:as best you can and make a lot of progress at the same time.
NR:All right.
NR:Let's get to our interview
Greg:My guest today is Garima Kapoor, co-founder and co-CEO of MinIO.
Greg:Quick math on MinIO, about 200 people, a billion-dollar valuation.
Greg:MinIO is the object storage layer for enterprise AI.
Greg:The company is over 10 years old, so clearly, unlike me, she's been in this
Greg:business, the AI business, quite a while in terms of building enterprise
Greg:infrastructure for the Fortune 500.
Greg:But that's not why she's here.
Greg:She's on the podcast to help us understand how she is leading and managing the
Greg:company, how she's building a super company, why she thinks about the
Greg:opportunity cost of tokens when she's deciding whether to hire, why she's been
Greg:able to get 100 times productivity, that's what she says, out of engineers with AI.
Greg:She's making the call that every leader listening to this
Greg:podcast is gonna have to make.
Greg:When does the next dollar go to an employee or to compute?
Greg:That's how super companies think.
Greg:Garima, welcome.
NR:Thank you.
NR:Thank you, Greg, for having me here
Greg:Okay.
Greg:Let's start with the description of what you guys do, 'cause you've been doing it
Greg:a while, and I'm r- especially curious around were you early or were you doing
Greg:something else and then did you apply it, uh, to this age of AI in terms of
Greg:enterprise A- AI in the last few years?
Greg:So, but first of all, tell, tell us, uh, what the company does today.
NR:Yeah,
NR:we, we are a data company, and that's why, you
NR:know, in
NR:the age of AI, it's only about evolution because AI is only as good as the data,
NR:uh, you know, that you train your models on or that you build your agents on.
NR:So for us, it has been, you
NR:know, an
NR:evolution from data to data and a memory, uh, company for, uh, enterprise AI.
NR:Um, just to give you a brief background, right?
NR:We started the company back in, um, 2014, 2015 timeframe, and, uh, you
NR:know, that was the time when cloud was growing and, uh, uh, for us, we wanted
NR:to bring cloud architecture close to wherever the data was getting generated.
NR:AWS's, uh,
NR:worldview was that bring all the world's data to us and, uh, you know,
NR:we will deliver you the capabilities on, and our worldview was that bulk
NR:of the data is going to be produced outside of public cloud, and we'll--
NR:we want to be closer to the data.
NR:And, uh, that has only proven true and true, you know,
NR:over the last, uh, 10 years.
NR:If you see, you know, the data that is getting generated just around
NR:us, you know, your devices are pr- generating data, the satellites around
NR:us are generating data, you know.
NR:So We started with a hypothesis that it's going to be a world where
NR:infrastructure is going to look same no matter where your applications are.
NR:And the key to making it happen is through open formats and industry
NR:standards, and that's how we started.
NR:You know, we started with the S3 compatible data store, and we are bringing
NR:the same thinking to the memory store when it comes to AI because, uh, the agents
NR:need to be able to access the data from open format, so different agents can
NR:operate together on the same foundation.
Greg:And so did, did AI provide this kind of boost or, uh, to your business
NR:definitely.
NR:I mean, the scale has…
NR:The scale of the data that we operate on has grown exponentially.
NR:You know, when we started, terabytes of data used to be like great.
NR:You know, the companies born before us, you know, tera, uh, Teradata,
NR:it was based on, you know, terabytes of data essentially, right?
NR:Old school.
NR:So now e-everyone is talking about the exascale of data, and it helps because
NR:we started the
NR:time that we started the company, it helped us build the product right,
NR:foundationally correct to be cloud native, and cloud is all about scale, right?
NR:And, and that helped us get
NR:a very unique position or set up uniquely in the market to address
NR:the scale of the data, unlike, you know, the traditional products
NR:which are still trying to bolt on additional code or additional
NR:capabilities to address this new market, right?
NR:So that-that's the superpower.
Greg:So when you raised that capital in 2022, that was, was
Greg:still pre, uh, AI in a way?
Greg:So, right?
Greg:S- uh, so just say…
Greg:I'm just curious.
Greg:Like, so you were, you were able to raise that capital.
Greg:You haven't touched it apparently.
Greg:Is that right?
Greg:Or is
NR:Yeah.
NR:Yes, yes, yes.
Greg:Okay.
Greg:And the, so the company is cash flow positive?
NR:Company is cash flow positive.
NR:We raised the round, uh, I-- we closed the round, uh, in Jan of
NR:twenty twenty-two, so that-that's when we raised, uh, hundred million.
NR:And of course, the balance sheet has only grown from there thanks to…
NR:I always say customer money is the best money.
NR:So, uh, and we were only 40 people back then.
NR:So from 40 to, uh, we've also hired quite a bit, and bulk of the hiring has
NR:happened in last two years, uh, more so on the go-to-market, uh, side of things.
Greg:Okay, so, uh, you got a great business clearly.
Greg:Uh
Greg:And let's talk about how you're, how you're running the business.
Greg:And, and, and let's talk about how you're expecting employees to use AI.
Greg:Obviously, engineers were the first team, I would imagine, where you said, "Listen,
Greg:we, we can really supercharge this team."
Greg:I think you quoted, you made it, y- you, you said something, I think it was on
Greg:CNBC, that your engi- your engineers are 100 times more productive, which I
Greg:th- I think is you being provocative.
Greg:100 times seems crazy.
Greg:I mean, I, I, I, I can kind of see 3 to 5X productivity in
Greg:our engineers in my company.
Greg:Are you really seeing 100 or you just, kind of, uh…
Greg:Is it, is it, is it, is it…
Greg:What's going on?
NR:So it, it was not for the, uh, uh, sake of
NR:being, uh, you know, provocative.
NR:Uh, I do see that productivity.
NR:Now, it depends, right?
NR:There are certain core group of engineers to whom you just enable with
NR:AI, and that's where you see the 100X.
NR:Of course, there are engineers that still need to ramp up, uh, in terms
NR:of using AI, uh, you know, in their,
NR:uh, workflow
NR:a lot more, and that's where you
NR:see the productivity difference, quite honestly.
NR:But at the rate at which the new products now can be brought and
NR:launched to the market, the time to the market has reduced so significantly.
NR:It's, it's unbelievable.
NR:So I do see that productivity gains across
NR:the team.
NR:In
NR:engineering, of course, you know, because, uh, um,
NR:that is the, you
NR:know, the right fit for-
Greg:it's the perfect fit for c- obviously
NR:perfect use case, right?
Greg:or for, yeah, for a use case.
Greg:Yeah, So i- if a time to market was 12 months, I'm just using that as
Greg:a timeframe, if it was 12 months before, what do you think that's
Greg:now become for a software company or for, or for, or for you guys?
Greg:Like, is it down to a month?
Greg:Is it down to a couple months?
NR:It is down to a month.
NR:And, uh, you know the thing, it all depends on, you know, the
NR:problem and the product that you're building at the end of it and how
NR:complicated the product, uh, is.
NR:But even for a backend, uh, uh, and data store and memory store are not,
NR:uh, easy products to build and launch.
NR:They require very hardcore engineering efforts, uh, in terms of, uh, end-to-end.
NR:Uh, and, uh, you know, it is a mission-critical product.
NR:If your, uh, stored memory systems are down, then you know your infrastructure
NR:is down and your applications cannot work.
NR:So even on those fronts, we are seeing time to market just
NR:like, yeah, accelerate by 10X.
NR:I mean, it's, it's
NR:amazing these
Greg:let's keep going though with, with the engineers, 'cause I consider them the
Greg:canaries in the coal mine, meaning what's happening in software engineering as the
Greg:team, I think eventually happens across the rest of the knowledge workforce, uh,
Greg:pr- you know, probably in marketing and other language-intensive functions first.
Greg:But let's stay with the engineers.
Greg:So what about…
Greg:And this might be going back now two or three years, obviously,
Greg:uh, their levels of anxiety.
Greg:How did you get through the resistance?
Greg:How did you get through the r- uh, anxiety?
Greg:Did you see it in terms of engineers worried about their, you know,
Greg:their future and their, you know, their, their kind of existence?
NR:yeah, I didn't see in terms of resistance, in terms
NR:of, you know, their thing that, you know, their job is threatened.
NR:I, uh, that resistance was not there.
NR:The, the resistance was more about I can do this better,
Greg:right?
Greg:Okay
NR:that's the engineering mindset, and those are
NR:the right kind of engineers you want in your team as well, you know.
NR:That was the resistance.
NR:It's almost asking like a chef, you know, and comparing, to bring a robot
NR:and, you know, do which meant, you know.
NR:So, and that, that's really the difference that comes into play.
NR:But, uh, we just mandated it.
NR:We mandated it.
NR:Either you use AI, uh, in your every step of the way workflow,
NR:or this is not the place for you.
NR:That was the only way we could go around it.
NR:I know it sounds very hardcore,
Greg:Sounds har- but harsh, but yeah.
Greg:But I mean, that's,
NR:But it felt right and, and it sets the culture, and it also
NR:sets the tone correctly coming from the leadership that this is how things
NR:are going to happen moving forward.
NR:It is not something good to have.
NR:This is something that we have to incorporate on every step of the way.
NR:And it's not that we mandate it only for engineering.
NR:We mandate it across the board.
NR:We told that you need to start using AI for marketing, for sales.
NR:Across the board, you need to be using AI.
NR:There is no other option.
NR:Now, the extent to it that you use, of course, engineering is the, the biggest
NR:customer of, uh, you know, of using AI internal, but, uh, it is throughout
NR:the organization we have mandated it.
Greg:And w-- And I'm just curious, is that Claude in
Greg:terms of the p- the primary AI?
NR:So we, yes, we do use Claude a bi- uh, quite a bit,
NR:uh, uh, b- for engineering, and then we use Codex, uh, uh, as well.
NR:So we have a mix, but primarily engineering is with Claude.
Greg:And what do you think about, uh, this idea of one primary LLM
Greg:or AI platform versus, you know, more than one, two or three?
Greg:Uh, 'cause I'm beginning to see, at least in larger organizations, maybe not i-in an
Greg:organization of your size, but in larger organizations, I think in fact, so the CIO
Greg:is confusing the everyday employee 'cause they're launching almost too much AI.
Greg:You know, Claude or, you know, and GPT and maybe Gemini or Copilot.
Greg:And at some point, I think the, the employee's overwhelmed and/or you've
Greg:got teams now building agents and workflows across different platforms,
Greg:and that doesn't kinda, you know, they sync up very well and so on.
Greg:So just what's, what's your take on, uh, how much has to be on a
Greg:primary platform, uh, you know, versus other, other, other tools?
NR:You know, it'll be a mix.
NR:I don't think so there is one model that can do it all.
NR:There are, are models that have specific advantages, definitely
NR:depending on the task that you, uh, that a team wants to accomplish.
NR:And at the end of the day, here is how I think about it, right?
NR:The frontier models and the race has already started to commoditize it, right?
NR:Uh, the, the value is going to move more towards…
NR:See, think of models as reasoning engines.
NR:Now, you need to provide it the right kind of context to give you the right
NR:kind of reasoning at the end of the day.
NR:So again, at the end of the day, it'll come down to how smart you make your
NR:agents, what kind of right context you give your agents to then feed
NR:these engines to provide you with the right kind of context, right?
NR:So we will see a world of multi-frontier models that are getting
NR:used even for same task, right?
NR:Because there are different workloads even when, uh, when one task is involved.
NR:And you might want to do, use different pieces of different models, right?
NR:Like Claude plus maybe Gemini for abstract reasoning.
NR:You know, different, different, uh, uh, models have their own different
NR:capabilities, and you want to bring them all together depending on
NR:the task you want to accomplish.
NR:But everything will come down to the context that your agents can
NR:have from, and the right kind of context that they can feed the models
Greg:Yeah, I mean, I, I sort of, I sort of agree, but I, I kind of…
Greg:When I think about the typical marketing organization, you must
Greg:have, I'm sure, 10, 20, 30 marketers or maybe maybe more, right?
Greg:Uh, it kind of feels like that, that team's gonna be more efficient and get
Greg:more out of their AI if they're all on Claude or GPT or Gemini, you know?
Greg:Or, or, you know, just…
Greg:'Cause otherwise, I don't know.
Greg:I-- That, that's my advice to CEOs, which is at least at the team level,
Greg:um, p- uh, particularly in the bus- particularly on the business side,
Greg:where the, the ability to, you know, the fluency and ability to shift models and,
Greg:you know, use inference from different sources, I think is not as strong.
Greg:I think an engineer should be able to figure that out.
Greg:But I think when I think, when I think about a marketing team, like get them
Greg:all on Claude or GPT and, and frankly, focus more on the transformation of their
Greg:work, the, you know, the building of agents, the, the main- the maintenance
Greg:of agents, which is a whole 'nother problem, rather than get them sort of
Greg:distracted by, "You could do some of this here and some of this there," and
NR:yeah.
NR:No, a-and you're right on it, uh, uh, because it all depends
NR:on the how complicated the task you want to perform, right?
NR:Right now the task…
NR:Even now, the tasks are pretty simple.
NR:I'll give you uh, a, a, a simple example on our end, right?
NR:Go-to-market is one of the hardest functions, and there are also like,
NR:you know, you onboard new teams, uh, members on almost daily basis to onboard
NR:them, to get them quickly enabled.
NR:Documents are on different places.
NR:How do you, you know, bring them together?
NR:How do you connect your, build your MCP server, connect them to a Cloud client?
NR:These are simple tasks still where you can, you know, build your MCP server on
NR:a Google Drive where all the documents are there connected to Cloud client.
NR:It's a simple application interface that you're building.
NR:Then the employees or the new employees or your sales team can onboard and
NR:start querying the data, asking, you know, "What is my competitor?
NR:If I have to write an email for this particular ICP, how do I do that?"
NR:They're simpler task, and maybe one engine can do that.
NR:But if I'm building even for QA, right?
NR:Like I'm I'm building a complicated task, I have to build this right piece
NR:of code, then I want it to be QA'd.
NR:You know, there are different engines that do it a little bit better, right?
NR:Or plus Codex maybe, or plus something else.
NR:so it really depends on the task and the, uh, yeah, and how you define the problem.
NR:So
Greg:Green, Green, are you spending so much on inference now that you
Greg:are managing that expense in terms of thinking about open source models
Greg:and Chin- you know, and, and Chinese models in terms of source of, you know,
Greg:cheaper tokens basically, or cheaper inference in order to lower your
Greg:overall sort of average cost per token?
Greg:How do you think about sort of managing that expense?
NR:just had a conversation about this today
NR:that I want to see, uh, per person productivity, uh, uh, as compared
NR:to the tokens that they're spending.
NR:So it is a very fresh conversation in my mind.
NR:So we are, uh, getting to the stage, you know,
NR:when the expense start adding up, you know, couple of million dollars you're
NR:spending on Claude, then you really
NR:start thinking about, you know, how
NR:much consumer
NR:engineering is spending and what are we getting out of it.
NR:So those are real conversations that are happening.
NR:I have not put a, a mandate in terms of the number of tokens that
NR:a per
NR:employee can use.
NR:That's because we want to encourage people to just be more and more AI native and
NR:to use daily so…
NR:basically, So we have not gotten to the point, even though we are spending
NR:quite a bit on, uh, on tokens right now.
NR:Uh,
NR:but we will get there.
NR:And if you see the bigger or-- And we are a s- relatively smaller team, right?
NR:Two hundred
NR:plus people.
NR:So it's, it's really a small organization, and I was just
NR:joking, if a small organization
NR:like us, 200 people plus is spending couple
NR:of million dollars,
NR:it's, it's a
NR:considerable amount of money that's going.
NR:And think about as the organization scale,
NR:how quickly that ex, uh, expense can add up.
NR:So I'm sure that, uh, others who are at a bigger size than us are
NR:definitely having those conversations internally in terms of how to optimize
NR:the token spend and depending on the ROI that they are seeing, right?
NR:At the end of the day, it's all about ROI.
NR:If you-- Like I said, time to market.
NR:If you can bring a product to market, monetize it, uh, uh, you know, in
NR:a month's time as compared to 12 months, uh, nobody, nobody minds
NR:spending money on those things.
NR:So
Greg:So I wanna spend more time here 'cause I think this is a
Greg:really, really, really important question, and I think a lot of
Greg:leaders are, uh, struggling with it.
Greg:So first of all, if, if you are 100X more productive, which is what
Greg:you said, and, and I'm not saying everyone is, uh, uh, but, and you are
Greg:bringing, you are bringing products to market that much faster and, and
Greg:as you said, monetizing them sooner.
Greg:Based on your current size and revenue rate, you know, w- when do you really
Greg:begin to scrutinize the token spend?
Greg:Meaning you're at 2 million now, you're beginning to pay
Greg:attention to it, let's say.
Greg:But, but it would still seem to have so much benefit.
Greg:Is there, is there-- how do you think about when does it start
Greg:to like, "No, wait a minute."
Greg:Is that as a percentage of your headcount expense?
Greg:That's one way that I often think about is how much am I s- am I spending on
Greg:inference when compared to how much am I spending on, you know, a fully
Greg:loaded, fully loaded headcount costs.
Greg:That would be one way.
Greg:Uh, there might be other ways.
Greg:How am I, how much is that expense when compared to my next biggest expense,
Greg:you know, on, on the expense side of the, um, you know, of the income
Greg:statement, which could be all other software or it could be, you know,
Greg:it could be other, you know, whatever those other big, uh, expenses are.
NR:I mean, engineering is the biggest consumer, so of course
NR:that is the biggest bucket in which the tokens are getting, uh, uh, consumed.
NR:But the conversation was with the…
NR:with my, uh, VP of finance, and I want to see the per person token
NR:consumption and also, like, then it also shows, You know, it can go two ways.
NR:First of all, you know, per person token consumption, then you can…
NR:That's a good sign that they are using, uh, AI, but are they really
NR:lost that they are using so much of tokens to get a s-simple task done?
NR:So it's also like
Greg:sure.
Greg:Yeah.
NR:a way of evaluation, right?
NR:So that, that was the conversation, uh-
Greg:So I, I get it.
Greg:So at, at Section, for example, uh, our engineers are spending around
Greg:2,000 a month on tokens as far as, you know, as far as we can tell.
Greg:We only, we only have, uh, 10 engineers.
Greg:A- um, and the, and the business users are on average 500 a
Greg:month in token, uh, token spend.
Greg:So, uh, a low of 100 and high of probably 700, 800, um, in
Greg:terms of the business side.
Greg:how do you measure the productivity improvement, uh,
Greg:based on that token spend?
Greg:Like, are you looking, are you…
Greg:For the engineers, are you gonna look at PRs or code committed?
Greg:Or,
NR:exactly.
NR:How many, yeah, uh, PRs, the PRs
NR:that got merged, how much revisions it went through, you know, the
NR:quality of those PRs and, and
NR:the good thing is all of this also can be automated through AI dashboards again.
NR:So I don't have to hire another engineering manager to do the job.
NR:So yes, so that, so that, that's the, uh,
NR:that's the goal because, you know, um, anecdotally we can, uh, talk
NR:about, you know, one engineer versus another, but as a system in place,
NR:you need to, of course, uh, you know,
NR:set up the systems
NR:in the place
NR:that you can objectively see, you know,
NR:how much
NR:tokens versus how much PRs, how much lines of code, the quality of
NR:those PRs, how much, uh, you know, how it is getting merged quickly,
NR:it is getting merged to the code.
NR:All that
NR:flow, of, uh, yeah, can be completely automated.
NR:So that, that's, that's, the plan.
Greg:So let's, now let's talk about the business users, the
Greg:re- the rest of the company.
Greg:Harder, obviously, in terms of we have less direct, measurement, and
Greg:even when you see, uh, KPIs improving, let's say salespeople being more
Greg:productive in terms of selling more, you're never really sure was that
Greg:because of the AI or is that because, you know, they had a good territory?
Greg:How, how do you think about it for the rest of the, uh, of the workforce?
Greg:Or are you not gonna worry about it yet, meaning their spend on tokens
Greg:is not nearly at the level that the engineers are, and so that kind of
Greg:degree of granularity of measurement you're not gonna worry about?
NR:No, and I think, uh, uh, you know, those things need to be done.
NR:It's a very inefficient way for a salesperson to go and look
NR:into 10 different sources, then work with their BDR in terms of
NR:personalizing the message and then…
NR:It's a lot of work.
NR:It's, it's very, m- very hour intensive, like time intensive, where
NR:You don't want your sales guys to be spending all this time in all this.
NR:You want to get up and running, get on a call with the customer, right?
NR:That, that's the end goal.
NR:You want them to be spending time there rather than doing
NR:all these time-consuming tasks.
NR:So that I think y- we have to do regardless, right?
NR:It is more about efficiency of the business than anything else.
NR:Uh,
NR:and I don't think so you can directly measure
Greg:can't measure it,
NR:it in a way of how things were getting done versus now, right?
NR:So it's, it's a hard measurement.
NR:Anecdotally, of course, you can
Greg:it's a hard measurement, to be clear for the audience, a hard measurement
Greg:to make in that it's, it's, it's, right, it's, it's, it's l- it's hard to
Greg:attribute, it's hard to attribute in the bus- in the business functions, marketing,
Greg:sales, HR, whatever, so the, the impact of AI, at l- at least at this moment
NR:E- uh, uh, yeah, right.
NR:If you see, uh, uh, one of the big customers for big data has
NR:always been marketing teams just because of, you know…
NR:And, and it started for a reason because there are like-- it's, it's a very
NR:data intensive from different sources.
NR:You collect the data, uh, you know, from, uh, different attributes,
NR:campaigns that you're running.
NR:It's, it's all disparate, and the big data grew because, you know, you want
NR:to collect it and run an- analytics on it, and then derive insights
NR:and make data-driven decisions.
NR:Now, with AI, you can get that to one more level in terms of efficiency,
NR:uh, uh, right, and productiveness.
NR:So that's how I see, like it's more like a journey that I see
NR:than like completely changing the way things have been done so far.
NR:In engineering, of course, it's a completely different way
NR:because,
Greg:if you had to predict in a couple years, let's assume your headcount
Greg:grows at a, you know, reasonable pace, you know, hun- 200, 250, 300, whatever.
Greg:If you had to guess in a couple, three years though, Garima, what
Greg:would, what will your inference spend be if it's 2 million today?
Greg:Do you think it becomes 5 million?
NR:we want the team to be as productive as possible.
NR:As long as I see the ROI growing, why not,
NR:right?
NR:It's not?
NR:it's it's not so much about the absolute dollar spend.
NR:The dollar spend will be, will increase.
NR:I mean, that just goes without saying.
NR:As long As we are seeing the productivity, that all it, that,
NR:it, that, that's all that matters
Greg:Do you think you will begin to distribute inference budgets?
Greg:One of my predictions, it might take a while, is that executives
Greg:will be responsible for an inference budget like they are responsible
Greg:for a headcount budget and a travel and expense budget and so on.
Greg:And they'll be expected to turn that inference, that AI, into business value.
Greg:In the case of engineering, we know what that is.
Greg:It's code.
Greg:It's, you know, it's quality code committed to production.
Greg:Uh, do you think at some point the numbers are big enough such that, you know, you
Greg:want it managed by executives and that you'll, you'll basically measure them,
Greg:assess them in, in their ability to manage inference just like you assess
Greg:them on their ability to build and manage their own set of human teams?
NR:A hun-hundred percent.
Greg:Yeah, and when does that, when does that happen for you?
Greg:Do you think it's in the next year or two where you're gonna actually say,
Greg:"Here's the $2 million or 3 million," whatever it is, "and I'll divide it into
Greg:the, you know, the different buckets and expect the, the VP or whoever your direct
Greg:reports are to manage that effectively"?
NR:No, it's happening right now in our organization, uh, uh,
NR:making executives responsible in terms of, uh, the token spend and, uh, they
NR:need to be accou- uh, held accountable in terms, you know, what you're bringing
NR:to the table, uh, after all the spend.
NR:You know, much like we need to do for all the expense
NR:items.
NR:So it's just one more additional line item.
NR:That's how I see it.
NR:Uh, so it, it's happening right now in our organization.
Greg:and do you think they need, as an executive, do you think these
Greg:executives need to be trained in some way or coached on how to do that?
Greg:Like, it's a new management discipline, isn't it?
Greg:Like managing inference to create business value i- in a way that's,
Greg:you know, co- uh, cost-effective, has leverage and so on.
Greg:Like h- how do…
Greg:Or do we, or should they just kind of figure it out?
Greg:I don't know.
NR:You know, I think it's such a evo-- it-- the industry is
NR:evolving so fast and, and especially like because I operate in a startup
NR:space, everyone is pretty savvy in terms of picking things up and, you know,
NR:iterating and, uh, going from there.
NR:So I don't think so in a startup world there'll be much training required
NR:in terms of how to think about it.
NR:Maybe in a traditional organization where often processes are set
NR:up in a certain way that do not allow people to go beyond that.
NR:So the-- it, it will require a little bit more, uh, either change of processes in
NR:bigger organizations, how you think about these things, and implement it in a more
NR:systemized fashion than, you know, in a, in a stop-- startup environment where
NR:everyone is just very savvy and agile.
Greg:So Garima, you did, you did the mandate a while back, I g-
Greg:probably a couple years ago or so.
Greg:Did anybody leave or did they ha- or did, did, did have to leave or people
Greg:basically said, "Yeah, I get it and I'm, uh, you know, I'll work in the new way"?
Greg:How…
Greg:I'm just curious how that worked out.
NR:Yeah, no, nobody left.
NR:I mean, yeah, everyone, everyone kind of…
NR:I think it just communicated the seriousness, right?
NR:I mean, uh, because I think, and the thing is that it has to come from CEOs.
NR:There is no other way it can get mandated across an organization.
NR:It has to come from CEOs directly, otherwise organizations will be lost, and
NR:they'll not, uh, uh, implement it to the depth that it needs to get implemented.
NR:So there was no, uh, nobody left.
NR:There was no, not much resistance, but I think it did communicate that we
NR:are serious as leaders, that this is how we, we want to run the company and
NR:either you're on the program or not,
Greg:yeah, I assume, and I'm gonna put you on the spot here.
Greg:I assume, though, that mandating it from the top, which is you obviously,
Greg:uh, as co-CEO, also meant that you were model- you were modeling
Greg:the behavior you wanted, I assume.
Greg:Meaning you, you must be working in Claude or GPT or whatever it is
Greg:every day yourself, and your team must have realized that it's not
Greg:just you're mandating it, but it was beginning to change the way you work.
Greg:I don't wanna put words in your mouth, but I assume that's what was happening.
NR:I'm not an engineer by profession.
NR:and both my co-founders are, uh, hardcore engineers, so they were kind of like,
NR:you know, examples, like, Exa-exactly.
Greg:modeling the behavior, particularly as it relates to the
Greg:for the engineering and technical
NR:Exactly.
NR:And showing by example, right?
NR:That this is what I could accomplish or how easy or simple my life has been.
NR:And engineers, you know, always want to tinker with new things.
NR:So, you know, it was, very encouraging for rest of the team to see that, okay,
NR:if, uh, you know, my fellow co-founders are implementing it and being so excited
NR:about building things over weekend and, uh, uh, you know, getting excited about,
NR:uh, even hobby projects and how meaningful it can be in the, uh, day-to-day life, so
Greg:is it changing how you're hiring?
Greg:I'm curious today when you think about hiring, uh, i- into the company,
Greg:either an engineer or in- into a business function, what are your
Greg:expectations around how, how AI-pilled or AI-augmented that candidate should be?
NR:the hiring process itself, uh, is, uh, very strange in, uh,
NR:times of AI because now you don't even know if the candidate is reading answers
NR:from a AI engine and responding to you.
NR:So it's been very very unique experience.
NR:So as part of the hiring process, there are two, aspects to that, right?
NR:First of all, you need to also be savvy as an interviewer to, uh, you
NR:know, filter down into the things that the person is not just coming and
NR:faking it, uh, in an interview because everything is, is remote nowadays, right?
NR:So it's very hard, uh, at least the first couple of rounds of interviews.
NR:So I think just training on the management side of things or the hiring
NR:managers, things that they need to spot.
NR:And engineers are pretty good at it.
NR:Whoever is interviewing from engineering standpoint, they
NR:can, they can figure things out.
NR:that if someone is not going too deep and just being very high level and reading
NR:answers from, uh, from the engine.
NR:But on the second aspect of this question is, right, how AI
NR:augmented a person needs to be.
NR:So in every round of interview, uh, now we are making it mandatory for every, phase
NR:of the interview, we just need to go deep and deep in terms of how much AI they're
NR:using in their day-to-day life based on the role that we are hiring them for.
NR:And more importantly, I think, uh, we have also enhanced, you know,
NR:in our, process of how we interview for the curiosity of the candidates.
NR:Because for now, I mean, things are right now, you know, in a certain
NR:way, it will change a year from now.
NR:It-- They're moving so quickly.
NR:So I think more than AI augmented, we need to also look for candidates who
NR:are extremely curious and inquisitive and be able to figure things out on the
NR:fly and not be afraid of the change.
Greg:And how do you, how do you interview for that?
NR:Very hard.
Greg:I
NR:Very, very hard.
Greg:I, I, are you looking in their past where they exhibited that behavior?
Greg:Or you're ask- are you trying to find out in the moment, y- you know, will
Greg:they be curious and maybe solve a, you know, solve a challenge or puzzle?
Greg:I'm just curious if you have any, any ideas there.
NR:Yeah, mostly through the past in terms of, you know, where they
NR:have, uh, shown curiosity, side hobby projects they have gone, uh, you know.
NR:So those things are one way to find out in terms of how genuinely curious
NR:a person, uh, is and what do they do outside of their, uh, work also.
NR:So yeah
Greg:are you doing sort of screen share type interviews even…
Greg:I'm, I'm particularly interested in, in the, not the engineers, but the
Greg:business functions where you're trying to see how people work with their AIs.
Greg:Like, are you curious to the level of how do you actually have a conversation
Greg:with AI or, or not necessarily yet?
NR:Not, not necessarily, I think because prompts are so specific
NR:in terms of, you know, the questions that you are trying to solve for.
NR:And also like, uh, we, we are in a very specific, uh, space when it comes to,
NR:uh, you know, uh, comes to our product.
NR:So yeah, so it has to be very specific and you need to have a lot
NR:of clarity when you are asking the prompt to get to the right answer.
NR:So what we look for is the clarity of thinking, what we look for
NR:again, like curiosity and clarity, understanding the problem more than,
NR:you know, the answer in itself.
NR:So yeah, rest of the-- prompt is a prompt at the end of the day,
Greg:I wanna zoom out to the bubble that we, uh, uh, are in, and you and I are
Greg:pretty close to the center of it, meaning in the Bay Area, obviously, and you are
Greg:in terms of building enterprise, uh, data infrastructure, um, in this moment.
Greg:But, um, what's your sense of the enterprise's ability to really
Greg:harness these technologies and pay for them at the rate at which the
Greg:frontier labs need them to do that in order to fund this build-out?
Greg:Are you worried about the build-outs obviously happening,
Greg:significant investments?
Greg:Uh, I mean, it seems to be driving at least 50% of our GDP, at least,
Greg:you know, that's, that's what other, that's what others think in terms
Greg:of the, the data center build-out.
Greg:Are you worried about there being too much of a lag between…
Greg:'Cause you're closer to it, I think, uh, uh, than I am in terms of,
Greg:uh, or you're as close as me in a different way to large enterprise
Greg:as they deploy these capabilities.
Greg:Is, is your sense they're gonna be able to get value from them
Greg:soon enough to generate, uh, the revenues that these labs need?
NR:I believe so.
NR:I believe so.
NR:If you see the consumption of tokens are only, uh, going up
NR:and there is a huge backlog even in terms of the hardware, right?
NR:And there is not enough compute in the market right now,
NR:memory in the market right now.
NR:So those things are very real.
NR:demand is only growing, and of course, it's the enterprise
NR:customers who are feeding the demand.
NR:So profit, of course, is, uh, uh, you know, built up in the market, but I don't
NR:think so we are, uh, in a state where, uh, things are going to go haywire.
NR:so
Greg:Good.
Greg:what are your favorite use cases for AI?
NR:For me, I think, uh, uh…
NR:So we have our, uh, internal, uh, sales hub, which I love because
NR:I'm more closer on the business.
NR:So it just such…
NR:It makes life so easy.
NR:And we have, uh, in terms of even management of the sales team, it's,
NR:you know, you can really get into the depth of the activity of the
NR:rep to, you know, where they're spending time on, to the productivity.
NR:It's just so seamless
Greg:meaning you're using, let's say Claude for now,
Greg:you're querying essentially Salesforce data via Claude or not?
Greg:How, like how's that actually happening?
NR:Yeah, so we built our, uh, uh, MCP server, uh, uh, uh, you know,
NR:with the, all the data that we have, and it is querying from different sources.
NR:It has, you know, also calendar integration, everything.
NR:So, you know, you can really track and monitor in terms of what's going on.
NR:So I think just in terms of management itself, the life has
NR:become so much easier and better.
NR:Uh, we have not, uh, integrated Slack to Claude yet
Greg:You haven't?
Greg:Oh, I'm, I'm, I'm surprised.
NR:I
Greg:more.
Greg:Why not?
NR:worry about, I worry about the privacy little bit on…
NR:Because Slack is just such a fun place to hang out, and especially
NR:as a remote organization.
NR:And I know you can have like specific channels
Greg:Yeah.
Greg:How, how about, how about the public Slack channels?
NR:In a remote, uh, culture, right?
NR:You want people to be open and vocal as you're talking to each other.
NR:You don't want, uh, them to be thinking someone is monitoring me from an
NR:AI tool, although it will become a part of life at some point of time.
Greg:Here, here's my vote.
Greg:Do it.
Greg:My, uh, meaning, uh, a public Slack.
Greg:We, we, we did it about, I don't know, probably a year ago.
Greg:Uh, it's, it's really, it's a lot of rich and important context
Greg:inside a public Slack, uh, for us around clients, products, and projects.
Greg:And we are really encouraging those conversations to stay in Slack,
Greg:in the, in the public channels.
Greg:So there'll always be private conversations about any of those things.
Greg:But the, you know, when the team is involved, let's do it
Greg:in public Slack and give access, uh, give access, uh, to agents,
NR:so by public Slack, uh, you mean general
NR:Slack channels open to community?
NR:Or these are your internal organization but, like, group Slacks?
Greg:Yeah, group Slacks.
Greg:So yeah, that's what I mean.
Greg:So pu- yeah, so, so group Slacks ver- versus DMs, that's all.
Greg:So, but it, it, if it's a channel and it's, it, and it's a channel
Greg:with, you know, multiple members, so it's not public to the whole
Greg:company necessarily, you're right.
Greg:Uh, I, I want, I want Claude to get access to that,
NR:Did it change the behavior of, uh,
NR:your, uh, team in any way as
Greg:Yeah, I don't-- Yeah, it's a, it's a great question.
Greg:I'm not sure I know, and, and maybe, and maybe we should ask.
Greg:Pr- probably in some way.
Greg:It's, it's a little bit like Granola.
Greg:We are running-- We are also, and I'm curious if you are, we're running Granola
Greg:now for every internal conversation
NR:Oh, really?
NR:Okay.
Greg:time.
Greg:I-- 'cause I-- want the, I want Claude to, I want Claude
Greg:to have access to that context
NR:Yeah.
NR:Okay, and that has not changed the
Greg:I, I
Greg:don't, I don't think it is.
Greg:Now again, we're smaller, you know, and it's always easier when you're--
Greg:to do these things when you're smaller.
Greg:Uh, we are fully remote.
Greg:Uh, and, um, I, I think it, it, it might have changed the behavior.
Greg:And again, I, I'm not sure I, I notice.
Greg:I do think, though, the team sees the benefit.
NR:Wow, that's amazing.
NR:Okay, now I'm going to look at…
Greg:I, I'm ra- I'm raising the bar on you,
NR:Okay, I love it.
Greg:Do you have an AI chief of staff?
Greg:Do you yet, do you yet have an AI chief of staff?
Greg:Okay, Garima, you, you gotta do this,
Greg:Okay
Greg:All right.
Greg:But, a- and when you, when you build an AI chief of staff, and by the
Greg:way, both, uh, OpenAI and Anthropic have made those, like, off-the-shelf
Greg:agents 'cause they can tell it's a great, it's a great use case, right?
Greg:When if you
Greg:connect…
Greg:But, uh, but
Greg:the more, the more context you connect, the, the more, uh,
Greg:performant the chief of staff is.
Greg:So I've been running an AI chief of staff about six months now.
Greg:I can't…
Greg:I would not live without it, and in fact, we've deployed it to
Greg:every employee if they want it.
Greg:They can turn on their own AI chief of staff, and that's one of the reasons.
Greg:And applications like that or agents like that are so much more valuable
Greg:when they have access to more context.
NR:You know, it's interesting that when, uh, we were
NR:thinking in terms of, you know, our next, uh, uh, product for AI, actually we were
NR:thinking of building a co-CEO or a board member, AI board member, or a AI co-CEO.
NR:So it's very interesting that I didn't know about the, uh, out of box, uh,
NR:AI chief of staff being available.
NR:That's right.
NR:I, I see that value Yeah, absolutely
Greg:Yeah, yeah.
Greg:I mean, that's kind of how I start my day now.
Greg:I don't wanna start my day in my inbox.
Greg:I mean, I do get my, my chief of staff sends me its update in the morning,
Greg:uh, via email and via Slack, so I'm still starting my day in the inbox.
Greg:But I do think in a year or two, I think it's, I, I think it's gonna be archaic.
Greg:Uh, it will look archaic that we start- that we're starting our days either,
Greg:in Slack or, or, or in, in our inbox.
Greg:I think we'll be starting our day with our chief of staff, with our AI chief of
Greg:staff, who will help us manage, you know, the, the day and the week's priorities.
Greg:I met an entrepreneur, uh, early, early this week, Greem, I, who I
Greg:interviewed for the podcast, and he's probably, he's probably m- you
Greg:know, early to mid-20s, running a $200 million a year revenue run rate
Greg:company, uh, about 130, 150 employees.
Greg:He actually starts his day inside a Claude desktop in- inside the app, and
Greg:with a conversation, uh, with Claude about what he should work on that day.
NR:Superb.
Greg:So, like, having a real sort of dialogue and, and Claude is looking
Greg:back, uh, over the previous week and in, and over the previous 24 hours, and
Greg:basically coaching this young CEO on, you know, how he sh- how he should sh- how
Greg:he should be spending his time that day.
Greg:I, I just found that fascinating.
Greg:I think we're all gonna end up working, working in some way like that, you know,
NR:I something here, I'm going to do that.
NR:I'm going to start doing that.
NR:That's very fascinating and a very productive way to
NR:get started, uh, right?
Greg:Yeah.
Greg:All right, I wanna talk about you a little bit more because, uh, first of
Greg:all, you have a co- you're a co-CEO.
Greg:tell me how that's working or how does it work?
NR:So it's great.
NR:I think I have another CEO who is in charge of the product, uh, and
NR:essentially I just joke, whatever he doesn't want to do, I just pick it up.
NR:so deliver on it.
NR:So, but our lines are pretty clear.
NR:I, uh, run the business and, uh, he's on the product
Greg:Got it.
Greg:you knew him?
Greg:Like there's three, there's three, three co-founders, I think.
Greg:Uh, how long did you know each other before you started the company?
Greg:I'm just curious.
NR:So I'm married to one of the co-founders, so
NR:that's how I knew him, and I was involved in his previous startup.
NR:And, uh, so this is-- He always jokes that, uh, I forced him into this
NR:because I wanted to do this startup, and I'm not an engineer by trade, so I
NR:pulled him into building a remarkable product to me, uh, well, for me,
NR:That we
Greg:it.
Greg:That you could, that you
NR:Yeah.
NR:Yeah.
NR:and other co-founder, we had known him from the previous startup, so it was
Greg:Oh, great.
Greg:So you guys had, you, you, you, you'd worked together and in
Greg:one case you'd lived together.
Greg:Uh, and then, uh, this is about 10, about 10 years ago.
Greg:Ha- So I w- I, I did, I did, I started one company with my wife.
Greg:Uh, it was small and, and v- very modest.
Greg:Uh, we had a very small exit.
Greg:We never wanted to do it again, to be honest, uh, in that we
Greg:just found it har- you know, everything blurred obviously, right?
Greg:Uh, you talked about the bus- you talked about the business all the time.
Greg:Uh, you, you guys have been able to obviously manage it.
Greg:Uh, and, and yeah, and, and n- now I wanna talk about your kids, 'cause it
Greg:can't be easy being a co-CEO of a very now fast-growing startup and managing
Greg:all your, your family obligations.
Greg:You have how many kids?
NR:Uh, I have two boys,
Greg:and do you wor- and, uh, uh, do you worry about them in terms of, uh,
Greg:maybe outsourcing their thinking to AI, outsourcing their homework to AI?
Greg:What, what, what are your thoughts on…
Greg:W- when, when I am talking in, in, in larger public forums, I would say it's the
Greg:number one question I always get, Garima.
Greg:It, it, and almost the first question as well, which is, what about our kids?
Greg:What about, you know, the, the, the value of education, and what should
Greg:they be studying, and, and what career path, path should they choose?
Greg:With a 15-year-old, I'm sure you're beginning to think about so- some of
Greg:those questions and, and, and maybe of course college is, uh, not too far away.
Greg:Any thoughts, uh, th- uh, ar- around for, and advice for parents?
NR:Yeah, it's, it's a very hard question, you know?
NR:And especially, like, I, uh, follow, uh, you know, certain people, Elon and Vinod
NR:Khosla, and they've been talking about 2036 there won't be any need for colleges.
NR:And I'm like, "Oh my God."
NR:But, uh, I think the only thing that we can equip our kids is with
NR:critical thinking and reasoning and, uh, giving them the right foundations.
NR:And, uh, I mean, I am telling my, uh, uh, sophomore just to be in pure sciences
NR:because you just cannot go wrong.
NR:At least you learn about the universe around you, right?
NR:Uh, and those are just good things to live your life by.
NR:And, uh, have strong foundations in mathematics.
NR:So those things will never go outdated, no matter AI or non-AI, so at least, uh,
NR:you'll be good set well foundationally for no matter what challenges come your way.
NR:So that's, that's my, uh, elder one.
NR:My younger one, on the other hand, he knows he wants to become…
NR:He's like, he wants to become a software engineer from the day he was born.
NR:So I have been talking to him, and he was just very anti-AI, that "I'm
NR:going to build my own things by hands."
NR:And I'm like, "You will have no role to play."
Greg:Right.
Greg:Right
NR:he's getting used to the idea of, uh, working closely with AI.
NR:He's still in middle school, but, uh, he just wants to be…
NR:He just wants to build things.
NR:And I'm like, "Okay, let's get through high school and we'll get there."
Greg:Garima, thanks for joining us.
Greg:You are clearly building a future super company.
Greg:Well, a- actually, it might already be one because what, what I t- what
Greg:I say to most CEOs is that growth is great, you know, revenue growth,
Greg:but you need to deliver earnings.
Greg:Like super companies are not ones that just grow quickly.
Greg:Anthropic is not a super company yet.
Greg:Anthropic looks like it could be a super company, and OpenAI as well, and I'm
Greg:looking forward to the IPOs 'cause we can learn a little bit more about actually
Greg:what's going on inside these companies.
Greg:But they're a long way from super company status 'cause y- to be a super company,
Greg:that means you gotta actually make money, and probably do it for not just a few
Greg:quarters, but a bunch of years, like the previous generation did, you know,
Greg:Google, Netflix, AWS, Amazon, and so on.
Greg:You're already profitable, so you're, you're off to a very good start.
Greg:Hopefully, uh, you can sustain that, and th- then you're…
Greg:You know, do that for a decade, you're a super company, or another decade 'cause
Greg:you've already been around 10 years.
Greg:Do it for 10 more years, Garima, you'll, you'll be one
Greg:of the world's super companies.
Greg:And, uh, and help us write the playbook.
Greg:No one knows how to run a super company or how to build one.
Greg:We're all doing it real time.
Greg:So as you learn what works, please share it with the rest of us
NR:Absolutely.
NR:And every journey is unique, right?
NR:Success and failures come later.
NR:I think just to be on the journey is, uh, great in itself.
NR:So yeah,
Greg:Agreed.
Greg:Uh,
NR:Yeah.
Greg:thanks for joining us.
Greg:Good to see you.
