A strange thing is happening inside engineering teams right now.
The best developers aren’t writing more code.
They’re writing less—and shipping more.
Somewhere between ChatGPT, Cursor, Claude Code, and a new wave of AI agents, the job quietly flipped. And founders who don’t notice this shift are about to build companies the old way in a world that’s already moved on.
This episode with Karan Grover, co-founder of Kanu AI, is really a front-row seat to that transition.
The Moment Everything Broke (and Got Better)
Karan tells a story that sounds almost fake:
A team scoped a product to take three months.
With Kanu, it shipped in two days.
That’s not a 10x improvement.
That’s a time-compression event.
And when time compresses like that, everything downstream breaks:
Roadmaps become obsolete
Hiring plans stop making sense
“Engineering velocity” stops being a constraint
It’s like going from horses to cars—not faster horses.
The Myth of “AI Will Replace Engineers”
Let’s kill that idea quickly.
What’s actually happening is more interesting (and more uncomfortable):
Engineers are being promoted… without permission.
When AI handles:
Code generation
QA testing
Deployment
Infra setup
What’s left?
Architecture decisions
Tradeoffs (latency vs cost vs scalability)
Product clarity
System design
In other words: the hard parts you used to avoid.
Karan describes this as engineers becoming “staff-level by default.” And if you’ve ever worked with a great staff engineer, you know—that’s not about typing speed. It’s about thinking.
The Real Bottleneck Was Never Code
Here’s the punchline most people miss:
Code was never the bottleneck.
The real bottlenecks are:
Poorly defined problems
Legacy systems no one understands
Organizational inertia
Endless planning cycles
Enterprise teams don’t struggle because they can’t write code.
They struggle because they can’t decide what to build clearly enough.
That’s why Kanu doesn’t just execute—it forces clarity.
It asks:
What are you actually building?
Who is it for?
What are the constraints?
Sound familiar?
It’s basically turning every engineer into a mini product thinker.
Amazon Got One Thing Very Right
Karan’s time at Amazon shows up everywhere in how Kanu is designed.
There’s a concept called PRFAQ (Press Release + FAQ), which forces teams to clearly define:
The product
The customer
The outcome
Before writing a single line of code.
That idea is now baked into AI systems.
Which is kind of wild.
We’re moving toward a world where:
You don’t start with code
You start with intent clarity
And the AI refuses to proceed without it
So instead of “move fast and break things,” it’s more like:
“Think clearly, then move insanely fast.”
The Hidden Startup Lesson (Learned the Hard Way)
Before Kanu, Karan built a startup in housing finance.
It failed.
Not because the idea was bad—but because it depended on:
Banks
Government programs
Interest rates
And when those changed, the business collapsed.
His takeaway is simple and brutal:
External dependencies can kill you overnight (we know this lesson well at Team Ignite).
This is one of those lessons founders only truly learn after getting burned.
And it’s especially relevant now—because AI makes building easier, but distribution and control still matter.
The Future: One Engineer, Many Agents
Imagine a team that looks like this:
1 product-minded engineer
3–5 AI agents handling execution
That’s not a prediction. That’s already starting to happen.
Roles are merging:
PM + Designer + Engineer → one orchestrator
Dev + QA + DevOps → one workflow
And instead of hiring more people, companies will:
Hire fewer, better thinkers
Equip them with stronger AI tools
It’s not about replacing teams.
It’s about shrinking the distance between idea and reality.
What This Means for Founders
If you’re building today, this shift changes a few things:
1. Speed is no longer your moat
Everyone is fast now.
Clarity becomes the advantage.
2. Hiring changes dramatically
You don’t need 10 engineers.
You need 1–2 exceptional ones who can think in systems.
3. Legacy is your biggest risk
Greenfield is easy.
Modernizing old systems is where the real opportunity (and pain) lives.
4. The bar just got higher
If execution is cheap, only:
Insight
Taste
Judgment
will differentiate you.
The Quiet Reframe
Karan’s story starts with him walking out of a medical exam halfway through because he realized he didn’t want that life.
Years later, he’s building something that removes the parts of engineering people don’t want to do.
That’s the throughline.
Technology, at its best, doesn’t just make things faster.
It removes the parts humans were never meant to enjoy in the first place.
And what’s left?
The thinking.
The creativity.
The decisions that actually matter.
If this trend continues (and it will), the future engineer won’t be the best coder.
They’ll be the best problem framer.
And the companies that win won’t be the ones who build the fastest.
They’ll be the ones who know what’s worth building at all.
👂🎧 Watch, listen, and follow on your favorite platform: https://tr.ee/S2ayrbx_fL
🙏 Join the conversation on your favorite social network: https://linktr.ee/theignitepodcast
Chapters:
00:01 Introduction & Karan Grover Background
02:25 Early Career Shift: From Pre-Med to Computer Science
02:58 Building for Impact: SoundVision & Accessibility Startup
03:47 Winning the Ericsson Innovation Award
04:58 Learning Concise Communication as a Founder
07:34 Early Career Lessons from Deloitte & Best Buy
08:48 Journey to Amazon & ML Engineering Experience
09:22 First Startup & YC Experience (Winter 2022)
11:17 Building a Housing Startup & Lessons from Failure
14:07 Returning to Amazon & Rediscovering the Founder Itch
15:54 Origin Story of Kanu AI
17:01 Finding the Right Co-Founder
19:41 Building an AI Engineer: From Prompt to Production
22:53 AWS Marketplace, Trust & Enterprise Adoption
25:01 Human-in-the-Loop AI & Engineering Abstraction
26:23 Amazon Principles & System Thinking in Startups
27:37 Early Customer Wins & 3 Months to 2 Days Insight
29:03 Real Bottlenecks in Software Development
30:06 YC vs A16Z Speedrun: Accelerator Insights
34:50 Where AI Engineering Tools Win First
35:42 The Future of Software Development Roles
39:02 Deploying AI Agents in Enterprise Environments
41:18 Long-Term Vision for Kanu AI
42:58 Skills, Beliefs & Founder Lessons
45:47 Surprising Use Cases & AI Capabilities
Transcript
Brian Bell (00:01.166)
Hey everyone, welcome back to the Ignite podcast today. We’re thrilled to have Grant Grover on the mic. He is a three-time founder, YC alum, former Amazon ML engineer, now the co-founder of Canoo AI, where he’s building an AI engineer that takes a company’s roadmap from prompt to production. Very cool. Thanks for coming on.
Karan Grover (00:16.949)
Yeah, thanks for having me, Brian.
Brian Bell (00:18.872)
Well, I’d love to get your origin story. What’s your background?
Karan Grover (00:22.543)
Yeah, so I started off in a suburb outside of Vancouver, Canada, born and raised, and then went to the University of British Columbia. I started off doing medical actually, and then switched over to computer science after taking a comp sci course and a little funny tid.
Brian Bell (00:39.699)
I have here that you walked out on your MCAT. What does that mean?
Karan Grover (00:42.925)
Yeah, yeah. I was sitting in the MCAT and I was about 30 minutes in and I just realized I don’t like this. I didn’t like studying for this. I didn’t like any part of this. And so I stood up and I went to the person who was kind of proctoring an exam. I said, hey, I’d like to go in. And she started yelling at me. said, if you want to go to the bathroom, you have to ask my permission. I said, no, no, no, I don’t want to go to the bathroom. I just want to leave. And then she thought that I was leaving because she yelled at me.
Karan Grover (01:12.762)
And so she took it back. said, no, no, it's okay. You know the bathroom, but I'm just telling you, you have to let me know. And I was like, no, actually, I just don't want to do this anymore. And so I left and I called my best friend and I told him, Hey, I'm going to get a job at Google. And I didn't get a job at Google, but I did end up at Amazon.
Brian Bell (01:19.064)
Wow. I mean, it's next best thing. Google's so hard to get a job at, especially when you're young. What was your undergrad in? It was like pre-med?
Karan Grover (01:35.867)
So it started off as pre-med and then ended as computer science. yeah, bachelor of science in comp sci at UBC in Canada. Yeah.
Brian Bell (01:38.955)
OK, nice. Wow. Well, I think we talked about this, but I’m Canadian. I was born in Seattle. My mom was born in Vancouver. And my grandparents are from Saskatchewan, of all places. And then my great grandparents are from all over Europe. But yeah, so I got Canadian roots as well.
Karan Grover (02:00.069)
Yeah, for a little while there, we were hiring Just Canadians accidentally. Like we just would hire and they'd make it through the interview process and somehow it come up and we'd realize we had like a four five person company of Just Canadians accidentally.
Brian Bell (02:13.826)
Oops. That’s a bias in hiring, know, hiring people like you. So what do you think were the earliest experiences that pulled you away from medicine and more towards like engineering and problem solving?
Karan Grover (02:25.829)
Yeah, you know, I definitely had really strong problem solving roots from my parents. They’re both, you know, very analytical, very grounded in solving problems. But I think also just going through an experience in comp sci for the first time, it just felt like going through and solving puzzles all the time. And so that that’s kind of what swayed me towards it. Just felt more fun.
Brian Bell (02:48.653)
Yeah. So you had a company before that helped blind individuals hear their surroundings. That’s pretty incredible. What drove you to work on that problem so early in your career?
Karan Grover (02:58.033)
Yeah, I’ve always been one to want to apply things right away. So as soon as I feel like I’ve learned a new skill set, I just really believe that the best way to hone it is by practicing it. And so when I felt like I had a good grasp on computer science, I thought, OK, let me go through and find something to apply this on. And a newsletter came across from the comp sci department and it said that this other comp sci student had this problem where there was a colleague of theirs who was blind and had come in one day to work because they’d hit their head on a construction side. And they wanted to solve for that problem. And I really wanted to help them do that. And so I joined their startup and got to applying.
Karan Grover (03:38.19)
That’s awesome. I have here in the notes that you competed in and won an Ericsson Innovation Award. What was that about and what kind of lessons did you take away from that?
Karan Grover (03:47.205)
Yeah, we actually got to meet the Prince of Sweden after we came first place. And so we got flown out there. And it was a really fantastic experience. We got to compete against 862 other universities and research teams from around the world, including MIT was there as well. And we got to showcase our research. And at the end of the day, we were the lucky winners and got to really get our name out there.
Brian Bell (04:11.63)
That’s really cool. And so when you think back to sound vision, what was that, what worked and what didn’t and what stayed with you in that experience?
Karan Grover (04:58.161)
Yes, important story. Okay, do you want to start with the question again or just should I kick it off with the answer? Okay, sounds good. Yeah, so when I was at SoundVision, the person who had originally started the company, the very first founder, when we were at the Ericsson Innovation Awards, actually, they sat down with me and they told me that I rambled too much. And that was very hard feedback to get at that time. And I didn't really agree with them. thought everything that I said was creating value. But I started to reflect and really look back at what I was saying. And I realized like, Hey, yeah, I would kind of just drag out for five minutes at a time on the topic. he said, you really need to look at every word that you're saying, and almost treat it like every word you say costs a quarter. And would you say as much? And so that's when I started to refine my sentences more. And that's a lot of the positive feedback I've started to get now later in my life is being able to take large topics and boil them down to a few sentences.
Brian Bell (06:00.718)
That's such good advice. I think about all the really amazing executives I've worked with over the years at Amazon and Microsoft and other places, unicorn startups. They don't say a lot. Yeah, they're just kind of quiet types. They'll ask a question. They might say a few words. They'll have these dramatic pauses, kind of Obama-esque kind of pauses, and not just blabber on.
Brian Bell (06:29.57)
Right, that's really good advice. Every word costs a quarter.
Karan Grover (06:32.369)
And it feels so, yeah, exactly. And it feels so impactful too over time because when people know around you that when you’re saying something that you’ve thought through it, they’ll pay a little bit more attention.
Brian Bell (06:46.882)
Yeah. It’s like if you silently pause, people will actually fill the space with the project in their own feelings and thoughts and emotions. And then you kind of get to pause and react to that again. And then when you finally say something, they’ll really listen, right? Yeah.
Karan Grover (07:02.385)
told him. What a great negotiation tactic as well, just saying nothing and then let’s.
Brian Bell (07:10.604)
Yeah, just say nothing and just kind of sit there and think about it. People are uncomfortable with silence, even just now, like we’re recording the podcast. You went silent for a second. I was like, I waited about five seconds. I was like, OK, maybe he didn’t hear the question anyway.
Karan Grover (07:14.203)
Totally. Yeah.
Karan Grover (07:22.437)
Yeah. Yeah. Yeah, that’s that’s a that’s a skill on its own, to be honest, is just being really comfortable with pauses. It’s not one I’ve mastered, but it’s a good skill.
Brian Bell (07:34.638)
Oh, me neither. I’m a big talker. That’s why I do a podcast. So you were also at Deloitte and Best Buy, and you saw large orgs up close. What did those environments teach you about shipping and leadership?
Karan Grover (07:47.237)
Yeah, definitely. So I was always one to kind of step outside of my comfort zone. I would always do that even at a young age. at Deloitte and Best Buy, something that I learned is just ask. If you go up and ask something that you think is absolutely outrageous, sometimes you get back, sure, we’d love to do that. And one of the examples is actually the internship itself. So Best Buy Canada at that time didn’t have an internship program. And so just a lot of pestering and a lot of asking, I was able to get in as their first intern in their engineering department and now dozens of other engineers have come after the fact. But that’s one example. The second is just going up to directors and VPs and even the president and just saying, hey, I want to have a conversation. Here’s what the agenda is. You’d be surprised how often it works.
Brian Bell (08:43.918)
That’s amazing. And then somehow you ended up at Amazon. What’s the story there?
Karan Grover (08:48.527)
Yeah, so I got to go to Amazon, actually just went through their normal channels. I put my resume into the bucket, went through the interview process and was lucky enough to land an internship and got to work on some very, very cool projects over there and make a lot of impact and carried through the just ask mentality. So I got to work with VPs and directors and folks that my fellow colleagues didn’t have access to at the time.
Brian Bell (09:13.07)
And then at some point you went to YC the first time, right? Because you’ve been to YC twice, right? And we’ll cover the second time, but the first time was back in winter 22. Tell us about that experience.
Karan Grover (09:22.799)
Yeah, so I’ve been to YC once and then A16Z speedrun the second time. Yeah. And the first time was, yeah, winter 22. So that experience was amazing. We got the opportunity to go in as Canadian founders, one, start the company in Vancouver, Canada as well, and got to really experience speed for the first time in, I think, my life. We went from this idea of, we’re going to help Canadians buy houses that they’d never be able to afford or purchase otherwise, and went within months to working with some of largest banks in Canada. working with VPs, getting million dollar facilities for purchasing houses and actually selling houses. And that kind of speed was just, looking back, I still don’t believe it.
Brian Bell (10:13.294)
And was that all kind of the wind in your sails was the YC network basically, going through YC and being able to kind of get those introductions and stuff?
Karan Grover (10:22.289)
Yeah, so I think YC really pushed that just ask mentality much further. So don’t just ask of your director or VP. It was, hey, you can just reach out to the mayor of the city and ask for a conversation or ask Brian for an introduction to the most powerful person that he knows. And Brian will usually help you out because he wants to see you win too. so that’s, yeah, exactly. Yeah, that helps. And so that’s what we did.
Brian Bell (10:44.098)
Well, especially if you’re in the portfolio, right? That’s what I tell all my founders. I’m like, I don’t care who it is. I’ll introduce you to the president of United States. I could care less.
Karan Grover (10:54.917)
I think a lot of people, not everyone, but a lot of people want to see founders win and, you know, we’re taking a chance and they’re willing to help out however they can. So we actually got to meet mayors, leaders in government and large banking execs. And that’s how we were able to get the facilities to be able to actually sell homes in Canada.
Brian Bell (11:17.292)
What happened after the what what what was the kind of the the story of true true place and and did you go to door dash and Amazon next or what was the kind of chronology or.
Karan Grover (11:29.071)
Yeah, so I've been at Amazon a few times. So I started off as an Amazon intern, and then I went and I worked at Amazon for about a year and a half. I got promoted to SD2. We worked on some really cool stuff there, like two hour shipping and even 15 minute ultra fast shipping and worked some cool problems. And then after that went to DoorDash to work on, you know, similar problems around last mile fulfillment and, and monetary refund automation and things like that. And then I quit. DoorDash nine months in, although I really liked it, but we got into YC and so I really wanted to do that as well.
Brian Bell (12:03.682)
Also a YC company too, DoorDash, so I’m sure they understood.
Karan Grover (12:07.023)
Yeah, exactly. Tony Zoo is awesome. And yeah, and the idea was kind of, you know, I watched my parents make a lot of their wealth coming from nothing to really establish in Canada because of housing. They knew that as soon as they got here, they wanted to buy a house and then they slowly upgraded their homes over time and worked really hard. And I wanted every Canadian to have access to that. And it just felt like it was so out of reach because the income was growing at 3 % year over year, but houses were growing at 7 plus percent year over year. And so it just felt like a problem that no one was ever going to win. And so the premise of the business was what if we just give you the last 50 K that you need for your down payment and you’re able to purchase this house, live in it, build a family in it and all these amazing things that come with home ownership. And then on the back end, when this house appreciates and you make a bunch of profit, then we’ll take a small sliver of that profit that you wouldn’t have otherwise had without us as the return to the fund. Yeah, exactly. Yeah.
Brian Bell (12:44.28)
Mm.
Brian Bell (13:06.706)
Plus plus the principal obviously right? Yeah Effectively like a second financing mortgage or something. I think I recall seeing your business Back when you guys did we meet when you guys are in winter 22
Karan Grover (13:20.977)
I don’t know if we met back then, but I think we talked about it recently. Yeah. Yeah, yeah. But.
Brian Bell (13:23.436)
Yeah, I mean, that was four years ago, so it’s hard to remember. I know I didn’t invest. But what do you think it was? I mean, what happened to the business? And what are some takeaways from it?
Karan Grover (13:35.119)
I think the biggest takeaway was that housing and problems where you have to take external dependencies are very difficult. So we took an external dependency on a very large bank. We took an external dependency on the government, on a government’s financing program. And any of these change, policy, interest rates, taxes, it crushes the business.
Brian Bell (13:56.547)
politicians, elections, yeah.
Karan Grover (13:59.673)
number one lesson I learned out of that business. And that’s what happened ultimately to us. We had interest rates change and the finance of the model didn’t work.
Brian Bell (14:07.438)
No. Yeah, this is this is something I’ve actually integrated into my startup scorecard. I have this startup scorecard I use to score startups with AI. And I have one that shut down recently. And, you know, we analyzed it and did a postmortem and, you know, I always need the founder. like, what did we get wrong here and blah, blah, And that was like one of the key things I’ve added to my valuation criteria recently is the regulatory single point of failure. You know, like what is the black swan or even gray swan event that can like come in and wipe this startup out. And when you have like a banking partner or a government partner or both.
Karan Grover (14:44.625)
Yeah. And it’s not predictable because it’s, you know, we did our back testing. We said in the last 30 years, the probability of it getting so bad that we have to shut the business down is like a sliver. And that sliver is what happened. So, Yeah.
Brian Bell (15:00.95)
Yeah, nobody anticipated rates to, you know, triple overnight or whatever it was. So. And then what did you do after that?
Karan Grover (15:04.965)
Yeah. Yeah. So I went back to Amazon after that and I worked on a lot of different, really cool problems. got to work on machine learning. We got to work on post-training bedrock for Amazon as well. And I also got to work with some really, really phenomenal mentors there. And that was the key part for me is I said, if I’m ever going back to not founding a business, I need to be learning 10 X what I would be learning in university or founding a business. As long as that was true, I was there. And then, you know, the itch kind of of being a founder just continued and continued to grow until one day I said that, you know, I need to leave again.
Brian Bell (15:44.564)
Yeah. And so you’re there and you’re seeing this thing with ML scaling. What’s the origin story around Canoe?
Karan Grover (15:54.085)
Yeah. You know, like every other engineer, as soon as the big chat GPT kind of bump happened, I started to see how I could apply AI to my life. And one of those things was obviously cursor, client, clock code, using code gen tools. And it was very fantastic. I mean, when I first started using them, it felt like my job was gone. Like surely engineers won’t be needed in the future. And then I started to see the problems that I would be solving day to day. It was more architecture, leadership, design. And I was a senior engineer at that point and I still felt my role getting elevated day over day. I felt like a staff engineer and then a distinguished engineer. And the piece that was missing from Cursor and Cloud Code was there’s so much of this job that it doesn’t do for me. The testing, the deployment, the terminal commands. There’s just so much of it where I’m like, that’s so easy. Why don’t you just do that for me? And so when I couldn’t find that tool, I said, okay, let me quit. And I’m going to start to build this myself. And that’s what we did at Condu.
Brian Bell (16:57.657)
So how did you convince your co-founder to join you?
Karan Grover (17:01.361)
I have been talking to Harsh since 2018 or 17, I think. We worked at Deloitte together in their first AI department. It was called Omni AI. It was the first of its kind in Canada. We got to work in the first year of it.
Brian Bell (17:12.048)
yeah. I think I remember that one. Yeah. Because I was leading AI at Amazon back in 2018 and building a bunch of AI before that.
Karan Grover (17:22.577)
And yeah, so they’re like a very strong AWS partner. so, yeah.
Brian Bell (17:25.324)
Right, yeah. That’s why I remember Abney. kind of remember the, I mean, it’s been seven years since I’ve heard that phrase, but yeah. Yeah.
Karan Grover (17:30.853)
Yeah, that’s fantastic. That’s really cool that it’s kind of like made its way through the market all the way to Amazon. Yeah, we got the chance to work on some really cool problems. We working on computer vision for large grocery chains and all those fun applications. And Harsh and I became very close at that point and we stayed in touch throughout our careers, you know, calling each other maybe once a week, even though we were in different locations. And the commonality was that every single week we would talk about how one day we’re going to start a business together and one day we’re going to start a business.
Brian Bell (17:57.016)
You mean you’re a guy and he’s a guy and you guys would talk to each other on the phone?
Karan Grover (18:01.241)
Literally, we would call each other and we would just be like, all right, we’re going to start a business one day. We’re going to start a business one day.
Brian Bell (18:07.148)
because I never talked to any of my guy friends like maybe a few times a year you know
Karan Grover (18:10.563)
Really, yeah, you know, I only have Harsh to thank for that because I don’t think I would ever pick up the phone and call somebody, but Harsh calls all of his friends on a weekly basis. He is so good at staying in touch with people. Yeah, good, good lesson, think, for all of us. And he called me.
Brian Bell (18:20.716)
Wow, that’s cool. That’s really cool.
Brian Bell (18:28.374)
I mean, the way I stay in touch with my really old my oldest friends is we play video games together. So I talk on the discord basically with them while we’re playing a video game, you know.
Karan Grover (18:33.381)
Yes. Discord. Before Konu, Harsh and I also used to play PS5, but the PS5 is like fully collected dust over the last like seven months.
Brian Bell (18:44.374)
Yeah, I mean, if you’re a founder, can’t you have no time. have to be a VC to play video games. You can’t be a founder.
Karan Grover (18:48.751)
Yeah, you said it, not me. It’s been so interesting, but like literally one of those days I was like, called him instead of him calling me and I was like, hey, we’re starting a business now. Like I’m quitting my job. I’ve already quit my job. You quit your job. And before I had finished the sentence, he’s like done. Yeah. And I was like, do you want to know what the idea is? And he’s like, yeah, okay, tell me the idea. We hashed it out.
Brian Bell (19:11.886)
He was like ready to join. He’s like, all right, I got this idea. Let’s do it. And he’s like, all right, let’s do it. And I’m like, wait, let me tell you about the idea first.
Karan Grover (19:15.654)
Yeah. Let me tell you, yeah, yeah. I was almost like, wait, I feel like you should know the idea before you say yes, but I think him and I had just, like, we were just so close and it’s always something that we spoke about. And so when the time came, it was like a no-brainer for him, no-brainer for me.
Brian Bell (19:22.904)
Yeah, before you jump. Yeah.
Brian Bell (19:33.004)
Yeah, what’s been the biggest engineering challenges that you guys have taken on taking something from a roadmap, you from a prompt to production?
Karan Grover (19:41.359)
Yeah, I think when you’re building out a coding agent, the success really comes where you can mimic a human behavior without having to verbosely write out a thousand tools for every little thing that we do. And so for us, the three big buckets we thought are engineering, which is code generation, that’s a solved problem, right? So we know how to do that.
Brian Bell (20:01.454)
Yeah, Opus 4.5 is ridiculous, right?
Karan Grover (20:05.947)
phenomenal model and really good at solving for code gen. And there’s publicly available, you know, resources about how to do code gen and optimize for that and all that fun stuff. So we built that out. We believe we built out a better version of it, but that’s not the game changer of Canu. The game changer is we said too, we’ve been working in cloud specifically like AWS for the last eight years. We know that every problem comes back to the cloud. Whether you’re abstracted away from it as an engineer because your DevOps team takes care of it, it’s still coming back to the cloud because that’s how most enterprises service their customers. And then the third was QA. It’s that unsexy term where no engineer wants to do the QA. But what’s awesome about Amazon, and you know this, Brian, is that Amazon forces you to do all three. You’re your own QA, you’re your own engineer, and you’re your own DevOps. There’s no one.
Brian Bell (20:56.686)
As an engineer that you have to do all three. I didn’t know that I did not. guess I didn’t know that about Amazon because I was by the time I got the Amazon I was one step removed from engineering. So my I was like a technical BD but they wanted somebody that built a I products to lead a I and launch stage maker marketplace while I was there. I was more like the customer facing guy on the technical team. So I did not know I did not know the engineers there are there their own QA and what else.
Karan Grover (20:58.851)
Exactly, as in SD1.
Karan Grover (21:20.163)
and DevOps. So if you look at a company like, let’s say DoorDash, you have a platform team where they take care of all the AWS stuff and you just write logic. And so now this DevOps role, I’ve always experienced it for myself, but now other engineers in companies like DoorDash are starting to experience their roles getting elevated and they now have to do QA and DevOps and engineering, things I’ve always experienced since day one on Amazon.
Brian Bell (21:51.0)
Well, especially because a lot of the logic is being generated now by the AI,
Karan Grover (21:54.947)
Exactly. So if you’re not doing the logic, then you have to think about latency and costs and architecture. It comes back to the cloud. It always comes back to the cloud. There’s a reason AWS is so successful. And so we said, okay, let’s build these three agents so that Canoe can manage all and it can kind of interconnect. So now what Canoe can do is if you say, I want to make a chat bot or I want to upgrade my existing legacy system from Java 8 to 21. What KONO will do is it will go into your AWS account. It’s a publicly approved marketplace application. You download it, it takes five minutes. Understands your AWS system, understands all your app code, your infrastructure. And then it just does it. And then once it’s completed it, let’s say we’re going with the Java 8 to 21 migration, it will call your APIs and say, oh, did I do it right? Oh, no, I forgot this one piece. Okay, I’m gonna go fix that. Okay, did I do it right now? And it just iterates on that until it’s Achieve the success criteria and gives it back to you and says, hey, I’m done. Let me know what you think. Yeah.
Brian Bell (22:53.346)
That’s amazing. So it’s like kind of done for you. How has the AWS marketplace been for you guys?
Karan Grover (22:59.217)
It’s been a little bit surprising in how long the approval process took, but I don’t mock them for it because I feel better saying, hey, we’re an approved marketplace application knowing what we had to go through to get approved.
Brian Bell (23:16.844)
Yeah, because they don’t just let anybody into the AWS marketplace. They do a pretty rigorous technical and security screening, right? Because you’re going to be deploying into customer tenants, into their infrastructure, and they want to make sure everything’s buttoned up.
Karan Grover (23:28.813)
Exactly. Yeah. And so it kind of felt like doing, you we’re doing our SOC 2.0 Type 2 and we’re in the audit phase there. It felt almost as rigorous, if not more rigorous, to get through the Marketplace application.
Brian Bell (23:39.49)
That’s amazing. then how has the do you think that makes it easier to acquire customers or for customers to find you? How is that? Like, what’s the I know the technical impact. What’s what about the business impact?
Karan Grover (23:50.587)
think it’s about trust. So we can go to a Fortune 500 company and say, yes, we’re in the audit for SOC 2 Type 2, but also AWS approved us. And we’re putting this application in your enterprise account with a lot of information that you’d like to keep safe. So not only do we have those security checks, but also, Canu has no egress outside of your AWS account. Everything is completely local there. And so that is really a huge burden.
Brian Bell (24:16.071)
So you’re deploying like Kubernetes, like containers inside their tenant basically.
Karan Grover (24:20.273)
Exactly. So we have our full stack, all model code, everything is all inside of their stack. And so they have to pay for the hosting, which is nominal if you’re a large enterprise.
Brian Bell (24:29.986)
Yeah, right. That’s cool. So it’s like a software license. How do you guys price it?
Karan Grover (24:35.141)
Yeah, we price it by AWS account. So how many apps you download?
Brian Bell (24:39.507)
app to download. What does that mean?
Karan Grover (24:41.669)
Yeah, so let’s say you have five different products that have five different AWS accounts. So we’ll charge you per the five AWS. So every time you download the app onto an account, you get charged monthly.
Brian Bell (24:51.726)
Got it, got it. Wow. That’s amazing. So how do you balance the abstraction and letting an engineer stay in the loop where needed?
Karan Grover (25:01.935)
Yeah, this is the problem that we are still perfecting. And I think we’ve done it better than every other company out there. What we realize is that key insight that I mentioned earlier is our roles as engineers are getting elevated. But one thing I will never let happen as an engineer is an AI deploy something to my production without me understanding what just happened. And so how do you find that gap where I don’t have to review every line of code, but I feel very confident about what’s inside of this code base. And so we have built up this system where it’s human in the loop. It will ask you to make key decisions. But one thing that we found is at Amazon, at AWS, they have so many different offerings. And there might be one kind of hidden in that corner that I don’t even know about as an engineer that could have solved all of my problems. And so what Connor will do is he’ll make you recommendations. I’ve looked through the internet, I’ve looked through AWS, this are your top three solutions to solving this problem. Which one would you like? And then it gives you the option to select and then it gives you the option to dive deeper. How would you code this out? Kanu will tell you. How would you do this? How are you gonna handle latency? And you just chat back and forth with it about these problems. And then once it feels, once Kanu feels confident and you feel confident that you’ve understood the problem properly and selected the right solution, then it continues forward.
Brian Bell (26:23.63)
How much of Amazon culture do you think you’ve ingrained inside of Canoo? Like, because of what you’re describing almost to me sounds like a PR FAQ like, like doc review.
Karan Grover (26:32.721)
Yeah, yeah, it’s, whether it’s intentional or not, I think Amazon has some really great principles. You make up your own mind about the company, but their values that Jeff Bezos kind of instilled at the beginning, I don’t know about you when you were there, but every leadership principle I knew in and out, and I could tell you everything about it, right?
Brian Bell (26:54.779)
yeah. Well, and people will pull them out. They’re like, I’m going to disagree and commit here. like, you know, like people just kind of pull them out, you know.
Karan Grover (26:59.675)
Totally.
Karan Grover (27:03.365)
And I think that’s what Jeff Bezos got right in the beginning days is he built out these systems where whether he was in the room or not, or whether the customer was in the room or not, those systems would be adhered to. The hiring is always the best because of barraisers. Security is always done because of security reviews. And this system he’s built on, and I think those systems really help startups succeed. And so that’s why we kind of follow some.
Brian Bell (27:27.713)
Yeah, I love that. What’s been the biggest surprise about building Condu so far? Like how customers have used it or just any surprises?
Karan Grover (27:37.935)
We got this one customer that we onboarded. They had scoped a product at three months with cursor to completion. And they told me a couple of days later, they got it done in two days to production. And I knew we had something special, but holy smokes when they told me that I was shocked. And it was a Fortune 500 company too.
Brian Bell (27:58.669)
Wow, so they had this project, they’re gonna build this thing with cursor. And they thought it’s gonna take us 90 days to fully build this, test it, deploy it, make it scalable in production, production ready, fully QA’d, all the infrastructure, dev opt out, if that’s a word. And they did it in a couple of days with your platform. That’s amazing.
Karan Grover (28:02.235)
Yeah, with cursor. Yeah. Yeah.
Karan Grover (28:15.259)
Yeah, two days to production. Yeah, I was absolutely shocked. And so now we speak to, you know, we’re speaking to Fortune 50s, for example, and we say, okay, what’s your problem? What are you facing? What’s the engineering department look like? And, you know, these companies have 5000 plus engineers. And it’s the same problem. Hey, you know, my, my engineer will do this with cursor, but then there’s like this huge part that they can’t do and they tell me they can’t do it. And so now my roadmap is delayed. And so what if you didn’t have to. What if while you were sleeping, Kanu was testing your Jackson upgrade?
Brian Bell (28:53.591)
Yeah, that’s really amazing. What have you learned from your early customers about the real bottlenecks in software development teams?
Karan Grover (29:03.525)
Yeah, it’s tough. You know, I think us engineers are so good at solving ambiguous problems. I mean, that’s kind of why we’re in this role. But I think that sometimes we get wrapped up in company like enterprise. For example, if you’re Microsoft, let’s say you’re fantastic at building software, but you also have software that’s been around for 15 plus years that nobody knows the ins and outs of. And eventually it starts barking and you say, okay, well, let me just staff a team of five engineers on this and they’ll get it done. And yes, they will, but those engineers aren’t happy about it because we want to be solving problems that impact the world, not how to do system upgrades.
Brian Bell (29:46.177)
Yeah, turns out the Python versions from 2016 and, you know, it’s barely running. This is kicking off codes left and right.
Karan Grover (29:51.601)
Yeah, and like if you have an AWS account, like you see the emails, how many times a week do you get like notice of deprecation, notice of deprecation or like end of version life? And you’re just like, geez, like I don’t want to do this.
Brian Bell (30:06.465)
Yeah, but that’s where you guys can shine. You can come in and kind of crawl the code base, understand, do the migration, do the testing, get it fully built out. So you’re one of the unique founders that have done both YC and Speedrun. I’d love to maybe walk us through the process of, I see a lot of people go back to YC for a second time. I even thought in the back of my mind that you went back to YC for the second time. What made you choose Speedrun the second time around? or an accelerator. And why even do an accelerator at all? Because you’ve already kind of been through one.
Brian Bell (30:34.657)
or an accelerator. And why even do an accelerator at all? Because you’ve already kind of been through one.
Karan Grover (30:39.653)
Yeah, for the accelerator question, was just to be honest, that YC experience, the amount of work you get achieved, you can achieve in a three month period. It’s just mind boggling. And it happens because you’re in that pressure cooker. Like your colleague beside you has just gone from zero to 200K ARR in the last three weeks.
Brian Bell (31:00.727)
You’re like your batch mate in your batch group or whatever. Yeah. Yeah.
Karan Grover (31:03.953)
Totally, and you’re not gonna sit there and be like, oh yeah, like there’s an excuse for it. You’re just gonna be like, no, how do I get to 200K or two?
Brian Bell (31:09.345)
Well, and then they can kind of tell you, like, we tried this marketing channel, we tried that marketing channel, we did this and that. And you’re like, cool, cool idea. you’re kind of all helping each other.
Karan Grover (31:19.057)
It’s this like weird experience, just like you said, where you’re competing against each other, but you are also helping each other. It’s like, if I can do anything to help you get a sale, I will do it. But also I want to beat you in revenue. So it’s so fun. And so I was like,
Brian Bell (31:35.245)
Kind of like gamifies. It’s almost like the, you know, like pledging for a, you know, like a frat in college or something, you know, like a brotherhood, you know.
Karan Grover (31:42.277)
Totally, Yeah, yeah, honestly, it is like it and you want the best for your brothers and then you also want the best for yourself. So yeah, it’s you know, we keep pushing. So that’s why the accelerator and then in terms of a 16 Z, know, why see is at the top of their game and a 16 Z are at the top of their game and. Five speed run five. And so I just wanted to experience both, to be honest, like maybe it’s a little greedy of me, but I just wanted to see.
Brian Bell (32:01.549)
Yeah, you’re in cohort four for speedrun? Five, okay, so it was pretty recent.
Brian Bell (32:14.167)
That’s awesome. And how do they compare? are your kind of some of your takeaways? Maybe there’s some founders listening that are applying to both or considering what maybe they have offers for both.
Karan Grover (32:23.781)
Yeah, to be honest, I don’t think you can go wrong with your decision. I’ve really liked my time at a 16 Z speedrun and it’s very fresh in my mind. So I’ll tell you about it. I had amazing partners. For example, Josh Lue, he was our partner for the cohort. I cannot imagine have built. I could not have built this business without Josh’s support and mentorship. And then amazing other partners like Macy, Tom, the Jordans, Bella, Ryan, Lester. These people that I just mentioned, helped us with everything hands on, like media, marketing, hiring. They’ll join hiring calls for you if you ask them to. And it’s just so many critical parts of business that I don’t have enough hands or brains to do. And they’ll just step in and just say, hey, we’ve, we’ve perfected this. This is our crafts. Here’s, here’s some help.
Brian Bell (33:12.565)
Yeah, we’ve done this hundreds of times. We know how to do this. That’s really cool. It almost feels like I feel like YC’s kind of veered a little bit earlier since 2022, since Gary Tan took over. It’s a lot more like teams and ideas coming in and kind of going zero to 100 or 200k of revenue. Speedrun for me felt like a lot more mature for some reason. Some of the companies were, I mean, they, you know, they’re all coming out at like 30 caps or whatever in both programs now. But I don’t know, like some of the speed run companies felt a lot more further along, almost like almost into priced round or series a territory.
Karan Grover (33:47.441)
Mmm. Yeah. And I wonder how much of that comes from A16Z being like such a prominent multi-stage, where I wonder if some of that kind of trickled into the founder mentality.
Brian Bell (34:02.155)
Yeah, because YC is not as much of a multi-stage. think they do follow on investments. I mean, A60Z is going to, they can be with you all the way to IPO, basically.
Karan Grover (34:10.885)
Yeah, yeah, no, it’s, it was a fantastic experience. I think my first hire was through A16Z partners. Our first customer was through an A16Z introduction. And then we pivoted within the batch. we kind of, maybe a progression is a better word than a pivot. And that came with a lot of guidance from our mentor on just reaffirming like, hey, trust your gut, trust what the customer told you and move in that direction confidently.
Brian Bell (34:38.303)
Yeah. Back to AI and engineering, where do you think AI engineering tools will accelerate first? So you have Greenfield apps, legacy systems, internal tools, or other things.
Karan Grover (34:50.235)
Yeah, I think to be honest, a lot of companies take on Greenfield because it’s a really easy problem. Like we had Greenfield solved to pretty high caliber very quickly, but the difficulty comes in legacy systems and existing systems. Exactly. How do you take something with a thousand cloud resources that has been around for 15 years and like 200,000 lines of code and understand that without summarization and knowledge graphs? Yeah, exactly. Yeah.
Brian Bell (35:02.967)
holding them forward, yeah.
Brian Bell (35:13.633)
I think you just gave a solution architect like a heart attack when you said that.
Karan Grover (35:19.769)
And that’s actually such a good point. Like, how do you do that and then convey to the solutions architect or the engineer how this has been done without throwing, you know, 10,000 line PR in their face and saying, hey, go figure it out.
Brian Bell (35:30.957)
Right. So kind of looking forward a little bit, how do you picture software development changing over the next three years? I mean, it’s already changed so much in the last three years.
Karan Grover (35:42.821)
Yeah, I think that with Canu, you’re going to see this elevation just how I felt in my early days of engineering, where you’re becoming the staff engineer of the team. Maybe we’ll see less junior engineers, less mid-level engineers, and more of these pizza teams, like two pizza teams, that’s the Amazon saying. You know, have eight engineers. Now maybe you’ll see a product and then one engineer orchestrating with an agent like Canu.
Brian Bell (36:09.089)
Yeah, it’s like one PM, one designer, one engineer, and they’re all orchestrating their agents.
Karan Grover (36:15.003)
Totally. I love how you said that because it’s almost like, you we were talking about DevOps and QA and engineering morphing. Now maybe we’ll see product and engineering and all of that turned into just one role.
Brian Bell (36:27.405)
Yeah, I was talking to another founder about that, tries to take your designs into code, bridging the gap between Figma and fully developed code. And he’s seeing on the ground that the designer and the PM are merging with maybe the front-end engineer. And it’s all becoming one person almost that orchestrates AI. And everybody can be 5 or 10x more efficient in all these roles.
Karan Grover (36:33.265)
Hmm. So good.
Karan Grover (36:45.969)
Totally. And like if I was graduating right now in CS, I think that’s exactly what I would do. I’d go through and one, learn all of these tools. For example, the AI for design, the AI for QA, Canu, learn how to orchestrate these very effectively. And then two, you have the entire corpus of the internet at your fingers asking very deep questions. How do you do best practices in DevOps and design? and just really elevating your role and treating yourself like a staff engineer today. And you’ll be hard pressed to find a founder that wouldn’t want to hire you.
Brian Bell (37:45.707)
Yeah. Can you hold on a sec?
Brian Bell (38:46.253)
Alright, back.
Karan Grover (38:50.033)
Hey, Brian. No worries.
Brian Bell (38:50.881)
Yep, sorry about that. what do you think founders and companies, Fortune 50 for that matter, often misunderstand about deploying agents in production environments like you guys are doing?
Karan Grover (39:02.309)
Yeah, I think you have to choose the right agent because you have to make sure the incentives are aligned. With traditional applications, what you’ll find is, okay, sure, it works if you prompt it really well or you learn how to orchestrate it or all of these difficult things. And then you’ll throw it into maybe a startup and it works fantastic. But if you throw it into an enterprise, we know this, you have 5,000 engineers, they’re not all on the same page. we need to make sure that this system is well adopted. Well, what are you going to do? Are you going to go do like decades of training to help everyone get elevated? And then by then the agent is obsolete. So it just doesn’t work. So you need your system to have incentives aligned with what you want to achieve as an organization. And that’s why Canu, you know, some engineers, when they first try it, they’re, frustrated. They’re like, it’s forcing me to define the product really well, or it’s forcing me to define what I want. It won’t just let me hit start. And the reason is, is because we know that that product portion of the problem solving is actually really important. Like you said, the PRFAQ, you need to understand what you’re building and the agent more importantly needs to know what you’re building before it can go off and do it. So knows how to succeed. And that’s, that’s why we built that.
Brian Bell (40:12.725)
No, PR FAQ for anybody listening doesn’t know what we’re talking about is a press release. Frequently asked question, it’s a six-pager document that Amazon uses internally to review project proposals, basically. So starts with the press release, and then you have like five pages of FAQs. And really, you have like 20 pages of FAQs, and then you sort of concatenate them, condense them down to five pages or so.
Karan Grover (40:38.619)
And it’s a forcing function. You have to understand what you’re building. You’re not gonna get engineering resources if you don’t know what you’re building. And so that’s what Condom says, is okay, you wanna build a chat bot. Great, what’s the goals? Low latency, customers, internal, external? And it’s just gonna push you and push you and ask you these questions because then when it gets to the QA stage after it’s done the code gen and the DevOps, it’s gonna say, okay, this person told me latency is a concern. I’m going to ping this API until I can get it under 50 milliseconds latency.
Brian Bell (41:08.135)
That’s amazing. Well, let’s look out long term. do you see? What’s the long term vision for what you guys are doing? And where do you think things end up five or 10 years from now?
Karan Grover (41:18.095)
Yeah, I think we’re going to be the lead in that change to elevating the engineering role across the world. And we’re starting to see it already. We’re seeing, like I said, the three month project in two days. Now they’ll be able to deliver so much more on their roadmap, that team. And so we’ll start to see that happening. And then two, I think we’ll be the driver behind really, really amazing problems being solved. Because if you can get out of the weeds of all these menial tasks you don’t want to do, imagine the problems you could think of, the problems you could solve.
Brian Bell (41:46.263)
Yeah, it was funny. was kind of giggling to myself when you said the, you know, the one quarter project in two days because, you know, 10 years ago when I was still a PM, you know, our quarterly roadmap or, you know, our four quarter roadmap really got like, we really got Q1 done in four quarters. So it take us like all year. We could just get the stuff that we thought we’d do in one quarter done.
Karan Grover (42:11.525)
Yeah. Totally, because Q1.
Brian Bell (42:15.051)
And now the inverse is almost true. Now we’re like, well, we just got like our whole year’s worth of stuff done in a quarter or less. like, so the velocity is just so much faster.
Karan Grover (42:26.129)
Yeah, I mean, like enterprises, you spend Q1 talking about what you’re gonna do this year and Q4 talking about what you’re gonna do next year. So if you’re lucky, you have six months in their actual work. But yeah, I like imagine what you could do with all that extra time. mean, our team, when we use Condu, we don’t do like menial features or like random tasks that we don’t want to do. We’re thinking about the big problems. How do we change the world?
Brian Bell (42:48.631)
Yeah. Well, let’s wrap up with some wrap up questions. I used to call them rapid fire, these tend to be kind of medium length questions. Which skill from your Amazon ML Day has proven most valuable at Canoe?
Karan Grover (43:05.305)
I think really learning how the LLMs work very deeply and there’s some really great videos on YouTube as well to kind of help understand that. But that helps inform why the LLM behaves the way it behaves sometimes.
Brian Bell (43:18.305)
like the three blue and brown videos. Those are really good, yeah.
Karan Grover (43:20.613)
Yeah. Or like Andres, like he’s amazing. Yeah, he has like a four hour video. He’s a UBC alumni as well. fact.
Brian Bell (43:30.039)
Yeah, yeah, that’s right. What’s one belief about AI agents you’ve changed your mind about in the past 12 months?
Karan Grover (43:37.542)
Hmm. I think I used to think back in the day that agents couldn’t do these long polling tasks. And the more I worked in AI, the more I worked in ML. I agree that, sure, we’re not going to see this overnight, you know, I robot type of history, but we are going to see a lot of the tasks that we don’t want to do being replaced with agents. And that’s a good thing. Like engineers aren’t going anywhere. Engineers are here to stay, but our job is getting elevated.
Brian Bell (44:07.917)
Yeah, I like that. Well, yeah, I mean, you were describing it. So junior engineers are becoming senior engineers and senior engineers are becoming staff and staff are becoming like architects, basically, you know.
Karan Grover (44:19.717)
You know, that made me think of my very first manager at Amazon. told me, I told him I’m going to be an SD2. My goal is to get to SD2 in record time. I don’t want anyone to have got there as fast as me. And he’s like, if you want to do that, then you better be thinking about how to get to principal engineer in record time, because you got to shoot for the moon, right? And then, yeah, or shoot for the stars, land on the moon, whatever the saying is. Yeah.
Brian Bell (44:40.81)
Yeah, I love that. When you think about dev teams today, what is the single most underrated engineering bottleneck?
Karan Grover (44:49.339)
think it is going to be the enterprise process. It’s talking about next year, debriefing next year or last year. Those processes really slow us down. We need to get back to the core of engineering.
Brian Bell (45:03.041)
Which founder habit from your first startup would you never repeat?
Karan Grover (45:08.043)
It’s the dependencies on the external companies. think that’s the that’s one I would never do. And then the second is I would make sure always have radically true communication. think bottling feelings up and how you feel about a situation or a person is just such a bad thing to do as a founder. You can learn how to communicate better and more empathetically, but you have to tell the truth because otherwise you’re just going to blow up one day and and let it all out and then your company’s over.
Brian Bell (45:39.713)
What’s the most surprising thing users have done with Canoo that you didn’t expect? I guess we had that three month, two day thing, but anything else?
Karan Grover (45:47.921)
Three months and two days, yeah. I’ve seen some really cool projects. Like I’ve seen things that used to take us years to build. Like I remember building Chatbots before ChatGPT. That felt like the hardest task in the world. And really, yeah.
Brian Bell (46:00.161)
I did that back in before I joined Amazon in 2018. I was at a company called Converse, they’re still around. But we when I joined, they’re an AI company. This is twenty seventeen, but they had all regular expressions under the hood. That was it. So it’s basically like text recognition, bag of words, stuff, you know. So we did actually revamp it all with intensive entities and, you know, some rudimentary sentence construction and stuff like that. But it was pretty pretty archaic compared to what we have today.
Karan Grover (46:33.509)
Yeah, it’s so... Yeah, like that’s what I was gonna say. You think back and you’re like, back in the days, but it’s like, no, back in the days was like maybe four years ago. Yeah, yeah, yeah, literally, yeah.
Brian Bell (46:46.305)
Yeah, I’m back in the day before LLMs. Yeah. Yeah, we almost have to like qualify everything like before LLMs and after now. It’s kind of like before cloud after cloud. It’s that big of a shift, you know?
Karan Grover (47:00.091)
Yeah, it feels like that the same, I know people like to say like the internet, like I definitely see the parallels there, it’s the invention of the internet, but I really like the cloud shift because the cloud shift was a huge paradigm shift.
Brian Bell (47:11.125)
Yeah, that was a huge I would say my theory is and I think other people have said this is that the cloud shift actually allowed for angel investing because I mean there were angels before that but that’s really when you could be a really early investor because you just have two or three founders with a laptop and the cloud just building stuff and delivering it. Before then you really had to spend millions of dollars to set up your infra just to get your like website out there. to open business. you’ve like literally your seed round was like millions of dollars just so could like buy servers and like get your infrastructure up.
Karan Grover (47:37.819)
Totally. Yeah. Totally.
Brian Bell (47:40.055)
to open business. you’ve like literally your seed round was like millions of dollars just so could like buy servers and like get your infrastructure up.
Karan Grover (47:47.685)
Yeah. And like the faster we’ve learned to build as engineers, the more the clouds have benefited over time. You know, like we’re all spinning up a hundred AI projects a week now and the clouds benefit AWS Azure GCP. But that’s also why they love Canoo so much because now we’re saying, Hey, even with cursor, if you can only do it in three months, now you can do it in two days. Well, the clouds would love that because that means more infrastructure running on there. Yeah.
Brian Bell (48:14.282)
More cloud revenue, more meters are spinning. So do you like spin up? Will a customer just at the end of the day spin up Canoo? How long can Canoo run on its own? Just run and do stuff?
Karan Grover (48:26.417)
I’ve seen some crazy amounts to be honest. We haven’t found the limit yet. It will run until it solves the problem. There’s a way to stop it and put limits on it, but I’ve seen three days, six hours for, I was trying to solve a high frequency trading problem and trying to get very ultra low latency. And so that involves so many convoluted resources being used from AWS and so much unique app code. But yeah, it did it.
Brian Bell (48:38.689)
Wow.
Brian Bell (48:51.629)
So you had some sort of quant trading fund using Kano to optimize their latency for trading.
Karan Grover (48:57.349)
So that one wasn’t an actual real customer. That was me seeing if I could solve a quant problem with Connor. But because quants get paid a lot of money and they spend a lot of time solving for this one problem. So was like, what if Connor just spent 72 hours nonstop trying it themselves? So I took off all the cost limits, time restrictions and everything and it just went at it and yeah, it solved the problem. I can’t remember what the latency was, but it was extremely low. Yeah.
Brian Bell (49:00.641)
Just messing around.
Brian Bell (49:07.277)
Great.
Brian Bell (49:23.969)
So if you had to mentor a speed run or YC company building an AI infra today as we enter 2026, what would be the first question you’d ask them?
Karan Grover (49:39.941)
I’d say what was the unique customer insight that you found? Because I think I’d want to know, one, they spent time with their customer, and then two, what did they learn from their time with their customer? Because that’s what this is all about, right? We’ve kind of leveled the playing field in how quickly we can engineer. Now it’s how well do you understand what you’re building?
Brian Bell (50:02.315)
Yeah. What’s a recent research paper that has influenced your thinking the most?
Karan Grover (50:08.399)
Yeah, this is kind of me not gatekeeping, which I consider doing it, but the ace paper from Stanford and Berkeley, think is.
Brian Bell (50:17.803)
I heard of that. Maybe you could explain what it is.
Karan Grover (50:20.761)
It’s this idea that instead of fine tuning, your system teaches itself how to do things differently next time. And so that paper, the results that they achieved without fine tuning was groundbreaking. And so when we read through that paper and then we reread it and then we did a whole team read and we realized, okay, we’re not going to do post-training. There’s other techniques that we’re going to try first before we go in that direction.
Brian Bell (50:46.593)
Yeah, I think that’s the natural extension. I mean, you had pre-training and then you had test time compute. And now it’s possible to kind of like let the AI kind of analyze itself and kind of edit its own weights and fine tune itself based on what’s working, what’s not. And then you could kind of do that self referential loop. And it’s kind of how we learn as humans, right? Like, I tried that thing. It didn’t work. What is it about that thing that I tried that didn’t work? What would I do differently next time? OK, let’s try that. OK, let’s analyze the results. that’s how we learn.
Karan Grover (51:19.761)
We have to take those cues from humans and even like software development lifecycle. That’s so human nature for us engineers. It’s just a process we follow to solve problems. And so we have to kind of build that in ourselves and post-training happens in our brain and we do fine tune over time. It’s not that we shouldn’t do that, but it’s just not the first approach that we took.
Brian Bell (51:42.669)
What’s one thing that Big Tech does extremely well that early stage startups should copy?
Karan Grover (51:48.219)
think it’s those systems that you and I were talking about earlier about Amazon. Those systems, the ability to step out of the room and just know that the things will still happen the way that you intended. think companies like Amazon, Google, Microsoft, they really nailed those down. Like BarRazors, for example.
Brian Bell (52:04.365)
So who’s your ideal customers? You’ve been talking to Fortune 50, Fortune 500 customers. So anyone listening, who are you looking for and who should get in touch? And how do they get in touch?
Karan Grover (52:18.415)
Yeah. So we’re working with enterprise customers today. Generally, most of the teams that we’re speaking to, they have about 10 to 20 engineers plus, and we’ve worked with companies as many as 5,000 engineers. And so that’s kind of like our ideal persona. Now, that being said, exclusive to this podcast, if you shoot me an email and say you were listening, we have a beta version coming out of a self-service. So this is for your startups or companies that or listening that want to accelerate their engineering, you can just shoot me an email. It’s karan, K-A-R-A-N, at getkanu.com, getkanu.com. Or you can just go to our website and book a demo call there. And then, yeah, we’ll get you on the self-service and you can try this out.
Brian Bell (53:02.736)
That’s amazing. Well, I learned a ton. Thanks for coming on, and congrats on all the success so far. I can’t wait to see what you guys accomplish over the coming years.
Karan Grover (53:10.981)
Yeah, thank you, Brian. Thanks for having us.
