Fantastic Interview of Sourcegraph founder and CEO Quinn Slack by Gergely Orosz (@GergelyOrosz) on GenAI and developers
While traveling a couple of weeks ago, I listened to a fantastic interview of Sourcegraph founder and CEO Quinn Slack by Gergely Orosz on his fantastic Pragmatic Engineer podcast.
It was a fantastic interview, for a whole bunch of reasons —
- Their vision for the last 11 years has been to make it so that everyone codes, and how it has resulted in tools such as Code Search and now Cody
- Anyone interested in coding assistants will appreciate his thoughts on why GenAI is important, and how much the entire dev landscape could be changing as a result
- How Sourcegraph grew from 25 employees in 2019 to currently 200 employees, and some retrospectives on decisions he would've made differently during the pandemic years of zero interest rate policy
- His thoughts on how GenAI can enable developers to do so much more, and may even end up rewriting the entire tooling landscape around them, focused on not humans, but on the GenAI tooling that will consume them — thinking logging, issue trackers, deployments, automated testing
- How developers are going to be become even more valuable, especially as they are able to do more things, often entirely by themselves
- How SourceGraph shifted focus to GenAI that resulted in Cody, and how Job Fairs was a mechanism to quickly rebalance where internal talent were working —?they were able to quickly change the levels of product staffing, and he talks about how Job Fairs create another problem…
- He gives some mind-blowing data points that should give confidence to any software engineer that they are capable of being company CEOs.
In the interview, Gergely asked Quinn about what it was like as the world changed in 2020, including the pandemic and the AI world affecting devs post-ChatGPT.
Quinn: "I feel like we've been doing the same thing that we set out to do 11 years ago, and there's a lot of consistency, I think that's been really good for us. We've had this mission to make it so that everyone codes.? Our product, Code Search, has stayed the same. We've done stuff on top of that with Cody. It took us a while to get that initial traction, and Uber was one of our first customers."
It took them company five years for Uber to become their first major client.??
That happened during the pandemic period, where Zero Interest Rate Policy continued to fuel massive investment in software R&D in companies
Quinn: "...that pandemic frenzy was leading to massive Dev hiring, massive investment, and everything software related.? We had history, we had staying power. It was clear that we were committed and we'd been moving in this direction for a long time."
On the current AI boom period:
Quinn: "We're still learning a lot about how LLMs are going to change things and how we're going to have to adapt our business to take advantage of them. It's definitely an exciting time."
They currently have 180 people on the team, having raised over $200MM.
Gergely asks Quinn about his "would haves, could haves, should haves" about what he'd do differently during the pandemic-era, knowing what he knows now.? They experienced 30x growth in revenue, but eventually their paying customers had to scale back R&D spending because of the economy.
Quinn: "I think a lot of companies, us included, really wish we had done a lot more of that back in 2021. There were some new features and new products we came out with that I don’t think should have made the cut."
"That’s on me as CEO. I should not have allowed that to happen."
Quinn talked about how early-career employees might have seen this, and been led to believe that career progression was easier than it actually is — that "grinding" is part of the process of ensuring their initiatives earn their right to survive.
Hey, and they talk about how pivotal it was that they hired @steve_yegge out of retirement, becoming their head of engineering —?
Quinn: "I think he was looking for this kind of opportunity because he built a lot of the stuff around code search at Google. When we were starting Sourcegraph, we thought he was our nemesis, and he was our biggest enemy. He was going around at Emacs conferences and talking about the thing he was building."
(Ha!? You can see this 2012 lecture Yegge gave at Stanford on this work, then called Grok: https://www.youtube.com/watch&v=KTJs-0EInW8 )
Quinn: "He was just such an awesome fit in what he wants to do and what we needed... He just naturally had that... ‘We are building something. Let’s just iterate faster. Do everything faster. Let’s hold the standards incredibly high.’"
"That’s the kind of mentality that I think, you know, we and every company probably needed a lot more of."
This is the part of the interview I found so exciting — Gergely asks Quinn talks about how GenAI is going to change life for developers, suggesting that more ambitious things will be undertaken.? (Which I have personally benefited from!? These videos and these notes are being created and written by tools I've written, with an ever-increasing amount from coding assistants, including @SourcegraphCody.)
But Quinn's answer was super-cool and unexpected, suggesting how many other ways the entire dev tooling landscape will change.
Quinn: "I think because you're going to see the tooling landscape get turned over a lot now that AI is going to be a primary consumer of these tools. Think about the logging tool, the issue tracking tool, the deployment, you know, the security scanner, all these things."
"They were made for nice UIs and enterprise features for humans to consume. But, what's the AI version of Datadog when you don't need all those bells and whistles, but you just need something that is really fast that can you know pipe some of the performance and log data back to AI?"
Not just those tools, but think about what's required to generate the fast feedback you need when AI is doing the work, and can do thousands of code changes per minute:
Quinn: "[for humans], a really fast build is something in the order of 30 seconds or a minute, and most companies are nowhere even close to that.? But, if you have a really fast build... that could run in 100 milliseconds or 500 milliseconds, then you could have the AI iterate, try a thousand different code changes."
That's a big, mind-expanding idea!
It's pretty exciting to think how software engineering organizations can change for the better with these new generation of tools.
领英推荐
Quinn: "The value of a software engineer who knows how to use these tools has gone way up, so much so that they might not need a boss, or they might, instead of being very far from the value that they create in the product that they create, they might be able to be way closer to that which means they're going to be doing the job of a PM [product manager] and an EM [engineering manager], and a salesperson."
This is a topic that is riveting to me — it's not just the phenomenon of "novices optional" that Dr. Matt Beane has discussed, but it means that in some cases, good engineers will be able to do so much themselves, without the need for other engineers, or other functional specialties.
Their knowledge of the program domain, intimate knowledge of their customers whose problems they are trying to solve, will let a solo developer go so much further than what was the norm over the last decade.
Quinn: "And all that means [those software engineers] are going to make a lot more money in the future"
Gergely talks about the explosion in the amount of code that will be generated — "code is not an asset, it's a liability," which Quinn acknowledges.
With the advent of GenAI helping developers, they created Cody as a separate product —
Quinn: "We could have built it as a feature in Code Search.? We could have had the same people trying to build both, but we wanted to make a bigger shift because we knew that it was the future."
Enter Job Fair, as a way to re-allocate, re-align, re-prioritize efforts within the company:
"Job Fair was this idea that instead of well-defined teams that had long-term ownership over areas, basically every month or every other month everyone would just rearrange to work on the most important things that we would stack rank."
"It was good to kind of shock the system, to get people really quickly working on the new thing, but... it's not something that works forever and it doesn't create that kind of long-term ownership.? And as an all remote company, it doesn't create that kind of team camaraderie, so it's clear that it would expire pretty quickly as a way to work."
Quinn described how everyone was eager to work on the most important projects, but found that even with willingness to switch, it negatively impacted quality and was at times difficult to manage.
They used Job Fair for six months and then transitioned back to long-term teams. While short-lived, it helped them accelerate their move to AI.
"We had shifted enough people over to Cody that we went back to the much more normal way of just long-lived teams."
Gergely wrote more about Job Fair here: https://newsletter.pragmaticengineer.com/p/inside-sourcegraphs-engineering-culture
Quinn considers AI to be the most complex product they've built, requiring significant effort to ensure model accuracy and safety.
Quinn: "There are a lot of things that you learn when you're trying to grow a product, and I think AI is probably the hardest product we've ever built in a long time because it takes a lot of work to make sure that the model actually learns what you want it to learn, and that it's safe for everyone to use."
Quinn talks about the forces that lead one to need Job Fairs, which is fascinating.
Quinn: "Some of those projects that we started in 2021 that I ultimately see or regret allowing us to get started with, they would not make the cut today... It's not that they were terrible ideas; they were good, plausible ideas, but they were not the ones that we were lined as a company around, as being the very top priority projects."
"At the same time, we had people working on them that were working hard, that were satisfying customers with them. And it's hard to change that."
And thus the need to create shocks to the system.? But, on the other hand, he says, "It mostly worked. Going back, there'd be a lot of things that we could do to communicate it better, just be more upfront with this as a shift."
Many observed from the outside how quickly they shifted to Cody and AI —?
"[Despite knowing they could have done it better...], there's a lot of people on the outside who say, 'Wow, you shifted to AI so quickly. Can you tell us how to do it?" And VCs are having me talk about this.'"
Gergely notes how inspiration Quinn is to him, because he's overseen this fantastic success and expansion of the company as a software engineer — Sourcegraph grew from 25 employees in 2019 to nearly 200 employees currently.
Quinn: "I think software engineers naturally can do a lot of things. If you're coding, even if it's not a software product for developers, you're going to be able to talk to your customers. You're going to be able to talk to investors."
"People are very comfortable and familiar with the idea of a software engineer who's now a CEO, even in the Fortune 500. Software engineer CEOs are very much overrepresented relative to most other majors, like the CEO of United Airlines is a computer science major. And you can find so many examples like that. So it's not that hard of a transition."
Whoa.? I didn't know that.? So in the CEO of the Southwest Airlines!!
https://www.swamedia.com/about-southwest/our-leadership/robert-e-jordan-MCQTBMVFZJPZFW5ORGJQDUK7XYSA
And he highlights the importance of advisors and mentors, who helped him trust his own intuition:
"There are so many amazing advisors and mentors who will help, and for me, I think everyone's personality is different, but for me, the real growth was trusting my intuition... Because as you grow, a lot of people have really good advice. And also, as you grow, they see a CEO who was CEO of the company when it had zero revenue, and now when it has this much, and you know, it's especially as a company is quickly growing, you don't think the CEO is some, you know, expert genius that knows everything in every function."
Thank you for such a great interview, Slack Quinn and Gergely Orosz!
Back to basics, sustainability and first principle reasoning
1 周Well worth listening to. A company where the CEO codes ! A company that employs Steve Yegge and develops tools that I wish I had had when I coded and was challenged with large legacy code bases.
Professor of Practice, Clemson University School of Computing | Advisory Board Member | Technology Leadership Consultant
3 周On my list for a listen!
Board Member | GTM Hacker | AI/ML enthusiast
3 周Multiple worlds collide, Christina Forney - your favorite author!
Quinn is awesome. I interacted with him a fair bit back at Stanford, and he was a smart guy with lots of talents. Not surprised he's doing cool things today.
Sr. DevOps Manager/Architect at EXL Services
3 周great to know 200 employees now ?? ??