Nine Months at a Gen-AI Startup
"The most valuable businesses of the coming decades will be built by entrepreneurs who seek to empower people rather than try to make them obsolete." - Peter Thiel, Zero to One (2014)
One of my three 2024 goals was to post here on 30 Tuesdays. I came very close, clocking in at 4 new posts. I took a writing vacation since March 2024, and by “vacation” I mean I joined a startup that has consumed most of my waking hours, with a small carveout for listening to an NFL podcast at the gym. I will be writing more in 2025.
Zero to One
Possibly the best pound-for-pound nonfiction book that I’ve read is Zero to One by Peter Thiel, co-founder of PayPal and Palantir. Thiel loves contrarianism, but not in the Kyrie Irving vein. Thiel’s ethos can be pulled from his book: “The most contrarian thing of all is not to oppose the crowds, but to think for yourself.”
(Incidentally, Thiel is on the Mount Rushmore of gay Republicans. I know you’re thinking, “James Buchanan,” but Buchanan was a Federalist turned Democrat, so not eligible!)
My five-sentence summary of Zero to One:
Human progress depends on technology, and technology is the creation of something that didn’t exist before.
For 10,000+ years, there was basically no technology being created, and then from the 1700s to 1970s there was mind-blowing tech creation (steam engine to phone to car to airplane to computer).
Since the 70s, most new tech has been in computing, so much so that we think of “technology” and “software” as synonymous at this point.
Terms like the “developed world” and “developing world” are a reminder that we have largely given up on dreaming big about building new technology, seeing the “developed world” as the end state—the physical world around us still looks like it did in 1965, with no flying cars to be found.
Startups are fundamental to new technology, and there is existential urgency for new technology (climate, medical, etc.) to emerge from startups.
My time at Codeium, a startup doing Generative AI for software development, has reinforced the ideas in Zero to One via firsthand experience. Today’s post is not about Generative AI (like ChatGPT, which we’ve all seen by now), though I’ll write about the AI industry in the near future. Instead, I want to focus on the “startup” part of “Gen-AI startup.”
Golden Handcuffs and the non-Michael Scotts
I entered the tech industry with zero formal qualifications in March 2020 (a process chronicled here), beginning at a post-IPO database company. My plan was always to join a startup, but two forces kept me there for four years:
“Golden Handcuffs”: big and big-ish tech companies are famous for giving perks and compensating employees juuuuust well enough to stick around and fatten the pockets of the Powers that Be. (Important tangent: In the Michael Scott theory of social class, these Powers that Be are referred to as the “sociopaths”—e.g. Henry Ford, Steve Jobs, Elizabeth Holmes of Theranos if her tech wasn’t fraudulent, and David Wallace on The Office. “Sociopath” is not derogatory here, and though it’s popular these days to hate billionaires, tech founders are largely responsible for the massive consumer surplus we all enjoy—e.g. iPhones, Google, and Prime delivery. Without their ambition, there’d be little breakthrough technology.)
Most SaaS (software as a service) startups sounded the same, i.e. uninteresting. Why would I leave a company where I know the product very well—an advantage in tech sales—to join retread SaaS company number 788? Most business-to-business (B2B) SaaS startups did not feel exciting; efficiencies in software felt milked to the point of saturation.
Enter Codeium
My nihilistic feeling that all tech startups were boring began to change in 2022, when Generative-AI apps like Dall•E and Midjourney started to blow people’s minds by creating images from human prompts. Then, in November 2022, OpenAI took the world by storm with its release of ChatGPT.
On the one hand, this AI is still just “software,” but on the other hand, it clearly has the chance to help us reshape so many needed fields—genomics, education, and generally all knowledge work. Despite these promises, I still wasn’t eager to join a random AI startup, as most products were just “wrappers” that used ChatGPT under the hood.
In early 2024, a tech friend put me in touch with Codeium’s CEO, Varun Mohan. Varun is a force of nature, a statement supported by top VCs and most people who have interacted with him. I walked away from our in-person “coffee”—Varun did not order anything, so I sipped on my latte like a loser while gripping my notepad full of questions—with genuinely no idea if it went well, but something about his intensity was undeniable, exciting, and literally impossible to fake without a felonious amount of amphetamines on board.
Within the next week, I quit my job of four years before my Codeium interview even happened. I fully recognized that like 99.999% of startups, Codeium would likely not become the next Google even if it survived. I also realized that there was almost no downside to going all in for the interview: the new gig would have health insurance and a salary (founders are the only people taking risks, in my opinion), and if I didn’t get hired, I could take a mini-vacation and go find another job (and if I couldn’t navigate a job search, I’d have a LOT of explaining to do).
At a venture-backed startup, the upside of being employee 28 versus employee 100 is significant for equity, responsibility, and relationship reasons, which was another reason I didn’t want to wait.
Why Codeium?
In between quitting my previous job and interviewing at Codeium (and starting that same day), I emailed the initial three Austin employees to be explicit about the reasons I was most excited.
Codeium Leadership: Though I was new to startups, I’d done enough observing and reading in my life to believe that the quality of a company’s founders is the strongest predictor of future success. I’ve written before about Steve Jobs’ unusual ability to distort reality in Apple’s favor. If there was going to be another Steve Jobs, Varun is on my list of potential successors. Frighteningly, Varun’s co-founder, Douglas (Codeium’s current CTO), was seen as even more of a math outlier amongst the math outliers at the top of their MIT class.
People: Immediately after starting as employee 28, I discovered that by any objective measure of intelligence, I was dumber than between 26–27 of my coworkers (trivia knowledge and jokes-per-blog excepted). At the time, roughly all of Codeium’s youthful engineers came from MIT, Stanford, Cal Tech, and Harvard. I, proud Centre College alum, am in sales, as you’ve inferred by now. On the sales side, I also saw a great opportunity to learn from a team I respected / had scaled startups like Grafana to nine-figures of revenue.
Product: Codeium’s early team came from self-driving vehicle startups, so they have firsthand experience of the pain of lofty promises that come too early—e.g. “full self-driving will arrive by 2020.” While there, they also gained expertise in low-level GPU infrastructure, which is critical for AI apps and is something that only a handful of teams in the world can say.
Stemming from lessons in the self-driving world, Codeium’s philosophy is not to wait for future breakthroughs to occur, but instead to add value to software developers now while also continuously building for the future. Notable competitors focus on “AI software engineers”—i.e., replacing engineers—which is similar to companies in 2013 who promised the imminent arrival of fully-autonomous cars (in the interim, Tesla gathered a boatload of driver data while the others died).
Impact on the Future: Investors are calling Gen-AI the “mother of all markets,” and software development alone is a trillion-dollar market that Gen-AI will fundamentally reshape. Giving laypeople the ability to create apps and websites using nothing but English is an early domino that’s falling now. Although we’re still in the first inning, I’ve pondered the future of human labor for years on the blog, and I knew I wanted to contribute to a future of more meaningful jobs.
The most valuable businesses of the coming decades will be built by entrepreneurs who seek to empower people rather than try to make them obsolete.
- Peter Thiel, Zero to One (2014)
Scurvy
If I’m being completely truthful, my decision to pursue Codeium was not as analytical as I make it seem in hindsight. I wanted to be at a place—i.e., a startup—where my specific skills and creativity were more impactful than they were at a company of 5,000 people, and when one came along that felt right, it was time to move.
Emotional impulsivity can be harmful in many areas of life, but being an emotions-led decision maker arguably helps in one’s career. People’s biggest career-based regrets are rarely the results of the actions they take, but are instead regrets about years or decades of inertia-fueled inaction. It’s easy to tolerate the mediocre when one is not in pain.
In other words, Codeium (like every startup) could stop growing and die, but the growth of each individual member is not killable. That’s part of the thrill of joining the modern-day equivalent of an Atlantic gold-expedition voyage. Yes, you’ll get scurvy along the way and consider walking the plank thrice weekly, but at least you can use the word “AI” in blog posts.