Back in February 2024, I wrote a private retro about my time at scale.ai. For context, I was one of the GMs and led the Generative AI businesses during the explosive growth of LLMs.
Recently, I shared some of these learnings on LinkedIn, and several people asked to read the full retro. This is a more detailed (and still redacted) version. If you’re curious about what it’s really like to leap from big tech to a turbocharged startup during the Gen AI wave, I hope you find this snapshot helpful.
Why I chose to join Scale
In 2021, I was searching for something that would truly stretch me—I wanted to build something meaningful from scratch and step far outside the familiar. Scale felt like the most “possibility-rich” place I could imagine: nothing was pre-ordained, just a blank slate filled with both risk and potential, surrounded by people who were relentlessly smart and hardworking.
When I asked what I’d actually be doing, Scale’s answer was, “We have a lot of new business ideas—why don’t you just join and figure things out?” By contrast, most other companies were telling me “You’d have a team of X to work on Y, and we are tripling the team this year to Z!“. I honestly didn’t know what I was getting into, but I did know that embracing uncertainty was exactly what I needed after 8 very predictable years at Google.
Joining and ambiguity (early 2022)
I started at Scale when the autonomous driving wave was cresting. My official title was “Product Incubation,” but once I joined, everyone simply called me the “Catch-All” person. I was paired with three engineers, working on anything that didn’t fall into existing business lines or teams—mostly serving Content & Language customers in areas like search relevance, social media, and other general language use cases (hint: you can guess what this really is).
The team moved fast to support existing customers. I knew I had to win trust by getting my hands dirty, so I focused on fixing bugs and supporting the team wherever I could. In those months, it seemed like nobody really cared what I did—I often felt like I was taking on the leftovers no one else wanted, which made me question why I was even here. “Oh no, what have I done?” I thought to myself, and my goal seemed nowhere on the horizon because everyone was too busy to mind me. But I decided it was too soon to give up. Maybe the hard, uncomfortable parts were just part of the test.
Breakthroughs through action (Spring–Summer 2022)
A breakthrough came when someone in sales needed help on a customer call: “I have a customer meeting lined up and this doesn’t seem to fall into existing lines.” I responded on Slack, “I’ll do it with you!” I had never gone on customer calls before, and couldn’t sleep at all the night before. But that single “yes” cracked open direct access to customers—and soon I was pitching all kinds of new ideas and learning on my feet.
Not long after, a SEV0 hit. The team found a quick fix, but the same bug happened again (and the team fixed again). The team wanted to move on but I pushed (against some skepticism) for a backend validation. We shipped—and the issue never returned. That moment reassured me that even though I lacked codebase knowledge and couldn’t code as fast as my team, my instincts could make a difference.
In my first month, I also had to let someone go. I’d barely ramped up when I was asked to take the conversation. Another sleepless night. It ended better than I feared—our relationship stayed strong, and we’re still in touch. That whole stretch was a lesson in being both decisive and kind, and in how leadership here, up close, is a uniquely uncomfortable experience.
Allies, setbacks, and new teams (Spring–Late 2022)
As I got pulled into more customer calls, I started pitching new use cases that fit customer needs—one of them drew on my Search background, and we signed a 7-figure contract (yay!).
Over the next few months, more deals followed, and I established a new business unit around subjectivity. For computer vision there's a ground truth, but for much of content and language work, it’s extremely subjective: is there a trend in this data? Is this a good chat response? One major project emerged, quickly becoming the biggest opportunity for the company. Three work-streams, engineers reallocated from multiple teams, and a successful POC that led to a paid pilot. Our approach of supporting all three projects as a single team (rather than one engineer per project, which was the default model)—proved highly effective (the project was intensive but manageable the entire time, nothing was greatly on fire).
Eventually, I became GM for a business unit. Business evolved, teams expanded, and new faces joined.
Meanwhile, I was attending plenty of customer events and public intros (Transform X, AI dinners—the kind of things I never imagined doing), often against my own instincts. December’s AI dinner, with excitement swirling about ChatGPT (launched less than a month earlier), left me thinking: this is Silicon Valley at its most electrifying.
Layoffs and reset (January 2023)
In January 2023, a 20% company-wide layoff hit—40% of my team was gone overnight (despite us hitting our revenue target). It was a gut punch, and I found myself starting at the ceiling every night for almost another week, though it reflected a broader trend in tech. At the same time, Generative AI began to explode; suddenly, everyone wanted the projects we’d been building around subjective content and language—demand was rolling in non-stop.
From April onward, the business took off, and much of the company began focusing on GenAI. I was tasked with figuring out how we could sustain this crazy growth across the board—a deeply exciting challenge. It was a fun, intense time: sketching architecture with our CTO over weekends, breaking the product into new scopes, and helping transform the company. The speed was dizzying. My ideas found traction, and the team’s response made me realize we were really investing in product, systems, and infrastructure for the long term.
None of us could have predicted just how sustained and explosive GenAI’s growth would be. Facing this demand, the only solution was to “toss bodies” at the problem and extremely long hours month after months—hiring aggressively and growing the team from under 20 to more than 80 people by year’s end.
Gen AI explosive growth (2024+)
Continuing into 2024, as Gen AI took off, we found ourselves executing on much of the original roadmap—despite all the chaos along the way. When growth is explosive, things are supposed to feel out of control most of the time. But even with constant delivery pressure, the core needs stayed clear: build sustainable infrastructure and core components, keep processes simple, invest in internal tooling, nurture the community, and create repeatable best practices across projects.
At the same time, having a front-row seat to LLMs made me want to get back to hands-on building. I knew I’d never get the same kind of learning through second-order observation as I would by building myself. The “what if” question came back: what if I set out on my own, with no platform underneath me? There are more unknowns out there, and that uncertainty deeply excites me.
Summary on learnings
Coming out of the chaos and intensity from the past two years, here’s what stands out:
I learned more in two years at Scale than in eight years at Google.
Built a business unit from scratch, reached $XXM ARR, and assembled a high-performing team.
Developed complex roadmaps, secured buy-in, and shipped against real ambition.
Decided to make “say yes” my default—even when I felt unready. Facing imposter syndrome—whether giving public intros as a non-native speaker, handling early customer calls, or re-architecting teams—led to some of my biggest leaps.
Exposure to Ops helped round out my mindset and gave me new confidence.
My biggest gain is my people—feeling love and being loved by my team, and knowing that recognition and trust outside the team meant I was on the right path.
What didn’t go well
Some relationships or initiatives didn’t survive two years of whiplash; persistence wasn’t always enough.
I wish I’d been braver in what I said and let go of unresolved things—sometimes saying yes too readily led to burnout.
Habits from big tech sometimes helped, but also sometimes kept me from seeing scrappier ways.
The regrets that linger aren’t for missed projects but unspoken words or rifts left untended. Next time: be braver, softer, more present.
As I left in February 2024, ambiguity was no longer my enemy. I don’t know what’s coming, and that deeply excites me. For years, I thought startups were “for other people.” Now, I know I can:
Come up with (somewhat) good ideas
Build deep relationships and teams
Foster environments where people thrive
Face hard challenges head-on
Here was the reminder I wrote for myself at the time:
Embrace unknowns and risk.
Keep learning and absorbing at speed.
Cultivate internal motivation instead of chasing external validation, focus on what feels meaningful, not just impressive from the outside.
Push further in Gen AI and human-AI interaction.
Take care of my body and mind—stay in shape, keep running, keep climbing, stay ready for what’s next outside of work too.
Create environments where people I care about can thrive—because none of the best work or growth happens in isolation.
The best things I learned all started—and thrived—outside my comfort zone.
If you’re reading this in your own messy middle: don’t fear the ambiguity or hard parts. All my growth came from there.
Thank you for reading. If this echoes your journey, I hope you find both grit and hope here.
Loved reading it.