3 Months After Quitting My Corporate Job to Build With AI
Summary
- My plan was a website and two apps in 90 days of building with AI; three months in, only the website has shipped.
- The time went to four projects at once and then a month of over-engineering my own agent loop, until I changed course and started writing.
- So ship one real thing before you build a system around it, and if you are leaving a job to do it, build the runway first.
On July 3, I left a job and a team I loved after almost nine years at Poshmark to build with AI full time.
Three months in, I can tell you two things. The plan I left with did not survive the first month. And of the three things I meant to ship, one has.
I am not an engineer. This is what the leap actually looked like: what went wrong, what finally worked, and what made it possible at all.
Why I quit my corporate job to build with AI
I caught the AI bug at the start of this year. It began with an internal group at work called the tiger team, where one of my esteemed colleagues showed us how much these tools could do for anyone willing to build with them.
Within a week, I was building an analytics Slack bot for the company on my own. Nobody asked for it. I figured it would help the company enormously to get the right insights, as fast as possible, to everyone who needed them.
Within a few weeks, a few people were using my prototype. A few iterations later, more people could use it, and we started putting it into production. It became useful to the product, growth, and executive teams, and its insights started showing up in weekly and monthly business reviews.
That was a lot of serotonin. It was also proof that AI plus my expertise and vision could build something people actually used.
After that, I could barely go an hour without thinking about what else I could build. Every off-work hour went into learning and building more. The more I did, the more I wanted to do.
I did not leave because I stopped liking analytics. I left because there was so much more I could do with this technology and my ideas.
I knew I could delegate, and if I could set AI up to reliably do things the way I wanted, I would be on a compounding ladder: make it personal, and have it build, market, and monetize. So I left.
The financial cushion that made leaving possible
Ambition, and financial safety. Without the second one, leaving would have been almost impossible in this economy and job market.
The safety came from the financial foundation I built over the past ten years. I have an investing process and rhythm I trust, and a systematic withdrawal plan that gives me a long runway to try things. Before I gave notice, I also set aside a year of expenses in cash, so nothing I build has to make money on a deadline.
That runway is what gave me room to try things when the next three months got messy. Work had become optional, and this felt like what the option was for. This site is where I write down both halves: the money decisions that made leaving possible, and what I am learning as I build.
My 30-60-90 day plan (and how wrong it was)
I was not under any delusion that this would be smooth. It is not, and it was never going to be. But I did expect a lot from myself in the first three months:
- 30 daysBuild the website
- 60 daysBuild an iOS app
- 90 daysBuild an Android app
Along the way, I would automate my workflows and run a local LLM setup on my new Mac. The system would do its job while I delegated every idea under the sun, reviewed, and shipped. Then I would expand exponentially.
In my mind, that was doable. It was not even close.
Month 1: a second brain and AI agents for four projects
I set up everything I had learned on my own machine. I built a second brain. I built AI agents and named them after experts I had worked with, so they were easy to remember. Then I ran four projects at once.
The first versions turned out great. They did exactly what I expected, and I was over the moon.
By week three, the cracks showed. I was spending more time fixing things than building new ones. Lessons were not being recorded, followed, and applied consistently, at least not every time. That does not scale for a solo founder.
Month 2: building my own AI agent loop, and over-engineering it
So I went into research mode: loops, workflows, harnesses, and all that jazz. I spent a ridiculous amount of time using Claude to build my own agent loop, one that could run on local LLMs instead of the one Claude offers.
It worked for one task, then a few. Then it stopped scaling too. So I added a harness around each step, then checks and balances around the harnesses. I was building scaffolding around assumptions instead of around real work, and there was no end to it.
By the end of month two, I was nowhere near my 30-60-90 day goals.
What I got wrong
Month 3: ship first, then optimize
Month three, I changed approach. A friend and former teammate suggested I start writing, so I started this website, writing about personal finance and the lessons that could help others. It gave me a north star.
I am still learning how the backend and the systems work. AI does a lot of the work, but every time something new goes in, I learn something. That taught me I do not have to get it perfect the first time.
Ship first, then optimize. Month two taught me that the hard way.
So month three went into writing and building an audience here. It is still very early. The fourth quarter is dedicated to reaching 1,000 subscribers and learning a lot about marketing. In parallel, I am still building reliable workflows on local LLMs, with frontier models where they make them more reliable. More on that in another post.
What life after quitting a corporate job feels like
Net net, it has been a great journey: lessons, falls, and getting back up.
I started working from coffee shops. I have met other ambitious people who want to start their own entrepreneurial journeys.
Two people set these three months in motion: the colleague whose tiger-team demo started all of this, and the former teammate who told me to start writing.
It is also a lonely space. Being an introvert, I kind of dig it. But I can see why it is not for everyone.
What I would tell anyone building with AI, at work or on their own
- Expect the plan to slip. Mine was a website and two apps in 90 days. I shipped one of the three, and only after I changed approach.
- Watch your fixing-to-building ratio. When fixing what AI built takes more time than building new things, the setup will not scale, whether it is a side project or a project at work. That was my week-three signal.
- Ship before you build the scaffolding. I lost a month to harnesses built around assumptions, not real work.
- If you are leaving, build the runway first. A cash cushion and a systematic withdrawal plan gave me a long runway to try things, even through a messy first quarter.
- Find your people. Building alone is lonely. Coffee shops, weekly catch-ups, and other people starting out help.
To my team at Poshmark: thank you for nine great years. Much of what made this leap possible, I learned working with you. Lots of Posh Love <3
That is my three-month journey into making work optional. I will share the next update at the six-month mark. If you want it when it lands, subscribe to the newsletter.