Two facts, placed next to each other, do something uncomfortable. A former Microsoft engineer walked away from a $200,000 salary to build an AI startup called Treasury, with no paying customers at the time of the jump. That same startup has since generated over $6,000 in revenue, according to Business Insider.
On a spreadsheet, that trade looks absurd. Six thousand dollars against a two-hundred-thousand-dollar salary is a rounding error. But the two numbers are not measuring the same thing, and confusing them is exactly the mistake that keeps most engineers employed forever.
Validation is a lagging signal
I spend most of my time thinking about how agents learn under uncertainty, and the structure of this decision is familiar enough to be almost textbook. An agent operating in an unknown environment faces a choice between exploitation, which means collecting reliable reward from a known action, and exploration, which means taking an action whose payoff distribution you cannot yet estimate.
The $200K job is pure exploitation. It is a known arm with low variance and a well-understood reward. It is also, critically, an arm that teaches you nothing new about the environment you actually want to operate in.
The thing people call “waiting for validation” is usually an attempt to explore without paying exploration costs. You want the reward signal before you commit the action. But in most real environments, including markets, the signal is generated by the action. You cannot query the oracle for free. Paying customers are not a precondition for building something people pay for; they are the output of having built it and put it in front of someone.
Sparse rewards and bad credit assignment
There is a second structural problem. Early-stage startup reward is sparse and delayed, and the gap between action and feedback is where most reasoning goes wrong. A developer ships for three months, hears nothing, and concludes the idea is dead. An agent trained on that credit assignment would learn nothing useful, because the absence of reward in a sparse environment carries almost no information.
Six thousand dollars is interesting precisely because it is small. It is not a business outcome. It is a first non-zero reward, which in a sparse setting is worth far more than its magnitude suggests. It tells you the reward function exists, that it is reachable from your current policy, and roughly which direction to move. Zero tells you none of that. The distance between zero and a few thousand dollars is qualitatively larger than the distance between a few thousand and a few hundred thousand.
Why the equity argument is really an argument about state
The founder’s own framing is the most technically interesting part of the story. In his words, if he eventually decides to raise money, he would rather do it after validating the idea with an audience on his own terms than give up equity before he has really tried.
Read that as a statement about state space rather than about money. Taking capital early does not just dilute ownership. It constrains the set of actions available in every future step: which problems you are allowed to pursue, how fast you must move, what counts as failure. He is choosing to keep the action space wide while the environment is still poorly understood, and to narrow it later when his model of the environment is better. That ordering is correct. You commit resources after you have reduced uncertainty, not before.
What the counterexamples actually show
This is not a case for quitting your job. The same source material includes Michelle Yeung, 29, who left a $250,000 software engineering role in New York to open Matcha House and now makes $33,000, having saved roughly $200,000 in liquid cash first. The savings buffer is not a footnote. It is the mechanism that makes exploration survivable.
The TechTual Talk podcast raises the risks of leaving a stable $200K job for a tech career leap without reporting how any particular bet resolved, which is honest. Leon Consulting lays out more than twenty routes to a $200K income, and notes that the real step change comes from moving from employee to owner. None of these sources tell you the odds, because the odds are not the useful part.
The part that transfers
What transfers to the rest of us is the decision structure, not the outcome:
- Exploration has a cost, and that cost is not optional. Budget for it in cash and in time, rather than hoping to avoid it.
- Treat the first non-zero reward as your real milestone. Not the first big number.
- Protect your action space early. Commit resources once uncertainty drops.
- Silence is not evidence. In sparse environments, no signal mostly means not enough attempts.
Treasury may compound or it may stall. The reported $6,000 does not settle that question, and nobody writing about it can. What it does settle is a narrower claim worth holding onto: the information you were waiting for was never going to arrive while you stood still.
🕒 Published: