
Before I worked with AI agents, I built a marketing agency on my own money, with no outside investors. Three habits carried over: ship the imperfect version, make risk cheap to execute, keep every door open. So did one constraint I still have to design around: a brain that prefers starting things to finishing them.
What survived the first business
I paid for each of the three habits in cash, in a business that either made payroll or did not.
Ship over polish is the one people misread most. It does not mean shipping broken things and shrugging. It means the imperfect version that exists beats the perfect version that does not, because only the first can be corrected. An agency does not get six months to workshop an offer before finding out whether customers want it. I shipped the imperfect version, watched what broke, and fixed the parts that mattered instead of the parts I had been most anxious about. That habit has its own essay, Ship the prototype, then listen.
Making risk cheap took me longer to say in one sentence. An entrepreneur is not in the business of minimizing risk. You minimize risk when you are protecting something you already have. While you are still building, you make risk cheaper and faster to execute, as in Risk is an experiment. That compounds, because you get to take the same bet ten times for what most people spend taking it once.
Keeping every door open sounds like indecision until you see what the alternative costs. A bootstrapped business has one asset a funded competitor lacks: nobody can force you to close a door you are not ready to close. No board seat, no term sheet and no investor timeline decides when the experiment ends. That is why I could later leave the agency and rebuild my working life around agents, on my own schedule.
Why move to AI agents instead of building the same business again?
Because the lesson of the first company was never the industry I happened to build it in. The lesson was that someone with no outside money and a willingness to make risk cheap can build something real if the leverage is right. Agents are the largest shift in leverage I have seen in my working life, larger than anything I had when I built the first business by hand. Not using them would have meant choosing the smaller lever out of loyalty to a business I had already shown I could build.
Today I run Hasna, which builds the infrastructure for AI work, so that AI becomes more useful for everyone. I work with a fleet of AI agents and write little code myself. My job is to decide what gets built, hand it to the right agent, and read the evidence when it comes back. The appetite for risk is the same.
The brain that wants a new start
The brain that built the first business nearly sabotaged the second act. I have a dopamine-seeking brain. I mean that as an operating constraint, not as a diagnosis. Every fresh repository, product idea or clever agent workflow gives the hit a slot machine gives, and the hit arrives whether or not anything shipped.
Coding with agents feeds the habit. Every agent pane is a lever. I pull it, something interesting comes back, and the brain that once wanted a new business idea every quarter gets the same thing from a new prompt every few minutes. Recognizing the habit did not cure it. It gave me something more useful to point it at than a rotating set of unfinished side projects.
After a year of coding almost exclusively with agents, I wrote down what I had learned. The dopamine problem opens the list. These are the lessons I still use:
- You have to finish things, or all of it is noise that looks like productivity.
- Code is temporary: every three to six months the tooling changes enough that I want to throw away what I built, so I build small things that help now.
- When a build feels stuck, I feed the model the same problem again and tell it to continue and fix the bugs and the bad logic.
- Architecture and planning matter more than any single detail.
- Being calm matters more than running more agents.
- If you want to feel satisfied, ship it.
Feeding the same problem back is the least intuitive of these, and it is what keeps a product alive past its first week. When a build is buggy, the instinct is to abandon it for a cleaner idea. Feeding it back gives me a new prompt every few minutes while the project stays one project.
The point about architecture is not really about agents. A founder who spends the week perfecting one form field while the system around it is unplanned is working on the part that feels productive. A model does the same. It will polish a detail for as long as I let it and never ask whether the structure around it makes sense.
Calm is on the list because at scale, with more agents, more machines and more parallel goals, the instinct is to speed up further. That is how I end up tired instead of finished. A healthy fleet is one whose owner can still think clearly enough to notice that three of its agents are working on the same mistake.
One product, rebuilt four times
The test of all this is a product I have rebuilt from scratch four times. Going into the fourth round, I told the team it is going to be beautiful because we are going to use everything we learned to build something that works. A dopamine-seeking brain treats a failed attempt as permission to chase a different idea. What I am proud of is treating it as data and starting the next version of the same thing.
Satisfaction never came from starting. It came from the quieter feeling that something exists today that did not exist yesterday. I had it the first time a customer paid for something I had built. I have it now when a product ships, whether I typed it or the agents did.


