The year is 2021. My partner and I are looking to raise money for our venture: a stock market trading algorithm built on an AI model we developed ourselves.
The deal that almost happened
We go through several excellent meetings with an investor, including serious due diligence with someone genuinely technical on his side. We close on terms: 2 million dollars for 27 percent.
Then we talk to our lawyers. They hint, delicately, that we really should not get involved with this particular investor. That he is a problematic person, and that if things ever go sideways, dealing with him could get ugly in all kinds of ways.
From there to here, we ended up turning the offer down.
The bug that showed up the next day
One day later we discover a bug in the system. Not a small one. The kind of bug that would make everything we had shown the investor look like we had lied to his face.
And the two of us, instead of being crushed about the bug and the fact that we now had to tear the system apart and rebuild it, were mostly relieved that we had not taken the money and did not have to explain ourselves to anyone.
The technical due diligence was serious, done by a capable person, and it missed the bug completely. So did we, and we built the thing.
In the end we kept bootstrapping, and later raised money on Shark Tank Israel instead.
What is the actual lesson?
The punchline I used at the time was that you cannot do due diligence on software. The serious version: even a rigorous technical review by a smart reviewer is a snapshot of what the system appears to do, not a proof of what it does. The people who built the system did not know the bug was there. A few hours of an outsider reading it had no realistic chance.
That experience shapes how I build today. It is why I care so much about control mechanisms, logging, and verification loops in every system I ship, including the AI agents I build for clients: not because reviews are worthless, but because the only thing that catches what everyone missed is instrumentation that runs all the time.