I find myself considering if I even want to turn my years of experience into a bunch of ML-ops rules for these people in the first place...
I find myself considering if I even want to turn my years of experience into a bunch of ML-ops rules for these people in the first place...
This was the most painful lesson we had to learn so far.
The bootstrapping is the hardest piece looking back on the last half-decade. Once you figure out how to make 3 different customers work using the same code pile, you are moving in a good direction. Implementing for 1 customer at a time in a serial fashion is a path for guaranteed failure if you are expecting to just copy that code pile to the next one and have it fit their needs. We tried to do this 3 times and only by the grace and understanding of our angel investor are we still doing business.
My new job is to ask the question "Will this feature/function/enhancement work for all current and future hypothetical clients?". Forcing the team to think in this way was a challenge, but it's kind of our default mindset now. When you guard for bullshit like "what if the account number is longer than 64-bit integers for some future customer?", you trivially-sidestep things that have literally killed other businesses. Clearly, you can take all of this too far, but only if you actually implement in code the proposal before realizing your mistake.