Today I had another interesting conversation with a person who is starting a company to unite diverse Quantified Self data streams into a simple, useful API. I was encouraged and discouraged at the same time. This is obviously needed. But there are some serious technical challenges that are invisible when these ideas are pitched and funded, but cause big problems later. These include: lack of insight into way derivation of end point data is achieved in various personal data systems, naivety about dependence on device makers / servie providers maintaining API stability; and, underestimation of the challenge of responding to API calls quickly and efficiently enough to support the app "ecosystem" that is typically envisioned. A bit more realistic dialog about these issues would help everybody.