1) The big if is written out in the abstract -- "if it is accepted that algorithmic complexity is an appropriate definition of the complexity of a programming project." Relating algorithmic complexity to "how long software takes to write" seems, to me, to ignore that the vast majority of my time as a developer is spent discovering and communicating requirements, handling human questions, not writing novel code. The conclusion touches on this, but ignores it.

2) Even if you accept that "if", this is like a halting-problem proof. Fine; it is impossible to estimate the complexity of ALL software. That does not mean that it's useless to quantify the complexity of software in limited but well-behaved problem spaces. How much of any commercial project is actually spent working on the cutting edge of computer science, facing complete unknowns? An estimate being wrong occasionally is worth most estimates being mostly right.

3) Why do you consider a 20 year old paper that's only been cited 16 times to be critical reading about estimation, when a vast body of research in forecasting exists, written by people who have measurements of the accuracy of estimates to base their theoretical models on?