He correctly notes that the biological fitness landscape changes constantly. Hence the inability to predict the direction of evolution. What he fails to acknowledge is that biological evolution operates on existing and nearly optimized systems, and point mutations are seldom sufficient to crash the system for an individual.
To continue with his factory analogy, a typical mutation might cause a fraction of a percent change in efficiency. Of course it is possible for a mutation to be fatal -- something that happens in biology -- but unlike the factory simulation, the species doesn't die when one individual is defective.
So how do the biological factories survive before reaching a nearly optimal state? The same way real factories survive in the absence of competition. When you are the first and only producer, optimization is not so critical. We see this all the time with new inventions. Before there is competition, products and methods of production tend to be primitive.
At the molecular level we already know that both self replication and Darwinian evolution can occur in the absence of cell membranes or "living" cells. Check out the Spiegelman_Monster. The gaps are getting narrower. Good luck finding new ones.
Across the board, the greater interest is in the information content, e.g. Yockey, the Wimmer experiment, Spiegelman_Monster -- either its origin or evolution.
Because Spiegelman - as with Wimmer - began with information content (RNA) as a "given" - their results are not remarkable, at least to me.
Rocha, Kaufman, et al are much more interesting.
Given that real cells have all kinds of lengths, what is it about the matrix ('primordial soup'), the early cells, or the 'competing moieties' which prevented such behaviour in "real" cells?
Cheers!