Actually, I was suggesting that automated communities of agents cannot function properly in a complex and fluid economic environment, if they communicate through price signals alone. My suggested fix was to automate sharing of knowledge of utility of various resources in different situations, by augmenting the value representation signaling system (pure economy) with non-scalar situational utility data flows. My early experiments show that this "hypereconomy" represents a more adaptive and efficient way to drive a complex multi-agent market.
I do not think that anybody reading New Scientist could guess this from what they quoted. But hopefully, a few interested people may look up my hypereconomy page. I just hope that the others do not get too scared by the "agent economy is unstable and dangerous" statement. Kephart wasn't really trying to make it either. He is just trying to run and study simplified market models, just as I do. I just met him at Alife VI conference; it looks like I may do some work together with his IBM group. In some distant future we may even really make some statements about the stability of the automated markets...