(Edit:) Yes, you can use the estimated number of reads (appropriately rounded) in downstream analysis tools like DESeq or EdgeR. Of course, Kallisto is built with Sleuth in mind, and so other DE tools won't be able to natively take advantage of e.g. the bootstrapped uncertainty estimates produced by Kallisto. Further, as Lior explains very well here, there are other reasons you should prefer Sleuth over using estimated counts with tools like DESeq or EdgeR; the modeling and distributional assumptions are different.