I started developing AFNI in June 1994, in response to the pleas of the neuroscientists at the Medical College of Wisconsin (where I was then a Senior Research Scientist). They needed a way to transform their FMRI results to Talairach coordinates. My original plan was to upgrade the old program FD2 to do this. I quickly changed my mind, and hatched the scheme of a program that would let the user browse through 5 "dimensions" of data: xyz coordinates, plus scanning run, plus subject. So in a frenzy of C coding, I wrote the original version (number 0.5) and had it working by September 1994 (just in time for the site visit for our Program Project Grant).
This earliest version used the HDF file format to store datasets. I discovered that this was grossly slower than direct Unix file I/O (a factor of 50 or so), so rewrote AFNI to use the .HEAD/.BRIK file combination. With the addition of the programs 3dmerge, 3dclust, and 3dttest, this collection of software became AFNI version 1.0. I released this on the Internet in February 1995.
3D+time datasets and bucket dataset types were added in 1996. These capabilities added a 6th dimension to the AFNI setup, and set the stage for AFNI to become an reasonably complete environment for the processing of volumetric FMRI data. This was also the year that I hired Doug Ward, who contributed so much to the AFNI package with his statistical analysis programs (3dANOVA, 3dNLfim, 3dDeconvolve, etc.).
Now, I've moved to the NIH (January 2001), but the development of AFNI will continue. (If you want proof, see the Latest News Page.)