The shape of the AGN luminosity function at high redshift has been hard to determine, as previous studies have been limited by small amounts of high quality spectra and/or by the small area of their surveys. The shape of this luminosity function at the faint end has implications for the contribution of AGNs to the ionizing budget during the epoch of reionization. Recent work in the HETDEX/SHELA field at z=4 has shown that the faint-end slope of the AGN luminosity function may be steeper than previously thought, but this result is degenerate with the shape of the bright-end of the galaxy luminosity function. This degeneracy can be broken by using spectroscopy to empirically measure the AGN fraction amongst objects at MUV = -23. We tackle this problem at z~3, where spectra from the Hobby Eberly Telescope Dark Energy Experiment (HETDEX) probe AGN-sensitive rest-ultraviolet emission lines. Here we present a sample of ~500 spectra for objects at z~3 across the 24 deg2 of the HETDEX survey in the SHELA field. We used an affine invariant Markov chain Monte Carlo (MCMC) Ensemble sampler to extract information from the spectra and a decision tree machine learning algorithm with a sample of known AGNs to classify the spectra. We find that these spectra allow us to classify AGN at fairly high confidence, and future work will allow us to directly measure the AGN fraction.