Abstract The evolution of the galaxy size–mass relation has been a puzzle for over a decade. High-redshift galaxies are significantly more compact than galaxies observed today at an equivalent mass, but how much of this apparent growth is driven by progenitor bias, minor mergers, secular processes, or feedback from active galactic nuclei (AGNs) is unclear. To help disentangle the physical mechanisms at work by addressing the latter, we study the size–M stellar relation of 32 carefully selected broad-line AGN hosts at 1.2 < z < 1.7 (7.5 < log M BH < 8.5; L bol/L Edd ≳ 0.1). Using the Hubble Space Telescope with multiband photometry and state-of-the-art modeling techniques, we measure half-light radii while accounting for uncertainties from subtracting bright central point sources. We find AGN hosts to have sizes ranging from ∼1 to 6 kpc at M stellar ∼ (0.3–1) × 1011 M ⊙. Thus, many hosts have intermediate sizes as compared to equal-mass star-forming and quiescent galaxies. While inconsistent with the idea that AGN feedback may induce an increase in galaxy sizes, this finding is consistent with hypotheses in which AGNs preferentially occur in systems with prior concentrated gas reservoirs, or are involved in a secular compaction processes perhaps responsible for building their bulges. If driven by minor mergers that do not grow central black holes as fast as they do bulge-like stellar structures, such a process would explain both the galaxy size–mass relation observed here and the evolution in the black hole–bulge mass relation described in a companion paper.