Saul Howard is vice president of engineering at Anterior, developing healthcare AI systems that safeguard sensitive information, preserve clinical oversight, and produce defensible records of their decisions. He also co-founded Deep Drama, which applies models of human conflict and cooperation to intelligent software.
Howard previously worked at startups in London and Asia, including Hailo, and from 2016 to 2021 led an Apple team building the CloudKit developer platform. His professional background also includes software development at Clear Line Tech, virtual-reality and mobile projects at Catalyst VR, and producing the feature film Brighton Wok.
With Deep Drama co-founder Manuel Puerto, he explored virtual agents capable of representing goals, conflicts, and cooperation, with potential applications in training simulations and interactive characters. His writing on intelligence analysis develops a complementary argument: language models should help people organize evidence, challenge assumptions, and compare interpretations instead of substituting persuasive predictions for human judgment. He has also described using language models to structure competing interests through Drama Theory.
- Immutable audit histories: Howard favors append-only event records that capture an agent’s actions, authorizations, and available information, making healthcare decisions reconstructable after the fact. The tradeoff is more complicated reads, which can require snapshots or cached projections.
- Separating patient data from orchestration: His architectural approach stores protected health information separately from operational events, which reference sensitive records without embedding them. That separation supports debugging without exposing patient information, customer-controlled infrastructure, access restrictions, and defenses against prompt injection.
- Clinicians and AI in the same workflow: Howard and Christopher Lovejoy describe architectures in which human and model agents operate from shared context, allowing clinical escalation without interrupting a process or losing its history. Their healthcare-agent architecture also supports privacy-preserving evaluation by replaying decisions and comparing human and model performance within protected environments.
- AI as the new product interface: Howard argues that AI changes how healthcare software gets built, enabling clinicians to shape workflows and evaluation datasets directly while making reliable, secure infrastructure more consequential.
His critique of generic AI-generated work extends that standard to individual contributors: as language models raise the baseline, distinctive human judgment and added value become harder to demonstrate—and more important.
