Bryan Hong led talent at Astranis, a satellite manufacturer, then ran people ops and built an AI services team for the whole company, becoming an engineering manager at 38 after 12 years in recruiting. His session covered the part nobody teaches: getting an AI workflow from demo to production.
Everyone has built the demo. Then IT says no, and the ticket disappears. Part of the gap is that engineers expect exact outputs and tightly scoped access. AI is non-deterministic, and recruiters already have access to everything in the ATS.
Gate 1: Is it worth it? Fix the process first, then use what your system already does, then native integrations. AI comes last. Good candidates are repetitive, high-volume and reversible if they go wrong. Make the case as cost per task: minutes saved minus build, maintenance and token costs.
Gate 2: Who owns it? Recruiting defines what good looks like, engineering makes sure it’s reliable, and security approves the data. Someone has to be on the hook when it breaks. His tip for security reviews: submit every connection in one request, and remind them you already have access to all of it through the browser.
Gate 3: Prove it. Pilot small and check every output. If it’s still not working after 2 or 3 weeks, move on, because it will probably ship natively within 3 months.
Gate 4: Rigor. Could you take a month off and come back to find it still working? Retest before every update, and never check a model with the same model. (Maker doesn’t equal checker.)
His advice: build infrastructure, not one-off solutions. A pipeline that transcribes audio and drafts a decision can serve debriefs, screens, calibrations and interview training. And find an engineering sponsor. Your recruiters already know which engineers are friendly.
