{"type":"rich","version":"1.0","provider_name":"Transistor","provider_url":"https://transistor.fm","author_name":"The WorkOps Podcast","title":"Kate Stewart reveals the trade secrets on building AI for HR tasks","html":"<iframe width=\"100%\" height=\"180\" frameborder=\"no\" scrolling=\"no\" seamless src=\"https://share.transistor.fm/e/0ba31d54\"></iframe>","width":"100%","height":180,"duration":1848,"description":"Summary\nOn this episode of the WorkOps Podcast, host Jeet Mukergi talks with Kate Stewart, Staff People Operations, AI and Automations Lead at Horizon3.ai, about what really happens when you put an AI agent into a live people-operations workflow. Kate shares the story of the offer-validation agent she built in Slack to check every job offer against approved comp bands and job architecture, how a Slack update broke it after two months, and the two weeks she spent secretly becoming the agent herself to keep the quality bar from slipping. Along the way she unpacks why a hallucinating agent is more dangerous than a silent one, why automation raises the bar instead of lowering it, and how to ship, break, and maintain automations without burning out. It's a candid, practical listen for anyone in people ops, HR tech, or operations moving from curiosity about AI to actually running it in production.\n\nChapters\n00:00 From the retail floor to people ops\n06:30 Staying curious without a second job\n07:45 The broken process behind every offer\n09:45 Building the offer validator in slack\n11:45 Two months in, the agent breaks\n12:45 Becoming the human agent\n13:45 Why a hallucinating agent is a liability\n14:55 Automation raises the bar\n15:45 From perfectionism to shipping V1\n28:45 Build it, let it break, protect the time\n\nTakeaways\n- When an automation breaks, you don't fall back to your old baseline, you fall below it, because automation raises the bar for what your team considers acceptable\n- A silent agent is confusing, but a hallucinating agent is a liability, a confident wrong answer is far more dangerous than no answer at all\n- Job architecture is not a data hygiene problem, it is an IT provisioning problem, the wrong title in the HRIS means the wrong system access on day one\n- Build one agent to do one job at the right moment, rather than spreading the same check across three people and three separate touchpoints\n- Ship V1, let it break, and protect the time to maintain it,...","thumbnail_url":"https://img.transistorcdn.com/Eh8J8csrLvAf-gfZx2fx-hND9gquazmC6b15qkZyyGA/rs:fill:0:0:1/w:400/h:400/q:60/mb:500000/aHR0cHM6Ly9pbWct/dXBsb2FkLXByb2R1/Y3Rpb24udHJhbnNp/c3Rvci5mbS9lMWNj/OTM3ZjEzMWFkMzQw/NjcwMzQyYTI4MTY3/MGRhMC5wbmc.webp","thumbnail_width":300,"thumbnail_height":300}