Weekend Listens: Where the Human Sits

Weekend Listens: Where the Human Sits

Three interviews about agents doing real work, and one question underneath: where does the person go? A health tech founder who built for autopilot. A Chick-fil-A operator whose agents stop one click short. A security veteran who says a human in the loop is now too slow.

1. The calls nobody had staff to make

CareTalk Podcast, Voice AI Is Expanding Access to Healthcare, with Munjal Shah, co-founder and CEO of Hippocratic AI.

Why listen: Shah's agents phone patients about their care, and he says the use cases that work are the ones health systems never staffed in the first place.

• Why he scoped the agents to a registered nurse's work: clinical, but no diagnosing.
• His case for voice over texts and apps, which patients learn to ignore.
• A worked example on patients lost to follow-up for lung cancer screening, followed to the end.
• Why he thinks the labor replacement thesis is wrong, and where the agents beat it, including 50,000 calls a day through a heat wave.
• The safety setup he describes: thousands of clinicians testing the agents, and a stack of models supervising the one that talks.

On the 2023 bet: "Everybody else was talking about co-pilots and we said, no, you won't get enough leverage unless you build an autopilot."

2. Agents that stop before submit

Restaurant Technology Guys, How a Chick-fil-A Operator Uses AI to Lead, with Jon Harmeling, owner-operator of a Chick-fil-A in Miami Shores, Florida.

Why listen: a former Army officer runs a team of 89 from 19 countries and points his agents at the part of the job that gets dropped on busy days. The people.

• Built on Claude Cowork: agents that read his Slack, watch delivery-app reviews, scan for nearby events, and order uniforms for new hires.
• The same overtime flag, read once as a cost problem and once as a reason to thank someone.
• Mining employee bios for career goals, then building a development program from them.
• His rules: trust but verify, run the old process alongside the new one, and keep every agent stopping short of send.
• The first third is his road from the Army to the restaurant. The agent material starts after.

"I just don't think enough of us are using AI for leadership and using AI for significance."

3. A thousand things at once

a16z, Building Defense for the Agentic Era, with Kevin Mandia, founder and CEO of Armadin and founder of Mandiant.

Why listen: after 30 years in breach response, Mandia now runs swarms of attack agents against client networks, and says defense has to go autonomous too.

• How an agent-led intrusion differs from a human one. A drone swarm, not a sniper. Sloppier, louder, more thorough.
• The "hyperattack": map everything once, then poll for change and attack what moved.
• His claim of more than 90 zero-days found in production since January, from outside, with no source code.
• How his team cages its own attack agents, and why most of the agents it kills are just wasting money.
• a16z is an investor in Armadin, and the last stretch turns to company building.

His line on the security operations center: "if you have humans in the detect and respond loop, you're going to be too slow."

Sources: Interviews from CareTalk Podcast (Oct 2), Munjal Shah of Hippocratic AI; Restaurant Technology Guys (Oct 5), Jon Harmeling of Chick-fil-A Miami Shores; and a16z (Oct 6), Kevin Mandia of Armadin.