Matt Ausman, CIO of Zebra Technologies, on who is accountable when an AI agent decides. Just after 8 in the morning on 26 June 1974, a cashier in Troy, Ohio scanned a ten-pack of Wrigley's gum. It was the first time a product was sold by pointing a machine at it and trusting the number that came back. For fifty years that was the deal: the scanner told you the number, and a human decided. This year the deal changed. The machines on the warehouse floor stopped reporting and started deciding — reordering stock, moving robots, repricing goods, on their own. Matt Ausman has run generative AI governance across Zebra Technologies for three years. Zebra's scanners, mobile computers and printers are the nervous system of the world's warehouses, shop floors and delivery rounds. He came up through General Electric's operating businesses, so he thinks like an operator rather than a lab — which means when the theory about human oversight meets thousands of people and a deadline, he is the one who has to make it real. In this conversation: The reporting agent that told Zebra it had booked ten billion dollars of revenue in a single segment. Why the big hallucinations do not worry him, and the small ones do.Human in the loop, on the loop, and out of the loop: the three-way split most governance frameworks collapse into one, and how to tell which one you actually have.Why data quality stops being an IT problem the moment one sensor says aspirin and the label says codeine.The audit agent Zebra built to check its own people, and what it found.Why three agents negotiating with each other frightens him more than any single agent going wrong. Context the conversation does not contain: on 2 August 2026 the EU AI Act's duty of human oversight over high-risk systems came into application. Article 14 requires that whoever oversees the system can interrupt it — in the Regulation's own words, "a stop button or a similar procedure". Amer asks Matt whether he has ever run for the power socket. And the prediction, on the record: within ten years, Matt says, we will each have a digital twin — an agent that sees what we see, reads our email, and answers for us. This is the show that calls its predictions in writing. (00:00) From barcodes to AI agents: when machines started deciding (02:27) Matt Ausman, CIO of Zebra Technologies, on his path (06:24) Barcode vs AI agent: what actually changed on the floor (09:04) How accurate does an autonomous AI agent need to be? (12:40) AI on the warehouse floor: what Zebra is running today (15:14) When data quality becomes a safety control, not IT (21:15) Will AI replace frontline workers, or add more hands? (24:57) Human in the loop, on the loop, and out of the loop (29:07) Who is liable when an autonomous AI agent gets it wrong? (34:45) The $10 billion hallucination, and the errors that hide (41:49) Why Gartner says 40% of AI agent projects get cancelled (48:44) Asimov's three laws and the Skynet question for AI (54:21) Can you actually switch an autonomous AI agent off? (58:55) The prediction: a digital twin of you within ten years Links Watch the full conversation on YouTube: https://youtu.be/1ne6AleqARU The Control Layer, free on Substack: https://thecontrollayer.arkava.ai Arkava: https://arkava.ai Zebra Technologies: https://www.zebra.com Matt Ausman on LinkedIn: https://www.linkedin.com/in/matt-ausman-63b15213 Hosted by Amer Altaf, Founder and CEO of Arkava and Managing Editor of The Control Layer. Views expressed by guests are their own. Some figures cited are the speakers' own accounts; independently verified statistics are attributed in the show notes. Sources: first UPC scan, 26 June 1974, Troy, Ohio (GS1 US). Gartner, "Over 40% of Agentic AI Projects Will Be Canceled by End of 2027", 25 June 2025. EU AI Act, Article 14, in application from 2 August 2026.
