Why Generic AI Fails Engineering: PTC’s 60% → 93% Answer

A generic LLM may deliver only 60% accuracy on engineering requirements. PTC says domain-grounded engineering AI can reach roughly 93%—and in safety-critical product development, that gap is the whole story. PTC leaders John Haller and Steve Dertien join Michael “Fino” Finocchiaro to explain what trustworthy engineering AI actually requires beyond a general-purpose model. The conversation covers deterministic results in regulated industries, engineering context, enterprise permissions and IP protection, advise/assist/automate, AI agents and MCP, Windchill, Creo, Codebeamer, FlexPLM, Jetstream, Orbit, change management, the digital thread, and the future of engineering work. CHAPTERS 00:00 The 60% AI problem — cold open 00:38 Meet Steve Dertien and John Haller of PTC 02:12 Why engineering AI needs domain context 12:02 Advise, assist, automate 14:56 PTC Orbit and Jetstream 27:23 Connecting the digital thread 34:00 The “OpenAI moment” for engineering 42:06 Startups, incumbents, and what comes next 49:11 PTC Next and closing thoughts GUESTS John Haller — DVP \& General Manager, Product Lifecycle Management, PTC Steve Dertien — EVP, Customer \& Technology Partnerships, PTC HOST Michael “Fino” Finocchiaro — AI Across the Product Lifecycle What is the biggest barrier to trustworthy engineering AI in your organization: accuracy, context, permissions, change management, or something else? Tell us in the comments. Subscribe for candid conversations about AI, PLM, the digital thread, manufacturing, and product development. \#EngineeringAI \#PTC \#PLM