Renan Delonier joins me to unpack what happens when you stop talking about AI as a buzzword—and start putting it into real factories with real consequences.
Most conversations about artificial intelligence live in demos and slide decks. This one lives in production environments, broken trust, failed deployments, and hard-earned iteration.
Renan is the CEO and co-founder of OSS Ventures, a venture studio that co-builds software companies exclusively for factories. After visiting more than 800 industrial sites and launching 22 companies inside 3,000 factories, he’s developed a structured approach to identifying operational pain—and solving it with software.
But this episode isn’t about hype.
It’s about what went wrong when they “bet the house” on AI.
We walk through the painful reality of pushing generative AI into production too early—70% error rates, broken client confidence, and discovering that prompting large models with raw data is not a strategy. It’s a demo.
The breakthrough came from rethinking the problem entirely.
Instead of asking AI to generate outcomes, they used it to generate code. Instead of trusting datasets, they extracted the undocumented rules living inside expert operators’ heads. In one factory alone, they uncovered 550 decision rules that existed nowhere in any ERP system.
We explore why most automation fails in manufacturing, why Excel survives inside billion-dollar operations, and why the real design target in AI systems isn’t the frontline user—but the manager responsible for auditing and controlling the machine.
This is a candid discussion about product-market fit in heavy industry, frictionless B2B offers, founder-led companies versus bureaucracies, and what it actually takes to deliver a 10x outcome in operational environments.
The lesson isn’t that AI will replace humans.
It’s that intelligence—codified correctly—can finally scale the complexity humans have been carrying in their heads for decades.
TL;DR
AI demos are easy. Production is brutal.
Prompting models with raw data fails at scale.
GenAI’s real leverage is collapsing the cost of code generation.
Most critical factory rules live only inside expert operators’ heads.
ERP systems automate the simple; humans absorb the complexity.
Design AI systems for control and audit—not flash.
A 10x offer changes how conservative industries respond.
Memorable Lines
“AI is not one prompt with data.”
“Cost of code has collapsed 100x.”
“The rules weren’t in the system—they were in his head.”
“Free until it works.”
“Design for the person managing the machine, not the machine itself.”
Guest
Renan Delonier — CEO & Co-Founder, OSS Ventures
Factory-focused venture builder launching AI-native software companies in industrial environments. OSS Ventures has co-founded 22 companies operating across thousands of factories worldwide.
🔗 https://oss.ventures
Why This Matters
Manufacturing runs the physical world.
And yet much of its decision-making still lives in spreadsheets, tribal knowledge, and undocumented heuristics buried inside senior operators’ experience.
As AI moves from experimentation to infrastructure, the advantage will not belong to the loudest adopters—but to those who can codify complexity without losing control.
For founders, operators, and executives rebuilding systems inside volatile environments, this episode reframes AI not as a disruption story—but as a discipline story.
The edge isn’t intelligence alone.
It’s structured intelligence that survives contact with reality.
