Manufacturing Happy Hour · 7 Jul 2026 · 1h 06m
Andres Naranjo, CEO of Software Toolbox, discusses his journey from a sabbatical sailing around the world to building the first full-stack industrial AI company. He explains why he acquired Software Toolbox to gain distribution and data moats, and outlines his framework for deciding when to build versus buy. Naranjo details the evolution of AI types—machine learning, computer vision, generative AI, and agentic AI—and argues for replacing static dashboards with real-time, AI-driven insights on the plant floor.
Where this conversation describes AI being used to build or deliver a product. Part of the AI in practice tracker.
Engineering & simulationMachine learning model (unspecified)
Predicted the number of 12-volt batteries to stock based on weather data to optimize inventory for a car parts client.
“we built a machine learning model to predict the number of batteries to stock based on what the weather is doing. So in 2018, I was like, wow, this is a really powerful tool. We no longer are slicing and dicing data, we are using data to train to get a prediction.”
Speaker 1 · Hear it at 8:34
Engineering & simulationMachine learning model (unspecified)
Built a product to predict causes of downtime related to labor and material availability rather than just machine failure.
“And so that's what we set out to do. And yeah, lo and behold, like all the causes downtime were because of that, the wrong operator didn't show up to drive the forklift. There's not enough bottles upstream in the in the in the warehouse.”
Speaker 1 · Hear it at 18:12
Engineering & simulationGenerative AI (early models)
Built a generative product to perform conversational analytics on a core SaaS application, providing shift analytics and charge recommendations.
“And we, and then we started building a generative product to do conversational analytics on their, on the core, uh, SAS application and it was doing, isn't, this is now 2 years ago. Two years ago. Yeah, it was incredible. It's like a lot of things that people are doing now, just like getting you shift analytics, giving you charge, giving you recommendations with some of the early models.”
Speaker 1 · Hear it at 18:12
Engineering & simulationComputer vision (Deep Learning)
Used computer vision to scan for defects and security hazards, such as detecting if people were driving forklifts in the wrong parts of a plant.
“That is tons of applications in manufacturing. We use it for the scan for defects, to scan for security hazards. Uh, I talked to somebody at, um, last week that worked in the chemical space. They look at to see if people are driving forklifts in the wrong part of the plants, things like that.”
Speaker 1 · Hear it at 39:49
companies scaling up
Other shows on the same ground, closest first.
Alex Sandoval, CEO and co-founder of Allie Systems, explains the distinctions between machine learning, deep learning, and generative AI, and how these technologies apply to manufacturing operations. He details Allie Systems' approach of connecting PLCs and sensors to predict downtime and optimize processes, while also addressing the cultural resistance to AI adoption among traditional manufacturing leaders. Sandoval advises companies to start by forming a transformation squad, securing historical data from machines, and partnering with technology providers rather than building solutions in-house.
Hosts Scott Wilcox and Matt Perrott discuss their collaboration between BuildPass and W, focusing on implementing agent-first workflows. Scott details using Paperclip to simulate an organization with CEO and employee agents, while Matt explains onboarding his entire company onto Claude to capture institutional knowledge. They discuss the strategic shift from general software to vertical, industry-specific solutions in construction, leveraging tools like Replit and Notebook LM to accelerate development and customer intimacy.
Christopher Savoie, founder and CEO of Zapata AI, discusses his career from early biophysics research to co-inventing the technology behind Apple Siri. He argues for specialized, ensemble-based AI agents over monolithic models, citing applications in motorsports with Andretti Autosport and grid management. Savoie highlights how predictive analytics in racing and power grids reduce surprises, while warning about the privacy and ethical risks of big data surveillance by governments and corporations.
Aron Semle, Chief Technology Officer of HighByte, explains Industrial DataOps as the engineering discipline required to contextualize heterogeneous data from factory floor equipment for IT systems. He discusses the challenges of scaling solutions across facilities with different machinery, using autonomous guided vehicles (AGVs) as a key example of where standardized data models prevent redundant integration work. Semle also details the rise of AI agents in manufacturing, describing how they can automate monitoring tasks and capture tribal knowledge to assist operators with aging workforces.
Sam Smith-Eppsteiner, a partner at Innovation Endeavors, argues that specialized vertical AI models are necessary for physical industries due to siloed, unstructured, and private data that general-purpose models struggle to utilize effectively. She outlines four key application categories: AI as a complex knowledge base for engineering, agentic automation for repetitive tasks like Hubflow’s trucker scheduling, engineering copilots such as Cadstrom’s PCB verification tool, and compliance automation exemplified by WeaveBio’s regulatory platform. Smith-Eppsteiner also discusses business model innovations, noting that selling 'the work' rather than just software allows companies to tap into larger labor budgets and achieve higher top-line revenue despite lower margins from inference costs.
The summary, chapters and topics are generated from the episode’s audio by this directory, not written by Manufacturing Happy Hour. The episode itself belongs to its publisher.