Apprentice.io's Angelo Stracquatanio on Building Manufacturing AI That Operators Can Trust
Manufacturing Happy Hour
In this episode of Manufacturing Happy Hour, host Chris Luecke talks with Angelo Stracquatanio, co-founder and CEO of Apprentice.io, about the company's twelve years building software for manufacturing and its newest product, an AI agent called A1. Stracquatanio's position is consistent across the conversation. In manufacturing, humans must stay the driving force, and AI has to be purpose-built, deterministic enough to repeat reliably, and earned through trust rather than hype. Along the way, Stracquatanio covers the company's pivots, its role during COVID, lessons about co-founders and teams, and how Apprentice rebuilt its own operations around AI before asking customers to do the same.
Why the Company Is Called "Apprentice"
Asked at the outset why technology should be seen as an apprentice to manufacturing, Stracquatanio explains that the name dates back twelve years. At that time they believed that humans in manufacturing would one day have to work alongside some kind of intelligent tool. They say they had no idea it would turn out to be AI as it exists today. The core belief behind the name, though, has only grown stronger for them: in manufacturing especially, humans remain the driving force and AI is a tool that supports them. In Stracquatanio's view, the name has not only lasted but fits the company's current focus better than ever.
A 200-Page Binder for a Cancer Therapeutic
The company began in pharmaceutical manufacturing because of Stracquatanio's wife, who works in biopharmaceutical (biotech) manufacturing. Twelve years ago the couple lived in Jersey City, and she would come home talking about what had gone wrong at work, which Stracquatanio politely distinguishes from complaining. Her site in New Jersey was running advanced, complex biotech manufacturing of a cancer therapeutic, and the process ran on a 200-page paper binder.
To Stracquatanio, then a software engineer, it seemed absurd that what they assumed was the most cutting-edge kind of facility ran on paper. That pain became the founding premise: bring excellent tools to people on the shop floor so that, ultimately, drug product gets out the door faster. Apprentice now works with manufacturers across many industries, but Stracquatanio stresses that it started with one simple, concrete problem.
From AR Headsets to an Agent Above the Stack
When Luecke asks how the mission has changed since 2014, Stracquatanio says one thing has stayed exactly the same: the company has only ever built tools for people who work in manufacturing. The product portfolio, however, has changed dramatically.
Apprentice started with augmented reality. Google Glass had just come out, and Stracquatanio says they "incorrectly thought" a headset would drive technology change in manufacturing. The reasoning was ergonomic. Workers need both hands, and holding an iPad or using a computer is less natural than wearing a headset. The headsets are still around the office. From there, the company moved into a manufacturing execution system (MES), which grew into a category Apprentice calls a "connected manufacturing network." Apprentice then acquired a manufacturing intelligence company focused on analytics and data processing, built its own edge hardware device to reach level one, and built a laboratory execution system and an L2 automation system. Its newest product is the A1 AI agent, which sits above all of these. Stracquatanio frames the next chapter as the same mission carried out with AI. They hope the company can guide customers through this transition as well.
ISA-95 and Why AI Can't Live in One System
Luecke asks Stracquatanio to explain the ISA-95 standard as if over a beer. Stracquatanio describes manufacturing as having many systems arranged in levels. Level four holds enterprise systems such as the ERP (SAP, NetSuite) and the quality management system, which spans the enterprise. At the site level is the MES, Apprentice's "bread and butter" for over a decade. Below that sit control systems (PLCs, SCADA, the automation stack), and below those, sensor-level data. The hierarchy is also organized by timescale, which Stracquatanio chooses not to get into.
The point Stracquatanio draws from this is that AI built into any single system is "trapped in a box." An AI embedded in the MES can't really interface with the DCS; one in the DCS can't really interface with the ERP. That led Apprentice to build a new layer that is not part of the current standard and that sits above all existing systems. The company is "playing around with" calling it level five, though Stracquatanio wonders aloud whether that goes too far. This layer connects to all four levels, letting one AI and a set of agents tied to people perform tasks across the subsystems.
From Engineer to Wartime CEO, and Back to Writing Code
Asked how they went from software and mobile app developer to entrepreneur, Stracquatanio describes several transitions rather than one. For the first four years, the company was just Stracquatanio and co-founder Gary, who still leads sales. Stracquatanio wrote every line of code and also handled deployments, support, and sales, with everything running on a personal credit card.
After landing the first customers, the pair had to build an actual company. They raised a first round around late 2017 to early 2018, when the team was under about 20 people. Then COVID arrived. Without naming the company, Stracquatanio says Apprentice helped manufacture 300 million doses of a COVID vaccine. The company grew dramatically and raised over $200 million, and Stracquatanio had to shift again from manager to CEO of a multi-product company with layers of management.
Next came what Stracquatanio calls a really big challenge. In 2023 and 2024, the life sciences market, which made up most of Apprentice's revenue at the time, slowed down. Stracquatanio attributes this in part to the Inflation Reduction Act. Because it let the government negotiate drug prices for the first time, Stracquatanio says, it "kind of froze the industry," which didn't know what the change would mean. Stracquatanio describes leading through this period as a "wartime CEO."
Coming out of that, the company ran straight into AI and has focused on it for the past two years. Stracquatanio says the story has come full circle. With A1 they are writing code again and working shoulder to shoulder with the engineers, so they have personally touched anything a customer uses. The moment feels like twelve years ago: new technology, learning on the fly, working directly with customers. Stracquatanio also notes that they didn't come from manufacturing. In the early days they spent literal years in manufacturing suites with customers, "gowned up head to toe" and sweating. Today, Stracquatanio notes, Palantir calls this a forward deployed engineer; back then Stracquatanio just called it learning. Now they are doing the same thing again, because they believe AI can have either a strongly positive or a negative impact and they want to get it right with customers.
"Predict and Prepare": How Apprentice Anticipated COVID
Luecke asks how others can position themselves to capitalize on an opportunity like the one COVID created. Stracquatanio points to an internal operating phrase, "predict and prepare," and a related process called PDS: predict, discuss, and solve. The principle is to always look around the corner. Working in the business is hard, but leaders have to spend as much time working on it, which means keeping a pulse on the market and on how customer concerns keep shifting. By Stracquatanio's count, the company has gone through this cycle of predicting something, aligning the business to it, and capitalizing on it about four or almost five times.
With COVID, in January of that year a team member named Frank, who now hosts Apprentice's weekly podcast, posted an article in Slack about something alarming happening in China. The team discussed it and concluded it could become a real problem. Before travel was shut down, they responded to rumors of travel restrictions by repackaging an existing but minor product, remote telepresence. With it, a worker on the shop floor wears a headset while someone else, possibly in their bedroom, guides them through the process. Apprentice made the product quick to download and start using, so when the pandemic hit it was ready.
Customers used it for the vaccine work itself, which Stracquatanio describes as trying to solve COVID with COVID all around them, and as "kind of scary, honestly." Stracquatanio says an even bigger concern was keeping global operations running. Supply chains were disrupted and people couldn't travel to sites, help colleagues, or troubleshoot in person. The vaccine is the headline, but Stracquatanio says the real stars were customers using these tools to keep drug supply chains open. They also briefly mention "Formula 45s," internal 45-day sprints across the whole business, before moving on.
Complementing Weaknesses and "Molecular-Level Trust"
On building a lasting team, Stracquatanio notes that despite the hard years of 2023–2024, average tenure at Apprentice is approaching four years, which they consider long for a software business. The chief customer officer is coming up on nine years, and some team members have been there ten. Stracquatanio credits two things. The first is the mission: the software is only one piece, and it serves people who build important products for people around the world. The second, which Stracquatanio says they learned through many mistakes, is that the leadership team needs to include people who complement their specific weaknesses.
Stracquatanio describes their younger self as "a pretty awkward dude," a software engineer who "did not know how to sell software, period, full stop." Gary was and remains the perfect complement. Over twelve years they have built what Stracquatanio calls "molecular-level trust": each knows how the other thinks, and conflicts get resolved quickly because they have worked through so many. Similar trust across the wider team means that when a new product launches, Apprentice can pivot the whole business overnight, adopt new tools, change strategy, and work around the clock "with zero drama." Stracquatanio emphasizes that this trust matters for getting through both good times and very tough ones.
Becoming AI-Native Internally: From Quarterly to Weekly Tempo
Asked for advice to manufacturers navigating the AI shift, Stracquatanio starts with their own company. It would be "a bit ridiculous," they say, to tell customers to use a product to become AI-native if Apprentice hadn't done so itself. For more than six months, the company has been going through a major internal transformation across software and operations.
Apprentice used to run on a "quarterly tempo." Every week of the quarter was pre-planned, and teams were coordinated, so that a product release, for example, was matched by marketing and sales webinars and announcements. Stracquatanio says they have compressed what used to take a quarter into a literal week and now run a weekly tempo. They attribute this compression to using agents internally, including A1 itself with agents and workflows built for internal operations. According to Stracquatanio, this cut manual work dramatically, raised people's throughput, and sped up the tempo and "takt time" of the whole business.
The company then overhauled its internal systems, building some tools in-house and upgrading others to new software, with A1 on top. It also changed people's roles. Engineers became "super stack engineers" who work across front end, back end, mobile, and infrastructure, and across business units. Stracquatanio's reasoning is that raising tempo, refocusing people, and using AI for throughput lets Apprentice serve customers better and speak credibly about how to do it. Anything else, they say, would feel hypocritical.
What A1 Is: Agents Mapped to Human Roles
In casual terms, Stracquatanio describes A1 as a new, largely standalone category that sits on top of existing manufacturing systems. A1 has sub-agents mapped to the roles in a manufacturing facility, which reflects what Stracquatanio calls a human-first approach. Examples include operator, quality, process engineering, and maintenance agents, along with an agent Stracquatanio names that is rendered unclearly in the transcript, plus supervisor, leadership, and site planning agents. The agents are meant to complement current workers by automating workflows that are tedious, manual, paper-based, or hard because they span many systems.
Stracquatanio gives two examples. The operator agent can take a set of paper-driven SOPs, find the relevant section when an operator needs to troubleshoot something, and turn it into a clickable, step-by-step troubleshooting guide. Stracquatanio says this goes from paper to a digital procedure "in a couple seconds." The quality agent finds exceptions, does an initial pass on the investigation, and helps the quality leader understand what's happening in the plant and close exceptions faster. The longer-term goal is to automate work across teams, functions, and systems, with the aim of making plants more functional and raising throughput.
Event-Driven, Not a Chatbot: The Alarm Triage Example
Luecke asks for a cross-role scenario. Stracquatanio first notes that teams sit at different points on digital and AI maturity curves. Some will use agents for basic floor-level tasks, while others are ready for advanced workflows. Stracquatanio then stresses that A1 is not a chatbot designed for prompt-and-response. It is meant to react to manufacturing events: alarms, MQTT messages from IoT, material flow, shift-to-shift handoffs, and movement between work cells.
The example is an out-of-the-box alarm triage agent. When an alarm fires, the agent investigates on its own. It pulls the active run from the MES, checks the control system, gathers any events flagged in the IoT publishing system, combines the findings into an analysis, and messages the manufacturing team that the alarm deserves a closer look. Stracquatanio says teams are "truly drowning" in alarms and, unfortunately, often don't look at every one. The agent does the first pass, so humans can decide which alarms need a detailed response, such as ones that might take a process off spec. Teams can start with a role-based agent and progress to trigger-based workflows that span many systems and people, producing outputs Stracquatanio says would be "next to impossible" today.
Constraining a Probabilistic System: Apprentice 4.1 and Structured Workflows
Stracquatanio explains that AI is probabilistic: the same prompt can return different answers. In manufacturing, which requires determinism, repeatability, and consistency, that is a problem. Apprentice constrains the AI in two ways.
The first is its own model. Apprentice fine-tuned a state-of-the-art model on twelve years of the company's knowledge, customer data, and expertise, producing a model it calls Apprentice 4.1. According to Stracquatanio, it gives hyper-specific manufacturing answers rather than general ones and takes compliance requirements and the relevant process into account.
The second is structured workflows inside A1. They follow a step-by-step form familiar to manufacturing: step one, triage; step two, get specific data from the MES; step three, contact the on-call supervisor for the shift. Customers can define their own, and Apprentice ships what Stracquatanio describes as hundreds of out-of-the-box workflows. Users don't have to specify every API endpoint, data point, or analysis. They describe generally what the agent should do, and the strict structure is what Stracquatanio says makes repeatability "shoot through the roof." In their view, that consistency is what allows teams to trust AI on the floor rather than getting different results each time.
What Makes It "Purpose-Built"
Luecke asks directly what makes A1 purpose-built. Stracquatanio walks through the layers. At the model layer, general models like ChatGPT or Claude are trained on broad knowledge and give broad results. Apprentice has used such general models inside its MES for two years and found they gave "general results." To address this, Apprentice used reinforcement learning with domain experts, people with a decade or more of manufacturing experience from both inside and outside the company, to train the model on what good manufacturing output looks like, covering context, data, process, and compliance.
At the agent or "harness" layer, which Stracquatanio describes as the shell around the model that performs actions, Apprentice built a workflow engine designed to connect to manufacturing systems. Stracquatanio contrasts this with OpenAI and Anthropic, which, in their words, want users connecting to generic tools such as monday.com and other collaboration apps. Manufacturing systems live at the lower ISA-95 levels. Apprentice therefore built an on-prem connector that understands those subsystems' data shapes, API endpoints, and data hierarchy.
At the output layer, the connector and agents produce what Apprentice calls manufacturing artifacts: troubleshooting guides, process flow diagrams, SOPs, trend reports, and shift summaries. Stracquatanio argues that manufacturing data is precise, not generic. If AI isn't precise, no one will trust it, and without trust it won't be adopted, which would keep the industry from gaining the benefits Stracquatanio believes it can deliver.
Trust, Credibility, and a Free Tier With No End Date
As the interview winds down, Luecke asks whether trust in AI is moving in the right direction. Stracquatanio answers from experience. Over twelve years, the hardest thing to learn wasn't software, entrepreneurship, or being CEO. It was earning customers' trust. Early on, customers didn't trust Apprentice to run their operations, and Stracquatanio says it took more than a decade to earn the right to run some of the world's most complex manufacturing processes. They expect the same for AI. Adoption won't happen just because it's a buzzword or because a C-suite or board is demanding it. There is, in Stracquatanio's words, "a certain gravity" to earning trust in manufacturing that can't be bypassed.
Apprentice's answer is to lower the barrier to entry. Stracquatanio says the free tier "can do literally everything": it connects to subsystems, includes pre-built agents, and supports MQTT triggers from IoT, all at no cost. It is not a 30-day trial or a limited version. When Luecke calls it a free trial, Stracquatanio corrects this: there's no end date. The reasoning is that manufacturing people need to explore and use a tool themselves before believing anything a vendor says. If a simple use case works, such as uploading some SOPs and getting correct results back, users may gain enough confidence to bring it to their team, then their site, and possibly their enterprise. Stracquatanio also invites feedback directly by email.
Looking Ahead: From Personalized Medicine to Hyper-Personalized Products
For the wild-card question, Stracquatanio shares ideas they describe as "early innings" and admits may sound a bit ridiculous. Some of Apprentice's life sciences customers produce personalized medicine using CRISPR-based technology. Stracquatanio describes it as allowing people, for the first time, to edit their own genes to treat and even cure disease. Manufacturing in this area happens on a one-patient basis, with an entire run for a single person, and Stracquatanio says the company has learned a lot from it.
Stracquatanio believes AI could extend this kind of personalization beyond medicine to products in general, on a per-individual basis, by enabling high-mix manufacturing. They acknowledge this doesn't seem realistic yet, because equipment, processes, and changeovers were designed for something else. Still, they are thinking about a future where manufacturing isn't the same thing made over and over consistently, but "a different thing over and over again consistently." They also mention having strong beliefs and questions about humanoid robots. Despite being a tech person, they say they are "not so hot" on humanoids and that, personally, they don't see them making sense in manufacturing. They leave that topic for another time. The conversation ends on these open questions, with Luecke suggesting they revisit them in a future episode.
So, we have an operational phrase internally called predict and prepare. You look around the corner at all times because even though working in the business is really hard and there's a lot of things you need to do, you have to spend as much time working on the business. Understand what your customers' concerns actually are and how they shift, constantly shift around you.
Today, we're interviewing Angelo Stracquatanio, co-founder and CEO of Apprentice. apprentice.io is one of our partners here at Manufacturing Happy Hour focused on purpose-built AI for manufacturing. Angelo is going to take us through his entrepreneurial journey, where his company has been, and how Apprentice has grown over the past decade.
Angelo Stracquatanio, welcome to Manufacturing Happy Hour, and you have a perfect background today of the New York City skyline from your view in New Jersey. So, in Manufacturing Happy Hour fashion, the first question has to be: if we were grabbing a beverage in your neck of the woods, where would it be? Paint the picture for us.
Man, there's this old-school Italian place called Leo's here in Jersey City. It's on a residential street. You'd never know it was even there. And you go inside and it looks like a classic diner from New Jersey, just 50 years older than that and Italian. So, to me, it's my absolute favorite place around here, not only to get a drink but to get a really awesome, in New Jersey fashion, a great Italian dinner.
Absolutely. I love New Jersey diner culture. I love the Italian food out in the Northeast. Well, let's say we're hanging out at Leo's, is what you said, correct?
That's right.
So we're at Leo's. Your company is called Apprentice. So, I want you to answer this as if we're having a beverage and some Italian food. Why do you believe technology is an apprentice to the manufacturing environment, if you will?
Yeah, so 12 years ago now, when we started the company, I named it Apprentice because I always believed that one day humans would have to work with some type of intelligent tool in manufacturing. It was a pretty simple idea at the time. And, 12 years ago, did I know it was going to be AI? Did I know it was going to be what it is today? No. But what I did know, and I feel still very strongly about, and somehow I'm even more rooted in this belief, is that in manufacturing in particular, humans still need to be the driving force, and AI is just a tool to help support them. So, to me, Apprentice, this name that we've had for 12 years now, north of a decade, has not only had staying power, but it's become even more suited to what we've been focusing on as of late.
Yeah, and I would say you're talking to the right audience. The Manufacturing Happy Hour audience understands that this industry is very human-centric, regardless of what technology revolution we're going through. So, I know people are nodding their heads relative to what you're saying. Now, Apprentice started with a focus on pharmaceutical manufacturing. Give us a little background, right? Take us through some backstory. Why did you start with that industry?
Well, it was because of my wife, my better half. So, she is in biopharmaceutical manufacturing, so biotech manufacturing. And, 12 years ago, as a naive software engineering husband, she was coming home, because we were living here in Jersey City at the time. So, we're in Jersey City, New Jersey, overlooks Manhattan over my shoulder. And she was, I'm going to politely say it, talking about the things that were going wrong at work, which can often also be said complaining about the things that were going wrong at work. Because at the time, this is a site here in New Jersey, really advanced, really complex biotech manufacturing, running it on a 200-page paper binder. And to me, as a software guy, this was insane, that what I thought was the most cutting-edge manufacturing facility out there, that's creating a cancer therapeutic, a cancer therapeutic, was using paper to drive their process.
And so 12 years ago, maybe naively, not so naively, I don't know all these years later, but we decided to start the company to solve that exact pain. We wanted to bring really great, amazing tools into manufacturing to help the folks on the shop floor, to ultimately, where we got started, get drug product out the door faster. Now granted, 12 years later now, we work with manufacturers across many different industries, but we got started simply because my wife had a real pain, I was a software guy, and we decided to go solve it. And so 12 years later it's very different now, of course, but it started with a very simple premise that today we've taken all the way through to the end.
Yeah, I'm curious to talk about the then and now. If you were to describe what your company's mission was when you started in 2014, what would you say that mission was, and how has it evolved to today, 12 years later?
So there's some things that are different, but there are a lot of things that are exactly the same. And what's exactly the same is that for 12 years we've been doing one thing and one thing only, which is trying to build tools for people in manufacturing. It's super simple. Now granted, over those years the product portfolio has changed dramatically.
We started initially with augmented reality, and if I took you on a tour around our office, you'd see these, I'm looking over my shoulder, but they're these headsets that, going back to 2014, I don't know if you remember, Google Glass had come out.
Mhm.
And we thought, and when I say we, I mean me, I incorrectly thought that a headset was going to be the driver of technology change within manufacturing, because I still believe that you need to work with your two hands, and holding an iPad or interacting with a computer, it's not as ergonomically centric as, say, a headset. So, we started with that as the initial thesis. It then evolved into a manufacturing execution system. That manufacturing execution system evolved to all these new categories that we created called connected manufacturing network. We acquired a company that does manufacturing intelligence, so a lot of analytics and data processing. We built our own hardware edge device to go down to level one. We built a laboratory execution system, an L2 automation system, and now our brand new A1 AI agent that sits above all of that.
But the common thread here is, even though the software has evolved, even though we're now at multiple different levels of the software stack, the kind of classic ISA-95 stack if people are familiar, what has not changed at all in 12 years is this is software made for people who work in manufacturing. And that's the one thing I think that over all these years I'm not only proud of, but continue to be excited by, because the next version of all of this is just going to be the same thing, just with AI. And so, if we stay true to those principles, I think not only will we help a lot of the people in manufacturing, but my hope is that as a company, we'll help these companies transform and guide them into this next transition, which just happens to be AI now.
Yeah. Yeah, let's go back to that diner for a second. ISA-95, I believe, I think most, I shouldn't say most, there's a portion of our audience that knows what that is. How do you describe that as if you're having a beer with someone?
So, in manufacturing, there's a number of different systems that we're all familiar with. And I think the big takeaway is that in manufacturing it's not just one system. First off, there's many. And there's different levels of systems. You have at your top, it's called level four, it's your ERP. We're all familiar with this, whether it's SAP, NetSuite, etc. Then you have your quality management system, also a level four system because it's meant to cut across the enterprise. But then the site-level systems, you have your manufacturing execution system, which has been our bread and butter for north of a decade now. And below your manufacturing execution system, you have your control systems, your PLCs, your SCADA, kind of your automation stack. And below all of that, you get down to actual sensor-level data. And all of this is categorized as a hierarchy, a four-level hierarchy, that is also then categorized based on timescales, but we won't get into that today.
But the point is, there's a lot of stuff in manufacturing. There's a lot of systems, a lot of data, a lot of interconnectivity. And we've taken the fact that we've seen this for 12 years, and we said, "Wait a second, AI can't just be within one of those systems." Because if it's in one of those systems, it's trapped in a box. It literally has four walls around it. And if we built it into our MES, well, it can't really interface with the DCS. If it's in the DCS, it can't really interface with the ERP. And now your AI is trapped. And because manufacturing is so complex and there's so many systems, it led us to building a new layer that, quite frankly, isn't even in the standard right now. It's a brand new layer that sits above all of those systems. We're kind of playing around with level five. I don't know if that's going too far here. But it's this new level that then connects to all four of those layers, all of the systems, and allows one AI and a series of these agents that are connected to people to perform tasks across each of these subsystems.
Yeah, no, I appreciate you taking the deep dive there. I want to go back to something you talked about a little bit earlier around the origin story of Apprentice. And we're going to get into the A1 launch a little later in this conversation, but let's get to know you a little bit first. We talked about how you saw a problem that you built a company around, but I'm curious, can you share how you personally went from being a software engineer, a mobile app developer, to an entrepreneur? And there's some follow-up questions to this, but let's start there.
So, looking back on these 12 years, it wasn't even just one evolution, it's been many. I went from just a good old-fashioned software engineer wanting to solve problems to then an entrepreneur, where then for the first four years, it was just myself and my co-founder Gary, who leads our sales team still, 12 years later now. It was just the two of us for four years, where not only was I writing every line of code, but I was out with the customers, doing deployments, support, sales, and everything was being driven off my credit card. So, phase A was just very simply getting off the ground.
Then, once we got off the ground, we got our first group of customers, very exciting, but then we had to build a company. And so, we raised our first round of financing back in 2017, already again many years into the business, almost pretty much early 2018. But then from there, we had to build a company, and that came with its own suite of challenges, because the first group was less than 20 people, give or take. It was a smaller company at the time. COVID hits, and all of a sudden, I won't say exactly which company, but suffice to say we helped manufacture 300 million doses of the COVID vaccine. You can probably figure out who it is. And that's when the company started to grow quite dramatically.
We then raised over 200 million dollars, a lot of capital, and the company scaled again, and I had to go from becoming a software engineer to an entrepreneur to initial manager and leader to now a CEO. And as CEO, I have layers and I have a management team and we have a multi-product portfolio.
And then after that we went through a challenge, a really big challenge, because in 2023 and 2024 the market slowed in life sciences in particular, which was a majority of our revenue at the time. The Inflation Reduction Act had come out and kind of froze the industry, because they didn't know what it meant, where the government can negotiate drugs for the first time. And so it was this really tough moment that created a large strain on the broader segment, not just Apprentice. And so I had to lead through some pretty challenging moments, to say the least. And through all those moments we came out of it and immediately ran into AI, where then for the last 2 years we've been focusing on AI in this journey.
And I feel like right now with A1, the story has gone full circle. I'm writing code again. I am in with my engineers. I am working with them shoulder to shoulder, so if anyone ends up using this product, I've touched it personally. Because where we are now with AI feels like where we were 12 years ago, in the sense of this is new technology, we're learning on the fly, we're working directly with our customers. And for me, not only did I have to transition from engineer to entrepreneur to CEO to wartime CEO to now this new AI thing, but along the way I had to learn manufacturing too.
I didn't come from manufacturing, and so those early days were just me physically in the manufacturing suite with our customers, gowned up head to toe with my safety glasses, sweating profusely, and I just spent literal years with the customers, which is now known as a forward deployed engineer, Palantir calls it this. I just called it learning back then. And so I spent years and years and years learning, but now with AI it's the same thing all over again. I am back with my customers, I'm in the product, because we need to invent this future together. We need to get it right, because it can either have a great benefit or can have a negative impact. I'd really like it to be powerful and positive for our customers and the people, the humans that actually use this thing.
Yeah, a couple questions I'm going to ask from that answer. One of the first ones being, you mentioned how during the pandemic you were able to scale up rapidly and be able to help with the vaccine at the end of the day. And I'm very curious, what advice would you give to the folks listening to this podcast on how to put yourself in a position to be able to capitalize on that type of opportunity when it comes up?
So we have an operational phrase internally called predict and prepare. There's a whole process that we designed internally called PDS, predict, discuss, and solve, but it all comes from this core kind of purpose, which is you look around the corner at all times, because even though working in the business is really hard and there's a lot of things you need to do, you have to spend as much time working on the business. And working on the business means trying to keep a pulse on what's going on in the market, understand what your customers' concerns actually are and how they shift, constantly shift around you.
And so going back to COVID, and interestingly enough, we've now gone through this cycle where we've predicted something, aligned the business to it, and capitalized on it about four, almost five times now. And COVID was a big one because back in January of that year, one of my guys on my team, Frank, he does our weekly podcast now for Apprentice, he sent out an article to Slack saying, "Hey, there's this really scary thing going on in China. Angelo, you should go take a look at this." So, we all got together and we said, "Oh, wow, this could be a real issue here."
And so, before travel was shut down, we built a brand new product, where that product was meant to be in response to travel concerns, because there were rumors at the time. And so, for us, it was allowing our customers to continue to operate and be in production and do so while they're at home.
So, we had one of our products, which was the remote telepresence product. Think of it as you put one of these headsets on, someone's on the shop floor, and someone can be in their bedroom, literally, guiding them through the process. And so, we took that product, which at the time wasn't a big product for us, completely repackaged this thing, created this way in which our customers can use it really quickly and download and get started. And then, as soon as the pandemic hit, we were there to serve our customers. And then, what ended up happening was they ended up not only using it for the COVID vaccine directly, which,
that was challenging in and of itself because we were trying to solve COVID with COVID around us, and it was literally, it was kind of scary, honestly. And so, helping our customers through not only the vaccine, but keeping the global operations open at the time was in many ways an even higher concern because you had a whole global supply chain that was impacted. People couldn't travel, they couldn't go to site, they couldn't help their colleagues, they couldn't troubleshoot. There's all these things that happen when people have to physically go to manufacturing suite. And so, we built out that product to help them with this.
And so, yeah, the headline is the COVID vaccine, but the real star was all of my customers using these tools and helping to keep the drug supply chains open in the middle of COVID. And so, it was really kind of wild to live through. But, it's not the only time we've done that. We've peeked around the corner some big macro, you know, kind of moments, and then we align the whole business around it, and we have these other crazy things we do internally called Formula 45s where we sprint on the whole business for 45 days, but you know, I digress. I won't bore you with the detail.
Well, no, that was an excellent example of how you've navigated a key pivot and how you were prepared to take advantage of it. A different question along the lines of how you've built your business is you mentioned that your co-founder is leading the sales team now. You've been with them for 12 years. I'm very curious what advice you have on finding a co-founder or really building a team around you like that that lasts. What tips would you give, 'cause we have some people that are in small lean companies. We have others that are in large companies, but I think there's some team building and, you know, finding the right folks to surround yourself with advice that's universal here.
Yeah, so even though I mentioned in 2023 and 2024, we went through some tough times, which every business goes through, our tenure is still incredibly long. And for a software business, the fact that our average tenure is coming up on 4 years where people have stayed with us. My leadership team, it's not just my co-founder who's been with me for 12 years. My chief customer officer is coming up on nine. I have team members that are at 10. And we've created, I like to think, you know, a culture that people not only believe in the mission because the mission is we're not just helping create software. Software is one piece of this story, but the software is meant to be used by people who build incredible things and produce incredible products for people around the world. And I think that mission has stayed true for all those years, number one,
but number two, and this is where I've made mistakes, a lot of mistakes around. Think what I've learned through the years is that I need people on my leadership team that complement my weaknesses specifically. So my co-founder for example, I'm a software engineer. I've learned over 12 years how to do enterprise sales and to be with my customers, but I was a pretty awkward software engineer 12 years ago, like a pretty awkward dude. And so I needed someone to complement me and my weakness, which was I did not know how to sell software, period, full stop.
And so he is and continues to be my perfect complement. And over 12 years we've built so much trust between ourselves that it's like, I like to say, like molecular level trust. Like I know exactly how he thinks, he knows how I think. When we get into conflict, we get through resolution really fast because we've just done this so many times now.
And it's not just him. We have people then on the team who all have over many years built up so much trust with each other that then now when we're launching a new product, it's immediate. We can pivot an entire business of all these people in this business overnight, adopt new tools ourselves, change our focus and strategy, work around the clock, and it's just like with zero drama. Because I just think that over time when you find the right team to complement, in my case, all of my weaknesses, then not only do we help force multiply, but we complement each other and through that trust it then can really start to accelerate to a level that it just feels really special when you have people around you who trust each other and can get through good times, but also really, really tough times. And it's the both that to me, I'm very, you know, fortunate through over a decade plus now having a team around me that thinks like that.
Yeah, this next question is a bit of a nuanced pivot question, but also a segue into the last part of our conversation more around agents in the A1 launch, but what advice would you give on navigating this current pivot that many companies are going through, the onset of artificial intelligence, integrating AI agents, etc. into their business? This is a pivot that a lot of folks are going through and quite frankly, you're going through it as well as an organization. So, what advice would you give to the manufacturers out there that are also navigating this change right now?
So, yes, we're building this product that's meant to help our customers become AI native. And we'll talk about that in a second. But to me, it was a bit ridiculous to say, "Hey, Mr. and Mrs. Customer, use this product to become AI native," if we didn't become AI native ourselves internally. And so, for the last six plus months, I'd say, we've been going through a pretty major internal transformation. And so, that internal transformation is at every level. It's not just software. It's operations, too.
So, what does that mean? So, we manage our business formerly with what we call the quarterly tempo. Every week in the quarter is pre-planned, every team has their respective goals, and every team is aligned to the other teams where, say for example, we have a product release, well then our marketing and our sales team is aligned to then do webinars and our release announcements and all these things that you connect many teams together. What used to take a quarter, though, we've compressed it down to a week, a literal week, where now we have a weekly tempo, where now we do the exact same thing just on a weekly basis.
And so, where did this time compression come from? This time compression came from now leveraging agents internally, which we use our own product internally. We use A1 internally, where yes, A1 is used for manufacturing, but we've built a bunch of these agents and workflows, which we'll talk about in a moment, for our own internal operations. And by doing that, what we found was we were able to even internally compress our cycle time, dramatically lower our manual work that we are doing, and increase the throughput of our people. And once we increase the throughput of our people, we're able to increase the overall cycle time, tempo, takt time, I'm going to use some manufacturing phrases, of the whole business.
And so once we did that, once we implement our own agents, then the next thing we did is we overhauled our systems. And we modernized even the internal systems that we have. Some tools we built ourselves, some tools we upgraded to new pieces of software. But A1 sits on top of all of it. And so we ate our own dog food, and we used our own product. But to help our customers become AI native, we had to transform our same operations, too.
And then on top of all of that, we changed the focus of our people. So our engineers, now we call them super stack engineers. Instead of just being front end, back end, mobile, infrastructure, now they work across all of these business units. And so, you know, at every level of the business we've had to transform because we believe that if you can increase the tempo, change the focus of your people, leverage AI to drive higher throughput, not only can we deliver for our customers better, but we can even tell our customers how we do this because without it, it's, I don't know, it just feels hypocritical. It feels just not credible. But anyway, yeah, that's what we did internally, too.
Yeah, no, great example of your own business becoming AI native because if that's ultimately what you're going to be selling, that's going to be the solution you're providing, it's helpful to be doing that internally to be able to talk about how that's playing out in your own business, as well.
We're going to go back to the diner one more time here. With the A1 launch taking place right now, your agents, I'm very curious. How would you describe this in the context of a casual environment like that? Because I love to go there and then get into some of the details after that.
Yeah, so it comes back to what we mentioned earlier about the existing systems. There's many systems that already exist in manufacturing. A1 sits on top of all of that. And A1 is a brand new standalone, in many ways, category where this A1 agent has a series of what we call sub agents. And these sub agents are meant to be mapped to each individual role within a manufacturing facility. We took a very human-first approach to AI in the sense that these roles are mapped to agents. For example, an operator agent, a quality agent, a process engineering agent, a maintenance agent, an MES agent. We also have supervisor agents, and we have leadership agents, and we have site planning agents. But the whole point is we want these agents to complement and help the people that are in manufacturing today.
And then these agents then automate a series of workflows that right now are either tedious, manual, paperwork, or just even really stinking hard to do when you cut across so many different systems.
And so there's different examples that we give, which is you have the operator agent, which can take a series of paper-driven SOPs. Instead of the operator having to go find exactly which SOP and the section of the SOP to then find how they have to troubleshoot something, instead the agent will go find all that and then put together a step-by-step and clickable troubleshooting guide. So it goes from paper to a literal digital procedure in a couple seconds to help the operator do his or her job faster.
Conversely, the quality agent will go find the exceptions that occur, do initial pass on the investigation, and help the quality leader help understand what's happening in my plant, how do we close out these exceptions faster, and go on from there. And we do the same thing with all the other agents, too.
But you kind of get the theme. The theme of this is there's this new layer that sits above your systems, helps the people automate these workflows that are previously manual, where then the goal over time, which we have these capabilities too, is to then start automating cross-team, cross-functional, cross-system things, which I can talk about. But the real goal is to support the folks on the floor, the leadership on site, and to ultimately help them get their plant, you know, more functional, more operational, and higher throughput.
Yeah, I'd love it if you could take us a little deeper on that because I think those were two really clear examples on the use cases around what an operations agent looks like, what a quality agent looks like. But let's talk about a potential cross-role scenario.
Yeah, so to kind of zoom out just a touch and then we'll go deep again, is every team has a different maturity model. You know, we're not saying that, hey, every team needs to go and do these workflows I'm about to describe. I just want to kind of caveat all of this by saying, we recognize that there is a maturity curve for digital just like there is for AI. Some of these agents can be used on the floor for some basic kind of capabilities, but then for some teams who are ready for some of these more advanced workflows, well, it can do that, too.
For example, A1 is not a chatbot. It is not designed to simply enter a prompt, get a response. Instead, it's meant to respond to manufacturing events, whether those events are alarms, MQTT messages coming off of your IoT, whether it's material flow movement, whether it's kind of shift-to-shift movement, whether it's work cell-to-work cell movement. All these different events happen throughout the day in manufacturing.
So, you can create these agents, for example, an alarm triage agent, which comes out of the box, where then it can respond to any alarm in the manufacturing suite. It can then go investigate that alarm on its own. It can pull data from multiple different systems, so it can pull the run, the active run from your MES. It can then look at what's happening in your control system. It can then pull up any events that are being flagged in your IoT kind of publishing system, combine all of it that it learned, put together an analysis, and then message the manufacturing team to say, "Hey, you guys should probably go take a look at this alarm in detail."
Because right now manufacturing teams are drowning in alarms, truly drowning. And what often happens, unfortunately, which is real, is that they don't look at every alarm. And so instead, this agent can do that first pass. It does the first pass on the alarms, it triages, it puts together initial analysis, and it then gives the humans an opportunity to say, "All right, we actually want to go take a look at this. Let's respond to this in more detail, because this is actually going to lead us off spec."
And so the takeaway is that yes, you can start small, and you can start with a role-to-role agent, but you can also then start to automate these trigger-based workflows that are all triggered off of manufacturing event, run a workflow, which I'd love to talk about a little bit more. What does a workflow mean in AI world now? And then from there, be able to cut across many systems, message many people, and get to an output that today would be next to impossible.
Yeah, well, let's go into that next step then. What does a workflow look like in today's AI world?
So, in general, AI is what's classified as like a probabilistic system. You ask a prompt, and you'll get many different answers, even if you ask the same prompt over and over again. In manufacturing, that's bad, because in manufacturing, you want determinism. You want repeatability. You want consistency. And you ultimately want to make sure that this agent is not going off and doing different things every time you use it.
And so what we did is we constrained the AI in two ways. First, we created our own AI model. We fine-tuned a state-of-the-art model and we trained it on 12 years worth of our knowledge, our customer data, our knowledge, our expertise. We fine-tuned our own AI model that we call Apprentice 4.1. Apprentice 4.1 is designed to then output responses that are specific and unique to manufacturing. So instead of this thing giving you a general response, it'll be hyper-specific. Instead of this thing not having any compliance built into it or any specificity, it'll be very thoughtful about the compliance requirements, what it outputs, and what process it's being used for.
And then we took that model and we built A1 around this, but within A1 we have these agents and then we have these workflows, where these workflows will feel familiar to folks in manufacturing. There's a step-by-step workflow that the agent performs. You can predefine it. We give like hundreds worth of these workflows that come out of the box, but the takeaway is that through this defined workflow it says, "Step one, do this. Go triage. Step two, do this. Go to your MES and get this data. Step three, do this. Contact the supervisor on call for this shift."
And it follows this very strict, structured workflow, but it does so with AI. So you don't have to define every precise API endpoint for your connector, every precise data point it needs to go find, every precise analysis it needs to go perform. You just tell it generally what it needs to do, but because it has strict, structured workflows, the repeatability of this thing shoots through the roof. So that now when people are using this in manufacturing, in an environment that craves and needs consistency, the AI then creates these consistent,
repeatable results, and that's how then teams can have confidence and trust that using AI in manufacturing will actually work to their benefit, not have these crazy results that are all over the place, that every time I use it, it's giving me a different outcome.
So, between talking about use cases, as well as how you don't need to write out all of these workflows, I think we're leading to the answer to this question, but I'm going to ask it directly. How is this purpose-built AI for manufacturing specifically? A lot of people are talking about artificial intelligence today and the agents are building. How is this one purpose-built for manufacturing?
Yeah, so we'll start with that model layer again. So, right now, when you use ChatGPT or Claude, you know, this thing is trained on a wide corpus of knowledge, and it's meant to give a wide range of results. We've been using those general models in our product for 2 years now. So, we've built in agents, like I mentioned at the top of the conversation, that are like stuck in a box. They're within our manufacturing execution system. But, those general models were giving us general results.
And so, the first thing that's specific to manufacturing is the model itself. And so, we used, it's a technique, maybe it's a little too technical, but it's called reinforcement learning. Basically, we took our humans that are domain experts in manufacturing, that have decade-plus experience not just in Apprentice, but external to Apprentice, and we trained the model to understand what is expected for an output within manufacturing. So, number one, the model itself understands manufacturing context, data, result sets, process, compliance, and other kind of more nuanced things that you'd expect when you ask this thing and you get an answer. So, number one is just the model itself is designed for manufacturing. That in of itself is differentiated.
But, then you get to the agent, which is also known as like a harness. Again, a little technical, but just bear with me everyone. You know, this is the shell that wraps around the model and can perform all these different actions. The troubleshooting, the artifact generation, all these things, it's the harness. So, we designed that specifically for manufacturing, too. So, what does that mean? This unique workflow engine that is designed to connect to existing manufacturing systems.
In manufacturing, you don't have a bunch of like cloud SaaS apps. You know, if you use like OpenAI and Anthropic, they want you to connect to like all these generic tools like monday.com and these other collaboration tools. But manufacturing systems live at those lower levels I was mentioning earlier. So, we created this on-prem connector that is meant to process the data that happens in those subsystems, understand the data shape, understand the API endpoints, understand the data hierarchy, like I mentioned the ISA-95 data hierarchy, and it can take all of that output and generate what we call these manufacturing artifacts. And these artifacts are things like troubleshooting guides, process flow diagrams, SOPs, trend and reports, brief summaries, like shift summaries.
And so, not only is the model for manufacturing, not only the workflow is manufacturing, not only the connectors and how we're connecting to those subsystems unique to manufacturing, but the output of all that, too, is unique to manufacturing, as well, because the data that we create holistically as an industry is not this generic stuff. It's very precise, and if we're going to use AI in manufacturing, it's got to be precise, otherwise no one's going to trust this thing. And if no one trusts it, it's never going to be adopted, and if it doesn't get adopted, I just believe that we're not going to as a broader industry set be able to, you know, reap the benefits of what I think this technology can positively improve for the folks in manufacturing.
Maybe this is a good question to start wrapping our interview on then. Do you see from your experience in the different industries that you've been interacting with lately, do you see that trust developing in the right direction? What's your take on that?
So, over the last 12 years, the hardest thing that I had to learn was not software. It wasn't even the entrepreneurship or the CEO stuff. It was building trust and credibility with our customers. That was really, really, really hard because we built this incredible, what I thought in my head to be this great tool. It did all these cool things, all this stuff. But in the early days, people didn't trust us. They didn't trust us because we didn't have the credibility to run their operations on this software. And it took over a decade to earn the trust and credibility, to earn the right to run some of the most complex manufacturing processes in the world right now. And that was hard-earned and hard-won. And so, the same thing I think is true of AI. This is not like all of a sudden you can just show up and magically people are going to adopt AI because, you know, everyone's talking about it as a buzzword and their C-suite is clamoring for it and their board is yelling about it. Like, it just doesn't work that way. And I think there's a certain gravity to earning trust and credibility in manufacturing that you just cannot get past.
And so, how do we build that trust and credibility? Number one is we lower the barrier to entry for people to even get started with this. So, our free tier can do literally everything. Everything. Connect to your subsystems. It comes with pre-built agents. You can do all the MQTT triggers from your IoT, literally zero cost, because before you pay us a dollar, I want you to have the trust and confidence that this thing even does what I say it's going to do.
And what we've learned through the years is that in manufacturing, people need to explore for themselves, they need to use this themselves before they even believe a word coming out of my mouth. And so we needed to lower the barrier to entry so low, not like a 30-day free trial, not like a, oh, you know, you log in and it has some limited functionality. No, like the whole agent completely for free where you can use every function that's in this thing. And we did that because I want whoever is using this not only to give us feedback, please let me know, angelo@apprentice.io, but at the same time, I want to hopefully build that bridge of trust.
Because if they initially trust it on a simple use case, upload some SOPs, get the results sent back, oh great, it does the thing I asked it to do. If it can build that basis of trust, well then over time they're going to feel confident enough to introduce it to their team, then their site, maybe their enterprise if we're so fortunate enough. But I just don't think you can get past the gravity in manufacturing that people need to trust this thing. You have to have credibility. It's only after we prove it, not just say words, that people will actually use this thing and hopefully, you know, use it in some things that we believe will be some pretty impactful next future state of manufacturing, because I think what's coming next is going to be pretty, pretty crazy, but, you know, we have to get through this phase first.
Well, you can tell you are a CEO with a software developer background building a product that builds trust. That feels like a real embodiment of that. We're going to have spots to connect with you, apprentice.io, spots to download, or download might be the wrong word, but start the free trial over at the show notes page.
Oh, it's not even a free trial. Like literally you could use this, like just use it. There's no end date. There's no trial end.
There you go. There you go. Thank you for adding that in there. That's probably the best way to describe it, is correcting my misunderstanding of that a little bit. I love it.
Let's see. So, as we wrap up, this is the wild card question. Is there anything you wish I would have asked you that we didn't touch on yet today?
So, I'm forming the early innings of what I think the future of manufacturing is going to look like. And I think, you know, right now it may seem a bit ridiculous what I'm going to say, but bear with me. Because what we saw in life sciences was we have a portion of our customer base that's producing personalized medicine, which means using CRISPR-based technology. And if folks are familiar with CRISPR, for the first time in human history we can edit our own genes. Like literally edit our own genes, not only treat disease, but cure disease. But the manufacturing of this is on a one-person basis. One. Where there is a whole run for one human, one patient. And we've learned a lot through that.
And I think what's going to happen next, and it may seem a bit crazy today, but I really believe that AI is going to be an enabler of this, is that not just medicine will be personalized, but I do think that products in general will start to become hyper-personalized as well on a per individual basis. Because if the AI can help create high-mix manufacturing, which right now, you know, again, doesn't seem real because the equipment was designed in a certain way, the process was designed in a certain way, the changeover between, you know, those different products happens in a certain way. I'm starting to spend some time thinking about what comes next, and I think there's going to be a future in manufacturing where it's not just the same thing over and over again consistently, but it may be something where it's a different thing over and over again consistently.
And so, I don't know, there's just some interesting stuff that I'm thinking about. And also, I have some strong beliefs and questions about humanoid, but that's a whole different thing. I'm not so hot on it, even being a tech guy myself, but it just doesn't make sense in manufacturing at a personal level. But, I digress. But anyway, yeah, that's the only thing. I think, you know, we've just seen some interesting trends at a personalized product level that I think AI will start to enable, you know, over time.
Yeah, I would tend to agree with your takes on, let's say, personalized manufacturing, right? I think we are getting to that hyper-personal space at this stage. And excited to see how it evolves. As that does happen, we'll have to get you back on to explore your end of the interview takes right there. With that, Angelo, I just want to say thank you so much for taking the time to jump on Manufacturing Happy Hour today.
Yeah, Chris, this was a great pleasure. Big thank you for having me on, and yeah, you know, hopefully we can one day meet at Laicos in Jersey City, New Jersey, and not only have a nice wine, but a nice spiedini al aroma.
Spiedini al aroma. I'll leave that for folks to Google later on. I look forward to taking you up on that offer. That's a great final action item for this interview, Angelo. Cheers.
Yeah, pleasure.
Article published