The Automated Factory: The Real Gap Between Workers & Robotics | The Factory Doctor w/ Alex Krutz

Squint founder Devin Bhushan joins Alex Krutz to discuss capturing tribal knowledge, improving throughput, and bringing AI to the factory floor.

Devin Bhushan: American manufacturing is going through a period where there's a wave of folks about to retire, and there's decades of expertise sitting in their heads that's about to walk out the door. I need to figure out where that knowledge lives. It's clearly not in the data. I kept asking around, and the answer I repeatedly got was, “We have to call Bob, who did this six years ago the last time it failed, and ask Bob what he did.”

Alex Krutz: We call that in the industry tribal knowledge.

Devin Bhushan: Exactly.

Alex Krutz: Welcome to The Factory Doctor. It's Alex Krutz, your host. I am honored to have Devin Bhushan here from Squint. Hey, Devin, how are you doing?

Devin Bhushan: Hey, Alex. Thanks for having me on. Doing well. I'm excited for this.

Alex Krutz: I want to tell everybody up front, you are the first person I've had on the show who's going to talk about AI. There are some people out in the world who can't even spell AI, and the fact that I'm going to have somebody on who's started a company doing that is fantastic. Really happy to have you on.

To tell everybody a little about your experience, you're from the Bay Area, studied computer science at Illinois in Champaign, worked at Yahoo, then Splunk, and that led you into your current work at Squint. Let's expand on that. Tell us more about your experience and your company.

Devin Bhushan: First of all, I'm honored to be the first person to talk about AI on the show. Thanks for the invite.

I grew up in the San Francisco Bay Area, in the heart of Silicon Valley, and technology was around me going through school. I really wanted to study computer science, but I also wanted to go to a university with a big sports program. After high school, I went off to the University of Illinois, a Big Ten school. Go Illini! I studied computer science there, got to learn a ton, and loved my time.

The thing that stayed with me through that experience was this love for sports. After university, I got an opportunity to work at Yahoo Sports on the fantasy sports platform. If anyone's played fantasy baseball or fantasy football on Yahoo, it's very likely I wrote a lot of the code you're touching right now.

Alex Krutz: My son is going to join me for the first year this year and be my partner in my league.

Devin Bhushan: That's awesome.

Alex Krutz: You probably wrote that on the Yahoo platform we use. That's great to hear.

Devin Bhushan: On so many levels, it was amazing. To show up to work on Sunday on the first day of football, you are working, because when that first touchdown happens, you have to make sure there are six million people getting push notifications.

It was a great experience for me. I learned a lot about building consumer products at consumer scale. I happened to meet a mentor who was the VP of engineering at the time. About four years in, he left and joined a company called Splunk as their CTO. He pulled me in.

What stood out to me was that this was a really cool opportunity to work on an enterprise software product. Splunk is a large data analytics provider. They work with large Fortune 500 companies all over the world. I think the stat is that 490-something of the Fortune 500 use Splunk.

I had a chance to do something new there. They were trying to break into the industrial space at the time. The plan was, can we figure out how to bring data from machines on the floor to the operator and help them troubleshoot faster?

They had a platform that collected data from machine sensors, such as vibration and temperature. My job was to figure out how to build a platform where the operator on the floor could access that data and get an insight. We built this platform that showed vibration, temperature, and all these things through the camera. You would point your iPad at a machine, and it would identify the machine and show you its vital statistics.

I thought that would be really useful. We all did. Then I talked to our customers and the operators. What I kept hearing from every single person was, “This is great. You've shown me that RPM is zero on this fan. Thanks for that, but I have a work order that says the fan is broken, and I can see it's not moving. That's not actually that useful. Can you tell me, based on the fault code that fired, why the fan is broken? How do I fix it?”

This question came up over and over. For me, the realization was, I need to figure out where that knowledge lives. It's clearly not in the data. I kept asking around, and the answer I repeatedly got was, “We have to call Bob, who did this six years ago the last time it failed, and ask Bob what he did.”

Alex Krutz: We call that in the industry tribal knowledge.

Devin Bhushan: Exactly. This concept of tribal knowledge came up. When I started Squint, it was 2021. If folks are following the arc of AI, this was after some of the initial research had come out around generative AI, but before ChatGPT. The papers had come out, and I was aware of that.

What I was realizing was, if this AI thing is really going to take off, the thing it's based on is context. You have to give the AI context so that it knows things to use to take action for you. I realized my software didn't even have the right context, because the context that's most important is tribal knowledge. Where does that live? It lives in people's heads, in their memories. How are we ever going to get that digitized so AI could do anything with it?

The thinking was, let's start this company that captures this valuable dark data that isn't captured anywhere, pulls it out of people's heads, and documents it. Especially because American manufacturing is going through a period where there's a wave of folks about to retire, and there's decades of expertise sitting in their heads that's about to walk out the door.

It felt like the perfect time and place to do this. I can talk more about what Squint does now, but that was the genesis of this whole thing. That's how I got here.

Alex Krutz: That's great. Before we move on from your company, it's about 70 people. Where are you in your funding cycle and growth pattern? Tell us a little about that. Then I want to get into AI before some other questions about what industry isn't talking about but should be.

Devin Bhushan: I'll walk through the company's status and what the product does briefly so everyone knows. We are a Series B startup based in San Francisco. We're about 70 people. While we're headquartered here, we're distributed because we work with large Fortune 500 manufacturers all over the country and all over the world. Some examples of customers would be PepsiCo, Continental Tire, Michelin, and Schneider Electric.

Alex Krutz: I have my Michelin Man right here from my time last year.

Devin Bhushan: That's awesome. I need one of those.

Alex Krutz: It's super cool. I'll show you a picture sometime at a reception, taking a picture with the big puffy Michelin Man.

Devin Bhushan: That's amazing. I have a picture of myself in front of the mining tires they make. The mining tires are about 40 feet tall.

Alex Krutz: Those are huge.

Devin Bhushan: It's a trip. Anyway, Squint is an industrial AI company. We start by capturing tribal knowledge. The simplest way of putting this would be, I could record a video of you, Alex, performing a task at a machine. Maybe you're the expert. You've worked on this machine for 30 years, and you're doing a changeover. I'd record you. The AI would watch that video, and a few seconds later, it would produce a procedure of exactly what you did.

It's capturing the stuff you only know how to show me but don't necessarily know how to tell someone, or haven't written down. Once we have that, we create this living context layer that is our knowledge base. This context layer is multimodal. You can add PDFs and manuals, connect it to SharePoint or SAP, and pull in old work orders, asset history, and all sorts of things. It starts becoming your factory brain.

Then we build applications on top of this that let you do things from designing curricula and assigning training to providing an agent that helps with lean manufacturing and does time and motion studies, line balancing, factory optimization, and planning.

Alex Krutz: Once we get to our technology section, we'll talk more. I'm interested to hear about CNC machines versus manual machines and that tribal knowledge. When you get to honing, threading, and fasteners, there's very manual setup of a machine. Maybe we'll talk about that in a bit.

I'm really intrigued by the AI piece. What's the one thing you're hearing and seeing that industry isn't talking about but should be, or something that's on the cusp? I'm not looking for secret squirrel stuff, but what's about to surprise us? What should we know when it comes to AI? It feels like something new is happening all the time.

Devin Bhushan: Something that's top of mind for me, that I'm thinking about every night when I'm going to bed, is this large narrative around the future of automation in factories and manufacturing. I think the bulk of that narrative is committed to talking about robotics and autonomous robots in facilities.

My hot take, or the thing no one's talking about, is that we're missing the low-hanging fruit, which is capturing that tribal knowledge and creating a bridge from the world we're in today, with veteran operators who have this hard-fought experience and know exactly how to run a facility, to this future everyone seems to be talking about, with a robot running the whole cell on its own.

I'm wanting to point out that there's a context gap. How are you going to get the way stuff works out of this person who's been doing it for 35 years into this robot that's been trained to pick up objects and handle them with care, but doesn't actually know how to work your facility, run your recipe, or run your production?

Alex Krutz: At my consulting firm, we work with clients to help them mature their processes. We're doing it through manual methods, traditional lean, and continuous improvement. You're absolutely right. Going from a 2D drawing on a manual machine, with somebody who knows the right settings, where to put certain things, the calibration setting, and what to do on the table, tool, or fixture to make sure it doesn't scrap out, there's a ton of tribal knowledge needed for a robot to eventually be successful there.

Devin Bhushan: You nailed it. We're going about solving that same problem, too. We would enable teams like yours to use technology to help speed up some of these processes. How do you capture that tribal knowledge, and then can you apply continuous improvement methodologies and lean? Totally agree.

Alex Krutz: Before we move on, I get asked a lot, “Can't we just put a robot into this operation?” The answer is, maybe at some point, but right now, no. There's a lot to get from A to Z, and there's a lot in the middle. I appreciate you bringing that up and trying to fill that gap, which I think is going to be important for industry.

Let's get into people. I always start with people, product, and process. We then go into strategy and technology. Let's start with your people. You've got 70 on the team. I know, as a small business owner, the challenge of leading a small team. How do you lead them? I imagine it's diverse, dispersed, remote, and sometimes nebulous around AI and creativity. Tell me about how you inspire the team and fix problems.

Devin Bhushan: It's a great question, and it's something I would say I've not cracked. It's ever-evolving. This is a very hard thing to solve, but the way we run Squint is that everything is focused on the outcome we're creating for the customer. What is the customer's problem? What is the outcome we're generating for them?

Typically, that boils down to reducing downtime, scrap, and waste, helping them hire faster and more efficiently, and helping OEE go up. If you're looking at those outcomes, you try to figure out what part of that problem we can solve.

That line of thinking applies to everyone who works at Squint. If you're a software engineer building a feature that's going to perform a time and motion study, what is the outcome the customer is trying to achieve? If I were in their shoes, how could I build this software so it's not in their way and is the fastest path to the solution for them?

If you think about problems that way, you'll overcome these things pretty quickly. We're all trying to win. You don't work at a small startup or small company if you're not in it to change the world or do something different. Otherwise, a lot of these software engineers in San Francisco can go work at Google or Meta and go to yoga at 1 p.m. You're here.

Alex Krutz: That's good. People are where it starts, especially in small businesses and environments like the one you're leading. It's all about the people. Having that goal where everybody is focused on the customer outcome is a really important point. There's probably not a better singular North Star you can put everybody on.

Devin Bhushan: Absolutely. It's the easiest thing for everyone to use as a universal dashboard. I'm not a big fan of sub-dashboards for every team where it's, “If I do my part, I'm good.” Then you zoom out and look at it, and everyone's done their part, but the customer isn't happy. How could this be? We've gotten lost in the sauce if that's the case.

Alex Krutz: That's good to know. Moving to product, you've explained that a little. How do you see your service and product now, and how do you see it in the years to come, especially with all the technology we have?

Devin Bhushan: Taking my own advice from a second ago, I'll start with the customer problem we solve. There's a customer we work with, an HVAC manufacturer in Texas, and they work with large hyperscalers, folks building data centers. This is demand that's come up over the last couple of years.

There's massive demand to build more data centers to power more AI. The downstream ripple effects are insane. Power generation companies are trying to quadruple their power output in a span of three or four years after averaging a 3% increase year over year over the last hundred years. HVAC companies are growing tenfold in revenue year over year. They have more growth than an AI startup in San Francisco.

Alex Krutz: Water cooling and climate control are huge around the temperatures needed for those data centers.

Devin Bhushan: Exactly. How does our product help? We start with that. If you could build more units, you would sell more units. They're supply-constrained. This business is trying to go as fast as it can. The most important thing for them is throughput and an ability to increase the number of lines that can generate these units supplying data centers.

We looked at that and said, we can capture tribal knowledge, but that feels like a nice-to-have at this particular moment. You're not trying to write things down. You're trying to deliver more units because this is a once-in-a-lifetime opportunity for your business to grow tenfold. How do we help you do that?

This context layer I talked about powers agents we build. AI agents are software that can run on the side and take action on your behalf. They automate things that used to be very time-consuming and require high levels of expertise and teams of people, because we can train an AI model to do this.

For this customer, we built a lean manufacturing agent. If they want to stand up 10 of these lines, they need to figure out what they're doing, perform time and motion studies, and do line balancing. They support 400 billion SKUs.

Alex Krutz: Wow.

Devin Bhushan: Four hundred billion permutations, because it's essentially all custom order. If you think about the people on the line, they're saying, “Four hundred billion? I can't memorize this.” Yet they have nothing written down, so they have memorized it.

We built this agent that takes a video of every station on the line. It watches that video and segments it into a gapless timeline of work done, value-added, non-value-added, waste, and the different things happening. It understands what the procedure was. It looks at it and says, “We could swap these things around, or you could take this process out of the larger process and prefabricate it so it was kitted and the whole thing goes a little faster.”

Then you could use this other line to prefabricate because they're standing there doing nothing right now. If your takt time is 10 minutes and theirs is four minutes, you have six minutes of production you could offload. There are things like that it automatically does that, in the past, would have taken a Six Sigma Black Belt six months, timers, clipboards, and spreadsheets. That's an example.

Alex Krutz: Here's a compliment. I never thought in my life, in a manufacturing discussion, I was going to talk with somebody from San Francisco who started at Yahoo in coding and is talking to me about Six Sigma and takt times. It blows my mind how well-versed you are in manufacturing concepts and the needs of your customers' businesses.

That also shows me there are probably a lot of other AI startups that really know what needs to happen and can talk the language companies and customers need to compete and increase throughput. In this environment, that's really what it's about. It's about throughput.

Devin Bhushan: One hundred percent. We have a company value called “Dunk the Three.” I told you I was a big sports fan. No one in the NBA has ever dunked a three, except Michael Jordan in Space Jam, when his arm gets stretched out and he dunks the three. But no one's ever done it.

The reason it's a company value is that we like to look at things that otherwise feel impossible, or not the way they're supposed to be, and do them anyway. Once you do something and it becomes a record, or a record is broken, people are going to break that record. It becomes normal.

I look at our customers, HVAC manufacturers who are going to grow their production tenfold year over year. It looks impossible right now, but once they figure it out, everyone's going to be able to do that. It becomes possible. We try to do that, too. Part of that is, we have to understand manufacturing if we're going to help. Otherwise, we're just in the way.

Alex Krutz: “Dunk the Three.” That's great. I was watching something on X or YouTube with Michael going from the free-throw line in the 1988 dunk contest, where he soared from the free-throw line. Nobody had seen that before. Now you're talking about “Dunk the Three.” I love that company value. Before I move on, do you have any other good nuggets like that? That's a really good one.

Devin Bhushan: That's how we run the company. The way we work is, we come on-site, understand the problem our customers are having, and try to think of it in manufacturing terms, like theory of constraints. We try to find the bottleneck that Squint can help with and figure out what we can do to make that not the bottleneck anymore.

We could build an AI that does time and motion studies. We could do line balancing. What else can we take off your plate? That's typically how we've built this product over the last few years. Now it's a product suite, and there are a bunch of things it does. We're happy to be involved and make an impact.

Alex Krutz: That's great. Again, I'm blown away. I never thought I would hear somebody talk about Goldratt's theory of constraints. Reduce inventory, reduce cost, and increase throughput, which are the three pillars of Goldratt's theory of constraints. The book The Goal.

Devin Bhushan: The Goal. There are 20-something copies in our office right now because everyone has to read it to work at Squint in San Francisco.

Alex Krutz: It's great. I had my team read Good to Great by Jim Collins. The hedgehog and the right seats on the bus. We all have a book that's good for our stuff. The Goal is a good one.

Let's move into process. If you have a quick minute on how you think about lean and process improvement in your business, it would be great to hear about that before we move on to strategy and technology.

Devin Bhushan: I covered a little about how we help with process. In terms of how we work, everything is continuous improvement here at Squint, too. We probably ship three or four updates a week to our platform. Out of the five business days, most days we're updating the platform because we believe heavily in iteration.

Let's get it out there. Let's get customer feedback. Let's address the feedback. Let's keep going. I often get asked, “What's your vision for six months from now? What's the roadmap?” I can tell you what it is right now, but it's going to change tomorrow. Our customers dictate that. I'm not creating a roadmap out of the blue. I'm responding to what we're hearing in the market.

We try to always move really fast. That's our process. Keep going. There's no such thing as a moat. It's just speed.

Alex Krutz: “No such thing as a moat.” I think that might get quoted.

Let's talk technology a little more, and maybe even outside Squint. I'm really interested, as you're my first AI guest. Feel free to expand horizons into technology. I just saw last night or this morning that Elon is doing a Texas fab, and what the Palantirs of the world and others are doing. Any insights around technology, where it's going, and where you fit in would be great to hear.

Devin Bhushan: What I can probably add that's unique is, let's talk about AI models and where we're headed. There are a lot of questions. I hear this a lot from our customers, “How do I decide between giving Claude to my process engineer versus giving them something like Squint?”

The off-the-shelf frontier models are very good at a lot of things. The difference between that and investing in a purpose-driven software platform is the enterprise gap you have to solve, especially in manufacturing. You probably don't want your CAD files, process maps, and value stream maps going into the training loop of a frontier model. You also probably don't want your operators getting advice from Claude that's sourcing that advice from Reddit. That could be problematic.

There's a lot of thinking right now around how to safeguard these frontier models from various angles. For enterprises and manufacturers today, I think the easiest thing is to find a software platform that's closed and private.

If you can take advantage of frontier models through an AI company that gives you a sandboxed workspace, such that the model answers questions for your operators by only accessing your knowledge base, doesn't hit the internet or Reddit accidentally, and eliminates the chances of that leakage, that's probably the right way forward. This is the nuance I'm most frequently asked about.

Alex Krutz: I'm going to ask a really dumb question, so excuse me. When we hear the term open architecture, what does that mean? Is that the frontier, the Claude models that have tentacles out into the worldwide web and different sources?

Devin Bhushan: There are a couple of styles here. There's ChatGPT and Claude that you can use off the shelf. Those are using proprietary, what's called closed-weight models that OpenAI and Anthropic have built. They're general-purpose, and those models are connected to the internet.

When you use them, there's a bunch of knowledge built in because they're very large. That's why they're so expensive to run. If they don't know something, they'll run internet searches and find you information.

There's also this concept of an open-weight model, which isn't a closed-weight model from OpenAI or Anthropic, but is open-sourced. This is a model where the parameters are published online. The open-source community reviews them, making sure they make sense. There's a lot of trust and connectivity of reviewers.

A lot of private software companies take these open-weight models and build software systems around them. The benefit is that they can customize them for their application, so they're cheaper to run. We can do things like train them on an industrial ontology. We understand this is a wrench, a wrench relates to this machine, and this machine has these procedures.

We've trained the model to understand the factory floor. We've given it lean methodology, and it understands theory of constraints. Because of that, it comes out flavored like something purpose-built to run in a factory. Those are the differences. I don't know if I'm answering the question super well.

Alex Krutz: Is your team taking that and helping with a solution for your clients?

Devin Bhushan: Exactly. We took an open-weight model, trained it for manufacturing and industrials, and now have a software system built around it. An app and a website let you interact with it.

I can say, “I want you to take this video, turn it into a procedure, and run a time and motion study.” Or analyze it for whether someone isn't wearing PPE somewhere they should. You can talk to Squint, and through Squint, Squint manipulates the model.

If I'm the end user or operator, I'm able to present it with problems. Squint understands what to do with that and invokes the model at the right time. It abstracts that layer so the operator feels like they're presenting problems and getting solutions.

Alex Krutz: Interesting. I always end talking about industrial policy. With my time in government last year, I had a lot of experience around industrial policy, what it means, and competition with advanced economies around the world, whether they're friendly or not.

What are you seeing on the reshoring front and industrializing America? Are your clients primarily here? What are you seeing from that industrialization standpoint? It would be great to get your perspectives and thoughts around government interaction with industry.

Devin Bhushan: We're super excited about the reindustrialization of America. Even in the last five years since I started the company, so much has changed. The pace has picked up. We talked about the AI supply chain, the demand for energy, data centers, and everything downstream of that.

It's exciting to see so many more midmarket American manufacturers experiencing a surge in demand. It's created a great opportunity for us to meet that demand before it moves offshore. We've got to do that in the next one, two, or three years. If we don't, more of these companies are going to go back offshore because they're not able to buy the HVAC unit they need for their data center they're building in Arkansas or wherever.

As it relates to policy, I think we need to allow innovation to touch manufacturing and industrials with fewer loops in between, as much as we can. Companies like Squint are building AI systems for manufacturers to use. The more hoops we have to jump through to get cutting-edge technology out to the floor, the harder it will be for American manufacturing to meet this demand.

We're trying to deliver this lean manufacturing agent. Let's make sure we can do that four times a week. I think that's important.

Alex Krutz: That's great to hear, and thank you for sharing that. I need to ask, because there are a lot of small and medium-sized businesses that listen to the podcast, what's the smallest you've gone as far as revenue size of clients you've worked with? Somebody's going to ask me, “Do they work with small or medium-sized companies, or just the big ones you stated?”

Devin Bhushan: I'll answer that twofold. The smallest we've gone is probably $100 million in revenue. The reason is that it's hard and expensive for us to find customers. It's easier to find bigger companies. They're easier to access and advertise to.

There's nothing really preventing that. I'd be excited to talk to anyone out there who's trying to apply these techniques and has a problem they're trying to solve. If we can help, we're happy to do it.

Alex Krutz: That's great. If somebody wanted to reach out to you or your team, is there a direct message, internet link, or contact page you could share to guide folks who want to get in touch, talk about your services, and explore more of what we talked about here?

Devin Bhushan: Our website is squint.ai. Anyone can go there and book a demo with someone on our team to learn more or discuss what we discussed. If anyone wants to email me directly, too, it's [email address unclear in supplied transcript]. I'd be happy to hear from you.

Alex Krutz: I think what you're doing, Devin, is fascinating. Even on the consulting end, we're just getting started with what it looks like and working with clients. It's been great to have you on this.

Not only are you my first AI guest, but you're probably now my longest conversation with anybody. I think you now have two records for The Factory Doctor, first AI guest and longest podcast.

Devin Bhushan: That's awesome. Thanks, Alex, for having me.