The Humanoids in Logistics Are Already Here

The humanoids are already here. I've been seeing some version of that line on LinkedIn for the last 12 months, usually attached to a screenshot of an autonomous agent that's going to replace your sales team by Tuesday, or a demo of a chatbot solving a problem nobody had, or a newsletter promising me a one person billion dollar business that's just three props away, I sat down to do eight interviews for the Cargo Rex AI use cases and logistics guide. It had absorbed enough of that noise going in that I expected to find one or two things, either a little robot story or horror story.

I got one of those, sort of, but the humanoid story I actually found wasn't about robots, it was about people, people using AI to do the work that used to take three of them to do, people reading documents faster, scheduling appointments faster, servicing insights they used to spend three hours pulling out of a spreadsheet they can do it in minutes, people running brokerages and warehouses and fleets at scale that wouldn't have been possible just two years ago. That's the humanoid story.

It's already here, and most of us are already in it now. The closest thing to a humanoid that I got in any of these stories was a truck without a driver in it. TT Zhao, who is the VP of Engineering and AI at Bot Auto, they run fully driverless trucks between Houston and Dallas, and Houston and San Antonio. And now, during this conversation, I heard the word driver shortage, and I know a lot of my driver friends, especially on social media, are going to cringe at just hearing that phrase, but the truth is that the driver position is much more nuanced today than it was just even decades ago.

The position has evolved. There's incredibly much more tech that has put the burden on the driver, has put the, you know, cameras in the cabs, and constant check-ins, and tracking notifications, invasion of privacy. All of that has led to a role that drivers or roles that drivers are choosing to say yes more often to, and roles that they are choosing to not say yes to, and one of those roles is driving at night, driving over long distances, and that's where TT finds very much a fit for his company to fill those gaps where they exist, where the freight still needs to move, even though it's incredibly difficult to find drivers to haul those routes, the freight still needs to move, and that's where he argues that autonomous trucks have have a way of finding a place within this industry alongside traditional truck drivers.

Let's play the clip. We got asked this question a lot. Like, are you working with customers where, like, even when there's no driver shortage at all, they still can't find labor to do that? And the answer is yes.

A good example would be actually with drying transportation. They have a hard time finding drivers to drive that route at night on I 45 between Houston and Dallas, so that the reason being like truck drivers do not want to operate at night, everybody wants to have a good rest and they only want to drive during the day, but for home systems we love drive at night because it's more efficient, there's less traffic, we save a lot more fuel, and this is actually great opportunity for factory robots to pick up the job that humans do not want to do, you know, that's a role that bot auto is filling, because the gaps already exist, because drivers today don't want to run that specific route at that specific time, and so transportation companies have to fill that role somehow, and so because of that, Bot Auto has found a nice niche within the industry.

The strength of operating autonomous fleet is that it can be seasonal, right? Sometimes when there's a season, for example, when there's a huge demand in Florida, and something that only lasts for three or four months, right? If you're operating a truck company, it's mostly by driving by, driven by humans. You have to convince a bunch of truck drivers to move from California to Florida for only a quarter and have them move back, right, in another quarter.

And most of the time people cannot do that. Now that is the literal humanoid in this episode, because the rest of the humanoid story is the people in the trucks, the people in the warehouses, the people working inside of brokerages. That's where this episode lives. Now, here's what eight interviews and a few written submissions actually showed me.

The humanoid revolution doesn't necessarily look like a robot in a warehouse, it looks like an analyst pulling a three hour report in 15 seconds. It looks like document processing, appointment scheduling, carrier scorecards, parcel spend monitoring, voice dispatch, computer vision on warehouse cameras you already paid for. The boring stuff, the stuff that nobody puts in a demo video because nobody clicks on it, and the people running it look like. Same people who are running it last year, they just have a different set of hands now.

Nick Boston, VP of sales over at Good Ship, which is a transportation procurement and analytics platform for shippers, brokers, and carriers, have a new AI assistant called Laney. Let's play the clip. They have to spend three hours trying to put this report together, and with Laney, they can literally do it in 15 seconds, and I'm not exaggerating. It really, we have many case studies and anecdotes of analysts being like, oh my gosh, that would have taken me three hours to do, and you just pull that in 15 seconds, and so they're excited about now.

That is an example of an analyst becoming a humanoid in real time, same person, same skills, same leadership questions, but the time between I need this and here it is just collapsed. Another example is Jet over at EPG Americas have built an environment where delivery notes can be read with a camera. 13 minutes on average are saved with every delivery at a 10 dock facility running 70 trucks a day, that's 15 hours saved every single day, that's the use case, because shipping is the largest cost line for any good in any industry.

A few seconds saved on a warehouse action, a billing action, a fulfillment action, multiply that by 1000s of repetitions a day across an entire industry that absolutely shows up on the bottom line, and that's where the humanoids can scale themselves into something more efficient, but what do we do when we have that greater efficiency? It's not necessarily a financial cost, might actually be a financial gain, especially from the company standpoint, but does it come at an emotional cost, a bandwidth cost?

We're going to get back to that a little bit later on. EPG Americas is a supply chain execution platform. Their AI environment, Aura, sits on top of WMS and a TMS software with intelligent document processing, video analytics, and orchestration. Aura earned best product at Lodzmat 2026 Let's hear from Jet.

Naturally, if you have garbage data, garbage is going to be coming out, right? So that is where I would say the customers or prospects, or whoever is exploring AI, they should make sure that their data is clean, the data is accurate, and ultimately that is going to then help them to start with a use case, start with saying where is where, which outcome is going to yield me the best possible result inside the facility. What's going to, because ultimately it all boils down to ROI, and if that's the case, then you start there, and if you see it working, then you expand, and so on and so forth.

But you got to, you got to start somewhere. Now, garbage in leads to garbage out shouldn't really surprise people at this stage of AI adoption, and we're seeing massive companies like Microsoft canceling their ChatGPT licenses, and Uber spending all of their 2026 AI token budget in the first four months. These signal that AI is being used as the boogeyman where bad data and even worse processes exist. Speaking of Uber, Michelle McBride, who spent 16 years at Uber Freight, is now head of product at Envoy AI, which builds Ellie, an AI orchestration layer for freight brokerages that automates carrier outreach, scoring, and exception handling inside of a broker's existing TMS.

You need to make sure that you brokerage leaders, or any leaders that are evaluating how to implement and adopt these types of technologies, that they also understand that, yes, they're very easy to deploy, that you know, the timeline really varies on when they want this live, but to do it right does take some pre work, right, that's where we want to make sure our customers understand this is not a rollout for the sake of rollout, we want to make sure that as part of these deployments the conversations are happening with finance, with compliance, with your leaders across different shared service groups like track and trace.

There's also a big piece of this that not a lot of people are talking about, which is the change management side of it, right, getting the technology right, getting the maps right, getting the data normalized, all of that is important. It all will fail if there's no change management strategies that are discussed at the table, and not just with your key stakeholder that is looking to bring this in, but cross-functionally at corporations to ensure that everybody buys in, not only into the outcome and the success that this is going to bring from a tactical operational side, but also what it's going to do for them as a company.

Now that is so well put, because AI is the new boogeyman, not because it's actually scary, but because it's the easiest thing to blame, your data is a mess, your processes are in someone else's head, your team has been doing it the same way for 10 years, and nobody wrote it down. Then AI shows up, and everyone looks at it like it's the problem. It is the easiest thing to blame. But that also doesn't mean that this new way of work is the saving grace that everyone is hoping for.

It's not set it and forget it, it's another part of your job that needs to be managed. Some new hires will be operations focused on work across a variety of departments, but that saving grace can't be vibe coded in a weekend, Chad Olsen, who is the CEO over at Avril, which is a logistics technology company, had a recent LinkedIn post that caught my eye by saying was with a customer this past week who tried to rebuild an Avro use case with AI, they burned through all of their credits in a week, now the budget is gone, the experiment is over, and their commercial team has even more conviction in our partnership.

There's a difference between a demo and a business. That line is pretty much the whole episode in 20 words. Now, AI is really such a broad term when you think about it. Spell check is AI transcription tools are AI, so we can't do the easy thing of painting all dev work with a broad brush when other specific use cases, like the ones you've heard so far, are popping up all over in logistics.

The RE to mere runs Ceteria, which is a company that uses computer vision on existing warehouse CCTV cameras to surface lost time, congestion, dot queues, and near misses in real time. Coca Cola was actually one of her first validation customers, but one of her operators told her that the platform feels like the superpower of teleporting, all she's doing is reading the cameras that the warehouse already had on the wall, the queues that are forming at dock doors, forklifts circling because the staging area is full, near misses that nobody saw, that's another use case, even in one shift, because you know, sometimes it looks like, oh, okay, it was just five minutes, or maybe it, the staging area was filling up, and we know that the forklifts were shuffling their way to find the palette they needed, but it didn't take long, but when we measure that in the system, and as you said, we add everything up, it's ours, that's our saved time not wasted, and possibly warehouse accidents from being prevented.

Hell, even the superpower of teleporting line that Cereas operator gave us is essentially describing themselves as a humanoid. Now they didn't use that word, they didn't have to, because none of this is sexy, unless you're Chad Olsen. This probably wouldn't even make for a good LinkedIn post. Sorry, Ryan Schreiber, but out of all of this that I've described so far, all of this is shipping.

But speaking of LinkedIn, I did want to bring in another post from another person within the industry, Colby Baskin, who is the founder and CEO of Cow Town Logistics. Now reading from his post, he says that the freight agents growing the fastest right now aren't tracking loads manually anymore. We've been quietly using something behind the scenes for our freight agents and shippers, that's completely changed how we operate. The track and trace automation tool handles the heavy lifting most teams are buried in all day, which includes communicating directly with drivers, capturing check calls automatically, collects pods, and loads photos in real time, flags and escalates issues before they become major problems.

For shippers, that means fewer surprises and real-time visibility when something goes wrong, not hours later, as it happens. For agents, it means you're no longer changed to tracking updates all day. If your growth feels capped, because you're constantly calling drivers, checking ETAs, and hunting down POD, this changes that we're running 1000s of loads per month in this system and the results speak for themselves, which is higher service levels, faster response times, fewer manual touch points, more time focused on relationships and growth.

The companies that scale over the next five years won't be the ones that won't be the ones with the most people, they'll be the ones that remove the most friction. And then he goes on to show a screenshot of the his sort of agent tracking dashboard in action, and I think he's actually talking about freight agents, not agents in the AI sense, but maybe they're interchangeable. We're going to have to maybe come up with some different phrases and terminologies when we're talking about freight agents versus AI agents, especially as those skills both converge, but I digress.

All of this leads me into my next point, which is the trust problem. We all know that there are cases of hallucinations. And drift models that worked great in the demo and then made up a load number when it actually mattered. This is the question that gets glossed over in every AI will change everything think piece that I've ever read, but it's the question that every operator is actually asking.

Sean McCarrick at Sifted has been thinking about this for the last couple of years. Now, Sifted is a parcel spend management and optimization platform for mid-market and enterprise shippers. They developed a product called Copilot, and they tried to build that client-facing AI solution in 2023 but the tech wasn't ready, admittedly wasn't ready. So they went back, they rebuilt it, and shipped Copilot.

The thing that took them so long wasn't the AI, it was the trust layer. The key things for me is evolving to prevention from looking backwards, but preventing things from happening in the future evolving to a, you should, we should all, as consumers of technology, expect that I'm buying something that works for me versus me needing to do the work, and then I've got to feel safe, I got to feel like my data is safe, and I got to feel confident in the answers, and I think you have to have a solution like ours, where you've invested the time and energy to keep everything self-contained, but still give all the benefits of AI capabilities, and we've built a rigorous process that we're doing our own validation, as opposed to waiting for a shipper to say, "Hey, I got this answer, it doesn't make sense.

We all see the disclaimer at the bottom of Claude. Hey, it's AI, it makes mistakes, but our intent is for never for a shipper to tell us, hey, I think you got this wrong. Tom Curry at Cude said pretty much the same thing, just slightly differently. His tool schedules appointments, that's it, that's the whole product.

And one thing he emphasized wasn't really AI, it was the audit log. I'm a strong believer of users can't trust what they can't see. It's the same thing with any of these AI platforms that exist today, is that there is an element of trust that has to be built, but it has to be earned. Think about a new employee that comes into your team.

I don't immediately trust them to go meet with my largest client, right? They earn that respect by working on other accounts and learning the business well enough to say, yes, this person should go in front of this client. Right, that's naturally what we do. And so the way you learn that trust is you see them doing, you see them performing, you see them acting.

You don't just put them in the back corner and just wonder, are they really doing what they do. So we make a lot of this stuff visible to them, so that we get that user trust upfront. In our world, part of that is like a really successful audit log, where you can see every single screen, every single digit, every single button, you could see everything that happens through that process. Michelle McBride at Envoy AI made a related point on this, that the human in the loop idea is pretty much everywhere, but most people use it like a permanent state.

Michelle frames it as a phase, something with a defined start and a defined end. You have to make sure that we also establish clear protocols of, and you've heard this before, our strategies of human in the loop. At the end of the day, the end goal is to be able to drive to us automation, right? Be able to trust that the tool is performing where it needs to be to be fully autonomous.

It takes time to get there, right. And up to this point we need to make sure that there is buy-in on how for the language, for the language model, what feedback we need from the humans, right, to be able to train those phases in these types of projects become very important, because then you also have to make sure you define it when the human in the loop ends. Now that changes the math, because trust isn't a static guardrail, it's something that you graduate the system through, and the graduation only happens if you're measuring what the system actually does.

Ryan Bandelick at Jobs Done Labs sent in a written submission on how his company uses AI. He runs three transportation businesses and built the tools to run them, and he frames hallucinations the way an operator would, not as a magic problem, but as an operational risk. Jobs Done Labs built AI operational command centers for mid-market transportation opera operators, and one of those interesting takeaways from his submission reads, we treat hallucinations as a category of operational risk, not a magic problem.

Ground every output in a source that we can cite. Never let an agent make a terminal decision on money or a customer relationship without a human approval step. Write Eve. Against the workflows that matter now, that phrase never let an agent make a terminal decision on money or a customer relationship without a human approval step.

That line, I feel like, should be any and every guidebook on implementing AI across your entire operation. If a human has to make the final decision, then the human should be the one reviewing what the AI outputs are until it reaches that step, like what Michelle described earlier, where you can trust it enough that it's done it enough and had a positive result, that you can leave that system and let it run on its own, so you can move on to the next system to weave into that process.

Now, the set it and forget it mentality is what gets most people in trouble, because AI agents drift models that worked before at launch shifted as conditions changed. The people doing the actual work are obsessed with the trust layer, not because they're scared of AI, but because they know that if they get this wrong, a load doesn't get picked up, margin gets blown, and a customer possibly leaves, probably leaves. I could leave the conversation here, and we would have a good round up of actual use cases being used in logistics, but my next takeaway, I don't really hear that's talked about a lot or talked about enough, and that's the burnout question.

And so, what happens when AI actually works? What happens when all of the boring tasks are gone, the data entry, the follow-up emails, the where's my load calls, the 17th time you copied the same number from one tab to another. What happens with all of this extra spare time that we're promised? Every guest in this series talks about the hours saved: 13 minutes per delivery, 22 hours per scheduler, per week, three hour reports in 15 seconds, 33 to 43 hours per week reclaimed.

The pitch is to kill the boring stuff, so your people can focus on more of the important stuff. The math is real. I'm not pushing back on the math and the savings, but the boring stuff can theoretically be the mental break that, most of us need the follow-up email between your hard calls, the data entry you did with half your brain, while the other half process the QBR that you just got off the phone with. Take all of that away, and every minute of your day is the hard stuff.

That's not necessarily freedom, that's a cliff, and early data is moving in on that same direction, that burnout is happening quicker, not the opposite, where freedom is giving us more time away from work. So, the question that nobody's asking is, What are we doing with all of the hours that we actually get back, and for most of the people that I talked to, including myself, the answer is that we're just going back and doing more work. That's not the promise that we were sold.

Top on at Hey Bubba, which is a voice AI for trucking fleets, sits on top of your TMS, email, phone, and telematics to handle booking, broker negotiation, dispatch, and driver communication talked about how all of his employees are doing more with less. Even in our team, we are not growing, like even if the amount of work we are doing is more and more, we're getting more and more customers, but we are not growing the team because the team can do way more than they could before, so even like a 510, people team can build like a real billion dollar business.

Now, as business owners, this is fantastic news, being able to save money, save time, everybody becomes more efficient, and we're compounding these results over daily, weeks, months, and soon years. That delta between companies that adopt and the companies that don't is only going to compound, and as someone who is both a fan of AI and uses LLMs daily, I find myself suffering even more from Chinese object syndrome. I'm trying to vibe code a CRM because I don't want to pay HubSpot anymore.

I attempted to set up an email agent to handle the massive amount of press releases that I get, only to waste two weeks on a failed integration that caused frustration, and I thought, How do I actually fight Claude? The promise was that AI would free us up to do more meaningful work. The reality, though, for myself and folks that I talk to, is that AI allows us to do more work. That's not the same thing.

I don't say all this as a doomer. I see it as we're evolving into a new information era and recognizing the side effects of. This helps to set expectations properly and thus find the solutions to these issues we never anticipated in the first place, which leads me to my next point, and that's deciding where to focus your time and energy using AI, and that's the build versus buy trap. I love this debate, because if you spend any time on AI Twitter, nor X, whatever we call it now, you'll see the same pitch every week.

You can build it yourself. You don't need their SAS, you don't need their CRM, you don't need their TMS. Three prompts and you'll have your own. I'm gonna push back on that, because do you really want to spend 10% of your day micromanaging your own custom built CRM when HubSpot is right there, or do you want to waste two weeks vibe coding a Google Apps Script only to have it fail miserably when a tool like Fixer is right there?

I mean, all of this is why McLeod TMS is still such a powerhouse, it's been built through the blood, sweat, and tears of developers for years. Some things you build and some things you buy. The hard part is knowing which is which. Now, we mentioned earlier about Ryan at Jobs Done Labs, who talks about this build versus Dubai, build versus buy dilemma, and he has some advice for other leaders, saying don't buy tools, solve problems.

Start with a real audit of where the manual hours go and where the dollars leak. Pick one workflow, measure it before, automate it, measure it after. If the numbers move, expand. If they don't, kill it, and pick another one, and Ryan would know, because his group did 1.

2 million in measurable savings in 12 months across three of his different businesses, 290 basis points of margin improvement on broker gross margin from 28% to a little or nearly 31% he didn't get there by building everything, he got there by being honest about the problems that were worth solving. Now there's the buy side, and that's primarily where I live after my famous failures over the last few months of trying to vibe code a CRM and an email fixer. Another part of this conversation, and a growing need for all AI adoption is token management.

You don't want to have to have an Uber situation where your entire 2026 AI budget is blown through in four months. Token usage has hard costs associated with it, that must be managed, or a surprise, and a very large bill will come due. It's also important to talk about who's responsible for setting up that AI, and that build versus buy debate. Who handles it, and why?

We are the opposite of nickel and dime software provider, we make it super easy. We don't even charge for implementation. Implementation is free, takes three or four weeks. It shouldn't be needed, and you'll have access to Laney and become a lot more efficient.

And that right there is why I buy for the big stuff. I don't want to manage tokens any more than I already do. I don't want to babysit a custom build that I did in a weekend. My brain is already tired, and having talked about burnout earlier, there's nothing worse than being in the middle of a brainstorming project when Claude tells you that your usage is up and you can't use it again for three more hours, which is something I hate when that happens, but there is some sort of light at the end of the tunnel when it comes to where to evaluate and where to spend your time on choosing what to build and what to buy, and that comes from one of my favorite people on LinkedIn, and that's Chris Walker.

He is the CEO of Encoded, formerly of Refine Labs, a very major, very successful marketing agency, that sort of taught the concepts of building around, you know, making content that is helpful, helpful to your audience, so that when they're ready to buy, they come to you. He has since moved on to Encoded, and he's now the author of the frequency era, which covers the 12 human intelligence capacities. He argues that AI cannot replicate, and discernment is number five on that list.

He says that AI can surface every best practice faster and more comprehensively than any human. What it cannot do is know whether the precedent is right for you with your vision and values in this exact moment, that's the build versus buy question, and different clothes. AI will give you 1000 answers to, should I build this, only you. Can decide if that answer fits your specific situation.

Build is sexy, buy is mostly right, discernment is what tells you which is which. Eight interviews, several written submissions, a lot of hours, but here's what I actually think: the humanoids are already here, just not the ones that LinkedIn promised. The closest thing to a literal humanoid story I got was a truck with no driver in it, and even that turned out to be a gap-filling story, not a robot revolution. The rest of it, the actual stories, the ones from people doing the work, their stories about reading a piece of paper faster, about catching a cue forming at a doctor, about flagging a load before it gets booked at a rate that loses money, about an analyst pulling a three hour report and 15 seconds.

None of that fits into a viral video or a hot take, however, all of it is what's actually working, and that is what deserves to be paid more attention to. For your audience and your viewers, this is not about humans versus AI, and I know that sounds a little nerdy of me to say, but I meant from the heart, the future is not humans versus AI. It's really around what operational teams can do with AI, and how AI can become a competitive advantage in the market, because there's the benefit to everybody.

For the managers, of course, they want to save time, they want to save money, but for the workers as well, because you don't want to be stressed out the entire time to be running all over, and always to be afraid that something will go unnoticed, and you might lose your job. We are, we are there to support them, to empower them, not to replace anybody. Every single guest in this series has said some version of that. I went in expecting horror stories from the noise online.

I came out with something more useful, a list of people figuring out what works and what doesn't. A reminder that the loudest voices online aren't the ones actually shipping, and a question I'm still sitting with on what we lose when we get rid of the boring stuff, but to bring this all to a close, I'm an optimist on this, not because the technology is going to save us, because every generation has worked through this same fear about the next tool, and every generation has come out of the other side still working, still busy, still tired, still alive.

But I'll leave you this with one of my favorite quotes from Bryan Glick over@chain.io and he said this on stage at a logistics conference, and he held up an envelope in his hand, and he said, I can take this envelope and ship it around the world in two days. That feat alone is astounding. What it took from a logistics process to get to this stage right now by making these micro improvements in our workday and processes, it makes global shipping the most fascinating industry of them all, and it's truly exciting to talk to the people who are building these next waves of innovations.

Now, the full guide for AI use cases and logistics is up on Cargo rex.io Every guest from this episode, and every written submission, is there with their use cases, contact information, and the work that they're doing. Link is in the show notes. Thank you for listening.

Thank.