To explain the value prop of Nuvo AI, we chose a fun analogy to the Roman empire to explain how Nuvocargo brings together human freight experts, our proprietary platform NuvoOS, and an army of AI agents to manage North American truckload freight end-to-end. Watch the launch video here!

This Substack post is designed to go one level deeper and explain why we believe our approach of being the full stack AI services provider is a powerful strategy - both for our customers and for our business. I believe the benefits of our strategy are very analogous to what will happen with AI in most other large, complex professional service industries.

In essence, our biggest learning is that the most important thing we’re building isn’t a product feature or a particular AI agent. It’s a full stack AI-native operating model. And after ~24 months of deploying AI in freight operations, we have a lot of conviction that this model is the right one - and some insights into why other approaches are not really working.

The smartest investors in Silicon Valley are arriving at the same conclusion. Sequoia recently published a piece called Services: The New Software arguing that the most defensible AI companies won’t sell software tools - they’ll sell the work itself. As Julien Bek from Sequoia puts it: “If you sell the tool, you’re in a race against the model. If you sell the work, every improvement in the model makes your service faster, cheaper, and harder to compete with.”

Foundation Capital has pushed this further with their System of Agents thesis — the idea that “Service as Software” is only truly realized when you orchestrate a coordinated system of agents driving outcomes end-to-end, not just individual AI tools.

That is exactly what we are building in North American freight. Here’s what we’ve learned doing it.

The alternatives - and why they’re structurally limited

When you look at the freight tech landscape, there are basically four models competing for the future of AI in managing & executing freight:

Pure TMS companies. Great software, but they have a fundamental problem: every customer uses them differently. In brokerage alone, you have the “split model” and the “cradle to grave” model as one example. Three companies can run on the same TMS and operate in radically different ways with completely different business rules, SLAs, and incentive structures. That makes it nearly impossible to build general purpose AI that drives consistent outcomes across your customer base.

Pure AI agent startups. These are layering AI on top of legacy brokers. The demos are incredible — and I mean that sincerely. Watching a voice AI negotiate a carrier down $50-100 on a load is genuinely amazing. But the leap between a nice demo and deploying that technology across thousands of daily transactions in a way that drives true outcomes is extremely hard. You’re at the mercy of change management and company politics. When something goes wrong - and things always go wrong in early AI deployments - the broker blames the AI startup. You’re the unproven 3rd party vendor; they’re the customer. Behind the scenes, a lot of these “AI for brokers” models are struggling to renew contracts and prove positive ROI. We hear constant private conversations about the massive tension between the AI vendors and their broker customers who signed on 1 year ago when discussing renewals - the ROI is hard to prove, and the big monthly spend on these AI vendors is hard to justify. And if AI agent startups decide to go full stack- they will have to decide to compete with their entire existing customer base.

Legacy brokers without proprietary tech or a legacy homegrown TMS. If you don’t build your own modern software and AI, you have zero proprietary advantage. You’re just another customer for the 50+ TMS companies and the AI startups that are all figuring it out. You’re adopting a ton of disconnected tools — TMS, rate visibility, carrier compliance, GPS, document processing — without any of them working together seamlessly. And none of it compounds into a real moat. There are a lot of scaled incumbents touting their $100M+ annual spend on technology in the last 5-10 years. None of that matters either if you haven’t radically readjusted your technology strategy and attracted high quality teammates fluent in AI in the past 24-36 months. Even if you have an incredible AI team, change management and iteration on your technology is highly complex & slow with a large headcount (more on that below).

Why the full-stack AI-native model is different

At Nuvocargo, we are the brokerage, the managed transportation provider, the customs broker, the TMS, and the AI orchestration layer — all at once. To get here, we built an entirely AI-native TMS in early 2024 and iterated for 2 years with an incredibly lean team - to avoid the trap of change management and slow iteration.

That combination has been transformational for Nuvocargo.

First, it has created real discipline on where to invest in AI. We are not paid for tokens, per seat SaaS, or for running a successful pilot. We get paid to deliver loads end-to-end at a competitive price, at a low cost to serve, either as a freight broker, or managed transportation provider. That means every AI investment has to drive real ROI on high-leverage tasks that actually move the needle. We can’t afford to build impressive demos that don’t translate to outcomes. Our P&L keeps us honest.

Second, it’s the only model where AI deployment actually works. Think about what it takes to truly deploy AI in freight operations. You need to control all of these variables simultaneously:

• The service outcomes

• The customer relationship

• The TMS and system design

• The orchestration layer

• The AI agents

• The humans operating everything

If you’re a pure AI startup or a TMS working with a legacy broker, you control maybe two of those six. Iteration is painfully slow. And when you need to change the UI, the business rules, the job descriptions of the people running the system - you can’t, because those aren’t your variables to change.

We own all six. When we realize that our platform now sources 90% of loads autonomously, we can simultaneously update the agent logic, the TMS interface, and the role of our carrier reps - turning them from phone-bangers into agent supervisors who tweak business rules and catch edge cases. That’s a completely different job, and we can actually make that transition happen because we control the whole system.

Why incumbents can’t replicate this

The most common pushback to these claims is: what stops a North American scaled freight incumbent from hiring a team of engineers and doing the same thing?

A few structural answers:

Headcount is a liability, not an asset, in the AI transition. When you can execute freight audit and pay workflows with 5 people instead of 50, the politics required to make that change happen inside a large company are immense. The leader of that organization has to a) embrace and learn the new AI tools, b) realize their headcount is bloated by 10x, c) actually have the incentive to make bold changes when headcount is a proxy for your influence, budget and importance in most large orgs, and d) take that kind of risk when they’re reporting to a public market with quarterly earnings. These transitions are not smooth. And the people whose jobs are at stake will not make it easy.

Even if they try to outsource AI to third parties - that’s not a proprietary advantage. If a startup builds your AI for you, you’ve learned nothing, you’ve built nothing, and the moment that startup raises prices or pivots, you have no leverage. You are not compounding AI learnings into a sustainable competitive advantage; you are just helping train the vendors who are also making your competitors smarter on AI. And in freight specifically, if technology isn’t already in your DNA, you will not attract the AI talent required to drive true and lasting differentiation.

The TMS problem. Most scaled legacy brokers run on old-school third-party TMS systems that are hard to iterate on. They can’t change the UI or the system design in response to what they’re learning about AI deployment. They’re customers, not builders.

The flywheel - and why timing matters

My biggest learning in the past 12 months as AI has gotten exponentially better is that our model benefits in multiple compounding ways as AI gets smarter.

Every time AI models get better, four things happen for us:

1. Our automation levels go up

2. Our cost to serve goes down

3. Our service quality to customers goes up

4. Our R&D team ships faster

The same model improvements that make our agents more capable also make our engineers and operators more productive. The whole system accelerates together. We already got a taste of that in 2025, and saw it compound only in the last 120 days with the launch of Anthropic and OpenAI’s latest models released in November/December.

If we were just an AI agent startup, faster AI progress would eventually commoditize us. Anyone could spin up a competitive agent. But because we run the agents, the TMS, and the service - there is no such thing as moving too fast. The faster AI progresses, the more we leapfrog the old-school competition and become the obvious, ROI-positive choice for shippers.

The M&A angle — and why AI rollups should pay attention

There’s also a consolidation dimension here that I think is underappreciated.

Because we’ve already spun this flywheel - built the AI-native TMS from scratch, trained the agents on real freight data, rebuilt workflows from the ground up, re-trained our people to work alongside AI agents - we have a structural head start that would take any new entrant years to replicate.

This is an important distinction that often gets overlooked when analyzing brand new AI rollups. There is a class of investor today looking at freight and thinking: “I’ll find a broker, bolt on some AI, juice EBITDA by 10-20%, and exit.” That’s not what we’ve done. We didn’t slap AI onto a legacy operating model to manufacture a quick return. We rebuilt the entire system from scratch, which is precisely why the flywheel is structurally hard for others to spin. If you just start slapping on AI tools on top of a legacy TMS for a new rollup, you’ll have the same challenges as big incumbents, just at a smaller scale.

All of this has exciting implications for Nuvocargo’s inorganic growth strategy. As our platform pulls away from the market, the calculus for a traditional broker increasingly becomes: you can’t build this yourself, and you can’t buy the time it took us to build it.

The most logical path is to consider selling to Nuvocargo and participate in the equity upside. Our equity becomes a compelling offer to brokers who understand the AI transition is coming for their business model whether they act or not. We become an attractive acquirer - not because we’re the biggest, but because we’re the only ones with a platform that makes a broker meaningfully better the moment they bring their shippers, carriers, and employees onto the platform.

What this looks like for our customers (shippers) today

If you’re a shipper and the market plays out the way I’m describing, I believe the implication for you is simple: the sooner you work with a full-stack AI freight partner, the faster you compound the benefits.

In our experience, the outcomes shippers care about right now are the following:

1. Spend less on freight

2. Improve service levels

3. Simplify your tech stack with less vendors and more accountability

4. Stay ahead on AI

A full stack AI partner like Nuvocargo holds ourselves accountable to all four of those outcomes. Hundreds of our customers have already seen:

• 7–20% freight spend reduction in year one

• 30–50% improvement in OTIF

• Hundreds of thousands of dollars eliminated in annual software spend

• Hundreds of hours of manual work eliminated for their teams

With a partner like Nuvocargo, the value to you as the shipper is only going to accelerate as AI models get better - thanks to the foundational work we have done.

You can get started with Nuvo AI with as little as ten truckload shipments. No long implementations. No rip and replace. We can help you see positive ROI and the power of AI in your supply chain in weeks.

If you’re a shipper or work closely with one and want to see NuvoOS in action, book a demo with our team HERE. If you just want some entertaining content about freight, AI, and the Roman empire - watch our launch video HERE!

If you’re an investor or operator thinking about the future of AI in professional services and freight or AI enabled rollups, I’d love to get your reactions to this post. Reply to this post or reach out directly.

While there are many answers to that question, today’s post is about one of the most important pillars for embracing AI in 2025: fostering a culture of builders.

Every day at Nuvocargo we speak with shippers, manufacturers, and even private equity firms that own shippers. They all ask us similar questions:

• How can I build a culture where all my teams embrace AI?

• How can I deploy AI with real ROI?

• Should I aim to build AI internally or partner with AI startups & 3rd parties?

• How do you approach budget and experimentation with AI vendors?

The only constant in technology is change. Today with AI, the pace of change is faster than ever before. The only real way to keep up and adapt is having a culture that embraces change instead of getting steamrolled by it.

In this post I’m sharing very specific things we’ve done at Nuvocargo to build a culture that has allowed us to embrace AI with many examples of quantifiable impact and positive ROI.

We will cover:

• Should you build on LLMs yourself or buy AI tools from 3rd parties?

• What is a “culture of builders?”

• Talent required to embrace AI

• Why small teams with a lot of autonomy & speed win

• How to evaluate and partner with AI startups

• Budgets for experimentation with AI

• Frequent company-wide demos to show what’s possible

• Executives: spend time learning & thinking about the future

The post is meant to be helpful to shippers, manufacturers, or private equity firms that own shippers; it’s also relevant for startups looking to reinvent service businesses with AI.

Build internally or buy from 3rd parties?

One of the first questions we get from companies trying to embrace AI is whether or not to build on top of LLMs themselves or rely on startups and other AI vendors.

Since Nuvocargo sits at the intersection of retail/CPG/manufacturing companies and Silicon Valley startups, we have a unique lens into the pros & cons of these approaches depending on what kind of business you are.

Over time, we have developed strong conviction that if you are a company where technology is not your primary differentiator or product offering, then you should heavily bias towards partnering with 3rd parties for AI adoption. This is especially true if you are just beginning to explore how AI can impact your business.

Building technology is complex, expensive, and requires a lot of things to go right for you to see real ROI. Its especially hard if technology is not already in your company’s core DNA.

The main reason we advocate for partnering with third parties is that we are living through an unprecedented boom in new AI startups and established technology companies layering AI onto their core products.

These are technical teams building on the cutting edge, moving super fast, and aggressively seeking traction with customers to validate their new products & features.

If you are effective at evaluating and partnering with the right vendors, you will learn very fast with a much smaller investment and with a lot less risk.

If you position yourself as a helpful business to startups or technology companies (more on how to do this below), you can learn a lot and get a lot of valuable work done with them as partners in a way that is much cheaper & faster than building yourself.

What is a “culture of builders”?

Simply put, a culture of builders at Nuvocargo means a culture where both technical and nontechnical folks think in systems and are empowered to solve problems for our customers and our internal teams with the latest technology.

As the power of AI and LLMs permeate most knowledge work globally, we believe it’s important to foster this kind of culture regardless of whether you are a tech company or a traditional retail business. LLMs can drive efficiencies across all knowledge work, not only in software products.

At Nuvocargo, being a builder means you can do two things really well:

A) Diagnose issues in our processes and workflows

B) Build prototypes and first versions of AI and software tools to solve said issues

Being a builder at Nuvocargo also means you’re constantly re-evaluating your workflows and processes, always pushing to find the most efficient ways to get work done so we can best serve our customers and our mission.

Lastly, a culture of builders is one where people are pushing themselves to deeply understand the new available technological capabilities.

New technology enables new ways of doing things that were not possible in the past. If you don’t make an effort to deeply understand new technologies, your imagination will be handicapped; you won’t be able to see the future possibilities when re-thinking how to improve products, workflows, and value propositions for customers.

How Nuvocargo fosters a culture of builders

1. Talent required to embrace AI

After many years working to transform a legacy service business with software & AI, we have learned that talent is one of the core tenets to really begin this process.

In short, we’ve built teams of smart generalists who are willing to roll up their sleeves and get extremely close to the business. These folks are willing to temporarily take on the service and operations work themselves, and then zoom out to help map workflows and rethink processes for the future with technology.

In terms of talent profiles, our builders come from a mix of roles you would traditionally think of as BizOps or Product Managers.

This year, we actually merged our BizOps and Product teams which we think is a trend that will happen across tech companies as AI proliferates.

We found that BizOps teammates were willing to get very deep in the weeds on operations and build scrappy solutions, while Product has the complementary skills of translating user painpoints to design, engineering, and data science. Product and R&D folks in general tend to think more in scalable systems with a more technical approach which is crucial to building great software.

The combination of these two skillsets and mindsets is what you also might find in teams called “Applied AI” across other technology companies.

As you’ll see in the visual above, we really think both sides are necessary to innovate in a complex service business with real velocity. One without the other just doesn’t produce results in freight; I believe this is also true for most other trillion dollar service industries where AI is finally having a real impact.

If you are a shipper or private equity firm who is not building software internally, then I would index on finding smart generalists who have the skillsets closer to the BizOps side (non-technical builders).

Make those teammates in charge of getting close to the areas of your business you want to transform with AI. Let them lead evaluating new startups and vendors and give them some autonomy over budgets and picking the tools they are leaning into.

In general - let them run with their ideas with the ability to fail fast.

Even if your company does not build software and have engineers, you can treat these people as your “non technical builders” in the diagram above. Then, through partnerships - treat startups & 3rd parties as your technical builders and let them feed off each other to help transform the business.

2. Small teams with lots of autonomy & speed

One very important learning is that the key to innovation is to have small teams with high caliber talent (as described in the prior point), but also with full autonomy to decide how to rethink or improve a workflow. Its really hard to innovate by democratizing decisions or by consensus/committee.

If you put very talented people who are willing to get in the weeds, understand new technologies & what’s possible, and then give them autonomy to try things fast, you will have more shots on goal and quickly learn what works and what doesn’t. It doesn’t matter if some ideas fail - the failure should lead to learnings that eventually get you on the right path.

It is your responsibility as a leader to create the environment and expectations that allow for failure and experimentation.

Side note: If you are going to try to foster a culture of builders, you also need to try to do the hard work to eliminate busy work and unnecessary process wherever possible.

Here is a good essay from the founder of Y Combinator about this called Maker’s Schedule, Manager’s Schedule. Whoever you are tasking with trying to innovate and understand AI, in my view, should have more of a maker’s schedule rather than a manager’s schedule.

3. Evaluating & partnering with AI vendors & startups

As a YC company, we’ve been fortunate to get inbound outreach from tons of new AI startups since early 2023 when ChatGPT was just a few months old. What started as curiosity for the technology has morphed into a well oiled muscle in the company where we frequently evaluate new startups & vendors to make a well thought out build vs. buy decision.

The process of evaluating AI startups can be an enriching one with lots of learning. These startups are building at the cutting edge. They can show you what’s possible and help you further your technical prowess of what’s possible today that may have not been possible 1-2 years ago.

Startups love to give demo’s, structure low risk pilots, and build custom software for early customers who take a chance on them. Pro tip: startups like to call early customers who give them feedback “design partners.” Try to be a great design partner!

While startups can add a lot of value to your business - one less discussed topic is how terrible most companies are at working with startups. Being a startup ourselves at Nuvocargo, I can tell you that a company who has empathy for startups can get things done much faster at much lower cost than the company that makes every startup jump through a 6-12 month death spiral of a sales cycle.

Here are some tips so that startups want to work with you too:

• Offer real feedback on their product: Position yourself as a company that will help AI startups and vendors validate new products & features with constant feedback. This way the startup gains feedback and you start learning at a lower cost.

• Fast decisions: Be fast to make decisions, don’t drag things out. Startups think in days, not months or quarters. Set a clear point of contact, make timelines as clear as possible, and try to show clarity on your company’s decision making process before engaging too often with a startup. This will help your company infuse some urgency which your executives and management are likely to appreciate as well.

• Clarity on budget: Be clear with your budgets and pricing up front to make sure you find vendors that meet your budgets on top of solving your needs.

• Understand & communicate your use cases: Do the homework internally to study areas where new AI use cases can clearly drive positive ROI so your finance teams will easily approve pilots and experiments. Conveying 5 clear ideas to a startup of where you want AI to make a difference can save everyone months of open-ended abstract discussions.

Big caveat: After analyzing 100+ vendors since 2023, I can tell you that there is often a lot of excitement and dreams in the first mind blowing demo when you see an AI agent do something you felt was impossible to automate.

That feeling of awe is frequently followed by one of disappointment when your teams do more digging into the solution only to realize that a lot more custom work will be required to really make this work. LLMs can do a lot, but they are not a silver bullet and almost all startups know that custom work per customer is required given where the technology is today.

This is why you’ll find most AI startups employ teams called “forward deployed engineers,” which is fancy word for consultants who will do custom work until the technology is deployed in a way that actually works for your business.

On the point of startups possibly underwhelming you - I would advise you to avoid becoming a pessimist and a cynic. It’s easy to point out what doesn’t work today. It’s harder and more rewarding to squint and make things work so you can get an edge on your competition.

Everyone in Silicon Valley right now understands that AI and LLMs are on an exponential curve. They keep getting better and better.

Startups are getting better at deploying these products in a way that is enterprise ready too. While deploying these tools requires more effort today than it might in the future, that is precisely where you can still gain an edge over your competitors before these tools become extremely wide spread in the coming years and decades.

Below is a list of some (not all) startups & tools that we’ve loved working with to help push the envelope on deploying AI.

I believe all shippers, manufacturers, PE firms, and other nontechnical businesses can gain ROI and learn about AI by adopting these tools:

• No-code workflows and integrations: Zapier and Airtable

• AI workflow automation: Gumloop

• AI agents build for nontechnical teammates: N8N

• Software engineering: Cursor, Claude Code, and Codex

• Vibe coding / rapid prototyping: Lovable, Replit, Supabase

• Marketing / content: ChatGPT, Sora

There are many more AI tools we use at Nuvocargo, but this list of vendors has been widely adopted across our builders, both for technical and non-technical folks.

4. Budgets for experimentation with AI

One small thing here is I’ve seen some big companies try to be too prescriptive here for the sake of consistency and process.

Example: “For all AI tools, we only use Gemini because it comes with our G suite subscription.” That means any enthusiastic teammate who wants to try a new tool launched this week is often held prisoner to the “corporate AI policy of only using Gemini.”

While this makes high level sense, beware of stifling innovation and your team’s ability to be nimble in trying new tools as they launch.

Tools are changing very fast and new things are coming to market every single day. Different vendors are known for being stronger at different use cases.

You might want your engineers to lean into Claude Code and Cursor for AI software engineering, and let your marketing teams leverage ChatGPT, Sora, or Gemini. Try to create the space for experimentation in a way that accommodates your own budget.

I like to think of our AI experimentation budget as a way cheaper budget than hiring someone new in house to get a really good feel for what’s possible in the market. The ability to quickly try out new tools with low bureaucracy on approvals is part of what helps foster a culture of builders.

If you’re worried about vendor creep - you can set a reminder every 2-3 months to clean up & opt out of the AI tools that no one seems to be using or getting value from. Push for low cost and short-term pilots wherever you can until you have conviction a certain tool is worth the long-term investment.

5. Frequent company-wide demos to show what’s possible

One of the most powerful things culturally has been allowing our builders to proudly showcase what they’ve been able to prototype and build to the whole company. These can be demos at all hands or demos every 2-3 weeks.

If presented correctly and in an engaging way, demos can help spark imagination across your company and spread ideas that improve your business like wildfire.

Tactical tip: I believe whoever is presenting these demos should be very engaging. They should also have at least 1 slide explaining: Why does this matter? Whose life is improving thanks to this? What is the expected quantifiable impact (time saved, money saved, customer outcome improved)?

Try to make the demos engaging and every week(s) can start molding your culture to one that is excited to truly embrace AI because the results start to feel tangible and consistent.

6. Executives: spend time learning & thinking about the future

This is a big one for me, but I’ve not seen a lot of writing on it. What has worked for me as CEO of Nuvocargo is to really get as deep and close as I can to the new possibilities with AI in freight; and then zoom out on the weekends and try to think “What do these new capabilities mean for our business model and our customers? What is possible now that wasn’t possible before?”

Here are a few conclusions I’ve arrived at:

• Business models: In most of B2B software, we are moving towards outcome based pricing or “selling work,” not SaaS pricing on a per seat basis.

• AI enhances humans: Today its very clear to see how humans will significantly enhance their productivity with AI. I think its less clear to see how AI replaces humans entirely. This is especially true in an industry like freight. Think “my carrier reps will be able to book 5-10x the loads while doing less grunt work” instead of “I don’t need carrier reps anymore.”

• Partnerships matter: In a world where we choose to partner with new startups and AI vendors, it pays to really think critically about who you are partnering with to run the core workflows of your business. Partnering with startups is amazing but it can also be risky. Before we make any startup or 3rd party a core part of our stack - we push ourselves to ask if we believe the vendor will be along for the ride as a great long-term partner.

Hope this helps you take some actionable steps to start fostering a culture of builders to deploy the extremely exciting power of AI!

If you are a shipper or involved with a shipper (such as via a private equity firm) and want to explore how we can help you adopt AI with positive ROI with NuvoOS, our AI native operating system for North America freight), please book a demo HERE.

Highway’s Q2 2025 Fraud Index showed fraud attempts are up 257% year-over-year, with 45,000+ fake carrier identities flagged in a single quarter and cargo theft losses now average $214,000 per incident.

Every fraud incident is terrible for shippers, carriers, brokers, and all other legitimate parties involved who are just trying to move freight and be about their business.

Fraud is rampant in the industry for 3 reasons:

• New carrier risk: Brokers and shippers are always searching for new carriers to reduce their costs and improve service levels; that means every new carrier you add runs the risk of being fraudulent if they dupe your compliance checks.

• Fraudsters are getting smarter: With the accelerated pace of digitization in freight, it’s getting easier to understand ways to “trick” actors in the system to bypass legacy & antiquated compliance checks.

• Dollar sizes are large: While freight brokers and carriers tend to have lower margins, the dollar sizes per transaction are large so fraudsters are highly motivated as they can easily make millions of dollars through fraud.

There are newer tools like Highway that have been gaining a ton of traction in the market to help combat fraud. Highway is a great partner to Nuvocargo and we respect them a lot!

However, fraudsters are getting more sophisticated and keep tweaking their ways to bypass these systems; this means we all need to do a lot more alongside Highway and others to keep helping prevent fraud in the industry.

Why is fraud getting worse?

Most compliance workflows in trucking haven’t changed in decades. They check if a carrier’s MC is active, if insurance is on file, and if authority looks valid. Those particular checks only confirm that a carrier exists, but not whether they’re the best or safest carrier for any particular load.

Across the industry we’ve seen:

• Spoofed dispatch emails that hand freight to imposters before it even moves.

• Insurance policies that look valid but are cancelled the next day.

• Carriers marked “active” bidding on lanes they’re not certified to run, like California reefers without CARB approval.

• Recycled MC numbers bought and sold by fraud rings, resurfacing as supposedly “new” carriers.

Current compliance workflows that both shippers and brokers use today miss all of this. It stamps a simple “pass/fail” while shippers are left exposed to fraud, denied claims, lost capacity, and reputational fallout with their own customers.

The impact is two-fold:

• False negatives (fraud risk): bad actors slip through because generic checks don’t catch mismatched insurance, spoofed contacts, or recycled MCs.

• False positives (lost capacity): fully qualified carriers get excluded because the system can’t tell whether a rule actually applies to that specific load.

Nuvocargo’s approach: Bid-level compliance with an AI engine screening for up to 150 signals

Legacy compliance is built around static checkboxes.

Instead, NuvoOS’ is built as an intelligent AI-powered system where agents are using AI & integrations to:

• Find new possible carriers for your open loads

• Get more bids per load (avg 10.6 bids per load)

• Negotiating bids with AI agents via voice and email

• Fraud Prevention Shield: Screening for up to 150 signals for compliance with AI on a per load and per bid basis (where most of the industry is just statically checking a carrier profile)

All of this happens fully autonomously inside NuvoOS for a large subset of our loads today.

Most carrier sales reps at brokerages are doing mountains of rote manual work just to have a list of 2-3 viable bids for any given load and to onboard a couple of carriers for a spot load.

They are manually procuring, negotiating, and then screening for compliance by logging into 7+ different platforms and doing a ton of manual work.

At Nuvocargo, our reps now log into our loadboard only to review:

• a) More bids per open load (avg of 10 bids per open load) sourced automatically through integrations which leads ultimately to more capacity for shippers,

• b) Lower rates as bids have already been negotiated against max buys with AI agents, and

• c) Pass/fail score for compliance: powered by Fraud Prevention Shield screening for up to 150 signals for every bid & for every load. Most of the industry just periodically checks a carrier’s profile on a system like Highway and doesn’t match that against a particular load or bid.

Our compliance engine is powered by AI to review all these signals and has a four-layer system of security that makes it much harder for a fraudster to bypass our checks & balances.

If you want a detailed understanding of how we built this and how you could benefit from Nuvocargo’s AI-powered compliance engine, please book a demo with our team here.

The Results

Our compliance AI engine is running across thousands of shipments. Carriers are vetted at the bid stage, decisions are instant, and onboarding now takes minutes instead of weeks, so shippers gain speed and security without compromising capacity.

The impact so far:

• 0 fraud cases since launch.

• Carriers onboarded in under 3 minutes, turning what used to take days or weeks into near-instant access to vetted capacity.

• Up to 150 compliance signals checked per load, with only the relevant subset applied, from lane-specific certifications to VIN-level insurance validation.

• 4 layers of security: AI reasoning, live data integrations, carrier sales review, and compliance team review.

• Automatic and autonomous refresh of contact graph carrier identity, insurance, and operating status, so approvals stay current and trusted capacity remains truly usable.

If you want a demo of how we’re significantly reducing carrier fraud and accelerating freight procurement in North America with AI, please book a demo of Nuvo OS (our AI native operating system for North America freight).

However, even when documents are correct, the handoffs across teams to have a smooth customs and border experience are broken. Customs visibility isn’t just about having the right document, it’s about how those updates move between different teams (or not) in a timely way to avoid delays at the border. Without that coordination, the equation is incomplete.

The real reason shippers still chase status

A single northbound crossing can generate 30–50 separate emails: invoices from the shipper, pedimentos from the broker, DODAs from SAT, FDA notices, trailer IDs, driver codes, each landing in different inboxes across ops, compliance, traffic, and finance. Some get answered immediately, others sit unread for hours.

Every team then re-keys details into its own spreadsheet or TMS column: ops tracking trailer numbers, compliance logging entry IDs, traffic chasing driver codes, finance recording duties. Each slice is technically managed, but no one has a live, unified view of the shipment.

That’s the reason shippers still chase status: customs visibility isn’t lost because documents are missing, but because information is scattered across silos. Without a shared source of truth, accountability slips and follow-up turns into endless back-and-forth. And that’s why coordinators spend hours every day asking the same basic question: Has this shipment cleared or not?

From inbox chaos to live full customs visibility

We designed Border AI Assistant to meet the problem where it actually happens: the inbox. Instead of asking teams to learn a new system or forcing brokers to change how they send documents, the assistant simply joins the same email threads customers and brokers already use.

Once copied, it securely processes every client-provided customs document; pedimentos, DODAs, invoices, clearance notices and turns them into structured, real-time data. That data flows into a live dashboard that every team can see, delivering unified visibility without integrations or IT projects. Just CC your assistant, and visibility starts moving.

Once looped in, this is what our AI agent does:

• Captures every update straight from email: Whether it’s a broker message, clearance notice, or a shipping document, the assistant processes it the moment it’s received. No downloads, no chasing threads.

• Classifies and organizes the flow: Each update is recognized for what it is (filing, clearance status, compliance event, operational detail) and routed to the right workflow automatically.

• Identifies key details with >95% accuracy: IDs, timestamps, consignee info, and status markers are recognized directly from the document, without manual entry. This eliminates the lag of copy-pasting between trackers.

• Attributes context: Every detail is tied to the correct shipment, broker, and team. No more wondering “who owns this” or “which load this belongs to.”

• Feeds a live dashboard: Instead of static spreadsheets, shippers see a single, real-time view of what’s cleared, what’s pending, and where action is required.

What you can see in real time

• Live Clearance Status: cleared, pending, or delayed shipments across MX and US.

• Customs Filings Dashboard: every filing in one place, with full history.

• Broker Performance Analytics: average processing times, SLA gaps, accountability.

• Duty Analysis: duties broken down by country, customer, SKU.

• Item-Level Visibility: SKUs, values, units tied to each crossing.

• Searchable Filing Archive: past filings instantly searchable by PO, entry, or broker.

The results

By removing the guesswork and manual tracking, Border Assistant frees up time and coordination between logistics and customs with measurable gains:

• 80% cut in manual status-chasing emails

• 45% faster first-response time on clearance issues

• 8–10 hours saved per coordinator each week

• 30% fewer shipments missing SLA targets

• 95% auto-classification accuracy on new filings

Why it matters

One of the most exciting things about the Border Assistant Agent is that it shifts the equation by working where the work already happens so it doesn’t force anyone onto a new system, no integrations, no software rollouts, no IT projects.

It’s rare in logistics to see something this simple unlock so much power and visibility this fast, that’s why the last few weeks of watching it run on real clients have been super exciting.

If you want to see it in action, we just recorded a session that goes deeper into how the system works and what early adopters are seeing: Watch the webinar →

Or if you’d rather skip straight to hands-on, you can request a demo from our team and see your own customs workflows running through Border Assistant.

If there’s one thing you learn fast in cross-border freight it’s that nothing can move to the next step of the process without the right documents: invoices, packing lists, pedimentos, DODAs, certificates of origin, etc.

However, all the processes around documents for freight tend to be very manual and cause a lot of pain, in both domestic US freight and especially in US-MX cross-border freight. Another example of manual document processes is explained in my previous post how we used AI to automate 90% of POD validation.

The pain

One mistake in the US-MX documents process and your truck can be stuck at the border for days or weeks. This problem impacts all sizes of companies: we hear stories every week from small shippers all the way to multibillion dollar corporations of how their shipments are stuck at the US-MX border for months.

The hidden cause is almost always documents and the handoffs related to those documents.

Specifically, the challenges are:

• High-volume: hundreds of loads a day means hundreds or thousands of documents per day.

• Messy: random PDFs buried in email threads, attachments, and unstructured WhatsApp group chats.

• Repetitive + error-prone: thousands of clicks to re-enter SKUs, HTS codes, consignee info.

• High-stakes: one mistake and you’re holding up trucks, missing windows, upsetting customers.

We solved this problem through a combination of building in the right business logic into NuvoOS for border crossing automations alongside a partnership with Extend, an AI startup focused on document processing.

Today we are seeing 97-99% accuracy with nearly zero human involvement through document intake, classification, extraction, and attribution to a shipment.

Why Extend

At Nuvocargo we build our own software and often build on top of LLM ourselves. However, in many cases, it’s smarter to partner with great AI vendors and startups. A recent example is how we integrated with Convoy to give our customers access to 30,000+ well vetted carriers with end to end automation in a matter of weeks. For AI document processing, we felt partnering with our fellow YC startup Extend was a similarly smart decision.

Extend has been pushing the frontier on AI for messy, real-world docs. By pairing Extend’s AI with our orchestration layer in NuvoOS, we achieved highly impactful results and near perfect accuracy much faster than if we had built it ourselves.

How It Works in NuvoOS:

• Ingests documents from Zendesk, email, and WhatsApp, eliminating the need for manual uploads or searching multiple channels.

• Classifies by type: Commercial Invoice, Packing List, Pedimento Layout, Timbrado, DODA, Entry, BOL, POD and instantly routes each file to the right workflow.

• Extracts structured data: Consignee, SKU, invoice number, merchandise description, HTS codes, quantities, unit values, currency, etc., so our teams work from complete, standardized data.

• Links to the correct shipment in NuvoOS automatically, so every document is in the right place for customs clearance.

Our orchestration logic decides when to run each step, handles exceptions, and continuously learns from human feedback.

The Results

• 4-5 hours per week saved per operations rep working on document processing

• 97–99% accuracy across classification, extraction, and attribution

• Tens of thousands of documents processed

• Faster customs brokerage filings, minimal manual touch

Today when our own operations teams or our customers log into NuvoOS, shipments have most of their key documents available, classified, extracted and we have significantly reduced any bottlenecks related to documents throughout the end to end lifecycle of North America truckload freight.

If you want a demo of how we’re processing documents with AI or of Nuvo OS (our AI native operating system for North America freight), please book some time with our team here.

We’re always excited to share any of our learnings and how we are helping North American shippers reduce freight spend, boost customer service levels, and adopt AI across their entire North America supply chain.

Compounded across thousands of shipments per month, this means many thousands of human emails that teams field day in and day out going back & forth to schedule appointments. Tons of errors, delays in responding, and typing information back into your TMS to make sure everyone in the value chain knows when the appointments will happen.

The pain:

Scheduling is high volume, repetitive, error-prone and time-sensitive. And because it’s one of those workflows that touches every shipment, even a small mistake can snowball into delays, missed appointment windows, and failing your customer commitments.

Specifically, this workflow is painful because:

• Results in thousands of manual emails every week.

• Lots of waiting around for replies from facilities.

• Mistakes: someone forgets to confirm, replies get lost, or details don’t get entered correctly.

• Delays: if one appointment falls through, a truck can sit for hours, or completely miss its delivery window.

What we built

Instead of adding headcount or buying another scheduling tool, we built an AI agent in-house to handle the back-and-forth for us. Think of it as a junior operations hire who only schedules freight, works 24/7 in English and Spanish, and never misses an update.

The AI agent reads inbound emails, figures out if they’re a confirmation, a reschedule, or a request for more info, and takes the right next step based on our SOPs. It can lock in the time with the facility, update NuvoOS so everyone stays in the loop, and also knows when not to take action and flag something for human review.

When action is needed, it passes the job to specialized sub-workflows:

• Initial request workflow: Generates and sends the appointment request email using GPT-4o for natural language generation and formatting.

• Reply handling workflow: Uses Gemini 2.0 Flash for classification, paired with GPT-4o for time extraction and slot selection, to confirm the best appointment in English or Spanish.

• Logging workflow: Updates appointment details directly in NuvoOS via API, labels the email thread for reference, and provides a short summary of actions taken. This gives Ops teams visibility without having to review the full email thread.

Before we let it touch live freight, we stress-tested it with thousands of real and synthetic emails. We used Gumloop to batch test multiple LLMs, GPT-4o, Claude 3.5 Sonnet, and Gemini 2.0 Flash, against messy edge cases like multi-stop changes, vague confirmations, and even intentionally misleading responses, to make sure it could pick the right action or know when to call in a human.

Every step the agent takes shows up in a dashboard, so our operations team members can see what it’s about to do and why.

The results:

• The agent handles about 80% of the scheduling process end-to-end with zero human touch. The other 20% still requires escalations & operations team reviews, usually because the facility’s email was unclear, unrelated, or required a judgment call.

• Freed up about a full day of work (~8 hours) per rep every week, which means more time solving real problems for our customers.

• Fewer delays, faster escalation when something goes wrong, and the same accuracy whether it’s the first load of the day or the hundredth.

• Agent works both for cross-border US-MX-CAN shipments which are more complex, and also for intra US freight shipments which are simpler.

• Path to 95%+ automation: Every time the agent flags an exception, we feed the solution back into the model for training and continuous improvement. That's how we move toward high-90% accuracy without losing a human in the loop for the real exceptions that AI cannot yet handle.

This foundation will allow us to extend the same architecture to tons of other workflows like carrier dispatch check-ins, driver info collection, capacity confirmation and bid collection.

If you want a demo of our AI appointment scheduling agent or of Nuvo OS (our AI native operating system for North America freight), please book some time with our team here.

We’re always excited to share any of our learnings and how we are helping North American shippers reduce freight spend, boost customer service levels, and adopt AI across their entire North America supply chain.

Today with AI, integrations are getting significantly easier and faster if you have an AI native engineering team and a proprietary platform that you can integrate with easily - like NuvoOS.

Most legacy 3PLs take anywhere from 3 to 10 months to integrate with partner technology platforms. When Convoy opened its API-only marketplace, we integrated NuvoOS in under four weeks. With the internal tooling we’ve built since, we could do it in about one week today.

Our pace and quality of integrations is accelerating thanks to AI and to the approach we took in building NuvoOS. After 18 months of making NuvoOS AI-native and API-first, every data point is structured, and every workflow is addressable by API. That lets us pair our own LLM workflows with deep integrations, without waiting on a legacy TMS to catch up.

On top of this, we will be releasing our own MCP for NuvoOS soon which will allow LLMs to interface directly with NuvoOS without relying on API integrations and simply using natural language to execute an action or get the data you need.

We’re still iterating on how to make integrations go faster and better, but the broader learning is clear: build the pipes first, and every integration will get easier and compound value. More specifics on Convoy’s integration below.

BTW: Huge shout out to the Convoy team for the recent deal with DAT and for building world class technology that moved our industry forward in a big way! (Shout out to Dan Lewis, Ryan Petersen, Bill Driegert, Dustin Teves, and the rest of the Convoy team).

The Pain: The old way of covering freight

For most brokers, covering an intra-US load is still a human intensive and manual process:

• Posting the load across multiple boards.

• Calling or emailing dispatchers one by one.

• Negotiating rates back and forth.

• Manually re-entering shipment info into the TMS.

• Endless follow-ups for tracking, PODs, and invoices.

This is slow, expensive, and creates inconsistent service. Even "tech-forward" brokers require heavy ops involvement on every single load today. Some startups are helping automate pieces of this puzzle but the vast majority of the industry still runs this way.

What we built: Plugging into Convoy

In Q4 2024, we plugged NuvoOS directly into Convoy’s API-only platform. This gave us direct access to their network of 30,000+ vetted US carriers, many of them owner-operators and small fleets with leaner cost structures with better rates for our shippers.

For our shippers, this was impactful. It instantly expanded their lane coverage and helped them unlock access to a high-quality carrier base that most shippers or brokers take years to build.

How It Works Now in NuvoOS

1. As soon as an intra-US load is created in NuvoOS, it’s instantly posted to Convoy’s marketplace (as well as to our own internal loadboard and other integrated partners for capacity).

2. Thousands of Convoy’s carriers see the load and then Convoy’s machine learning handles the negotiation dynamically with AI/ML and with rates we specify (“max buy”).

3. Once there’s a good fit between the load and a carrier, a match is booked automatically back into NuvoOS. Zero emails, phone calls, or messages needed.

4. End-to-end tracking flows directly from the driver’s app into NuvoOS, with every milestone time-stamped for full visibility every 5 minutes.

5. PODs and invoices flow back automatically and kick off the same AI-powered validation process I wrote about last week. All without a single manual touch.

This only works for a portion of our loads today, but the share of no-touch loads is growing. As NuvoOS gets more powerful and as LLMs get smarter, we are seeing a clear path to automate most workflows for all types of North America truckload freight (domestic US, US-MX-CAN cross border, FTL, LTL, dryvan, reefer, multi-stop, specialized, etc.).

However, it has required a really powerful foundation to make all this possible.

How we integrated 4x faster than competitors

Most brokers run on decades-old TMS systems. They can’t speak to a modern API without heavy middleware and months of custom development. Every workflow from booking and tracking to invoicing, has to be stitched together manually.

NuvoOS was built API-first from day one, so every shipment, milestone, document, and rate is already structured and ready to connect. This meant we could plug Convoy’s workflows straight into ours with almost no new work.

We don’t rely on a legacy vendor to build workflows for us. We do it ourselves and even heard from the team at Convoy that we were their most technical partner in comparison to other 3PLs at the time. This is another one of countless reasons we feel there’s a huge advantage to building your own proprietary technology.

The results:

• Access to 30,000+ vetted US carriers.

• World class fraud prevention technology for intra US owner operators

• More competitive rates, often 5-10% below traditional spot prices

• End-to-end automation & visibility on every load directly integrated into Nuvo OS

• Lower cost-to-serve, letting us scale efficiently and pass on savings to our shippers

• All of this built in less than 1 month (and would take ~1 week today)

If you want a demo of our Convoy integration or of Nuvo OS (our AI native operating system for North America freight), please book some time with our team here.

We’re always excited to share any of our learnings and how we are helping North American shippers reduce freight spend, boost customer service levels, and adopt AI across their entire North America supply chain.

It sounds simple, but today the process for validating a POD before executing a payment is one of the most broken, manual workflows in freight and until recently, we were handling it the same way most logistics companies do:

• A carrier would send a blurry photo or scanned PDF.

• Our ops team had to track it down, upload it manually into the system, interpret what it said, cross-check each line against shipment details, and

• Only then mark the load “invoice ready” so a payment could get executed

This was happening across thousands of shipments per month. It was slow, inconsistent, error-prone, and wasted countless hours for our ops teams every week.

The pain

Specifically we were seeing these painpoints:

• Dozens of hours per week lost to parsing POD emails and matching fields manually

• High error rates and missed data points, depending on carrier and doc format

• 2-3-day delays to reach “Ready to Invoice” impacting lead times for working capital

• Zero structured data to track performance, benchmark vendors, or resolve disputes quickly

What we built

We didn’t plug in a generic OCR tool or document parser.

Instead, over several months, our R&D team ran hundreds of experiments combining multiple vision-language models (VLMs), custom prompt architectures, and proprietary datasets. We built dynamic validation logic based on specific customer needs and continuously retrained the system using patterns from human-rejected edge cases.

We ran side-by-side experiments on real PODs and found Gemini 2.0 Flash consistently outperformed other models on visual parsing, especially with hard-to-read scans and lower-resolution docs. GPT-4.1, on the other hand, was more reliable for text validation and generating explainable outputs. So we combined both.

What we came up with was a fully integrated three-module pipeline inside NuvoOS: Reception → Validation → Action.

1. Smart intake

Emails are sent to a dedicated inbox. From there, GPT-4o extracts shipment numbers, sender names, contact info, and identifies associated loads. The system routes single and multi-load documents automatically, pulls metadata from NuvoOS, and archives everything in Google Drive and Airtable.

This removes the need for email triage and data entry, which used to make up nearly all of the manual effort in processing incoming PODs.

2. Intelligent Validation
We use Gemini 2.0 Flash to extract structured data from POD documents, then apply GPT-4.1 Mini to validate each field against expected values. Every pass/fail comes with a clear, explainable output.

The system creates a fully auditable trail, flags anomalies without slowing down the flow, and gives our teams complete visibility for QA and reporting.

3. Automated Fulfillment
Validated docs are instantly marked “invoice-ready” in NuvoOS and stored appropriately. Exceptions are routed to Slack for review, without interrupting the flow. Human feedback loops directly feed model retraining and continuous improvement.

The results

Today, this system runs live across thousands of shipments per month and the impact is very real:

• 95%+ field-level accuracy

• 90%+ of POD flows are fully automated, end to end

• PODs are now uploaded and approved 3.4× faster, saving 8+ hours of ops time per day and accelerating downstream workflows

• Fewer disputes, cleaner inboxes, and ops capacity redirected to higher-leverage work

If you want a demo of this workflow or of Nuvo OS (our AI native operating system for North America freight), please book some time with our team here.

We’re always excited to share any of our learnings and how we are helping North American shippers reduce freight spend, boost customer service levels, and adopt AI across their entire North America supply chain.

At Nuvocargo, we've been obsessed as well and we've built a TON- and I’m finally going to start sharing our learnings in this new Substack newsletter (subscribe below) and follow along on my LinkedIn as well.

Since launching Nuvo OS 18 months ago, we’ve quietly rebuilt Nuvocargo’s entire tech stack to become AI-native across our 3 products: managed transportation, freight brokerage, and customs.

Nuvocargo is uniquely positioned to deploy AI end to end and capture its full potential because we’re both a logistics company and a tech startup. That means we build directly on LLMs ourselves, but we also partner with many great AI companies & vendors.

We also have our own proprietary TMS and AI workflow engine (Nuvo OS) which was launched in 2024 so we don’t rely on a legacy TMS vendor that is decades old to modify UI’s and build AI native workflows for us - we do that ourselves. And unlike pure SaaS players, we can redesign org charts, processes, and team incentives to unlock the full potential of AI.

We’ve analyzed 100+ AI vendors, deployed many with real ROI, and built our own LLM-powered tools across the entire lifecycle of a truckload (email, WhatsApp, docs, voice, etc).

The Substack newsletter is intended to be short, tactical posts, mostly focused on where we’re seeing real ROI from AI across our platform and customer base including some demos and highlighting which startups & vendors we are having real success with.

As another important point, Nuvocargo grows both organically and via M&A (see here and here) so if you are interested in the AI buyout or Growth Buyout strategy at the intersection of PE and AI, you will find the newsletter interesting too!

Lastly, writing helps me clarify my own thinking and showcase what our R&D team has been shipping. They’ve built so much in the past year that it's been hard to keep up!

Please check it out and subscribe if you are someone thinking about how to deploy AI in the $1T+ North America truckload freight market.