Warren Buffett famously said, “Be fearful when others are greedy and greedy when others are fearful.”
Perhaps no two words could better define the AI era we find ourselves in than fear and greed.
And while maybe not an investment in the strictest sense, the path companies choose to take might well define what is to remain of their existence.
From my seat as CEO of a parcel intelligence company, there’s plenty to be fearful of.
AI can read a carrier contract. It can scan an invoice for billing errors. It can compare rates across services and make a recommendation. These are things my company does for a living, and six months ago, AI couldn’t do any of them well. Today, it can do most of them competently. Six months from now, I don’t know where the line will be.
That’s the honest truth. AI is a genuine threat to my business model. Not a theoretical one. Not a “someday” one. A real, active, getting-better-every-quarter threat.
And yet, I’ve built my entire product strategy around it.
I should level with you here. I know how this might read: a guy whose business competes with AI telling you that AI can’t do what he can do. I understand the perceived conflict of interest, and I want to address it directly. That is not why I’m writing this. Call any of my customers and they’ll tell you I have a love for problem solving that borders on obsessive. I’m not writing this article to tell you why you need me. I’m writing it to share the challenges I’ve faced while trying to implement AI day in and day out in my own organization. The things that went well, the things that didn’t, and what I’ve learned in the process. I probably can’t convince every reader of that, but I should at least state my claim.
The Two-Front War
Here’s the part that doesn’t get talked about enough: I’m not just competing against what AI can actually do. I’m competing against what people think it can do.
I’ve long come up against customers who think they can do it themselves. AI makes that problem worse. But coming from someone who once tried to remodel his own bathroom, rarely does anyone fully appreciate the complexities of a DIY project.
A VP of Operations reads an article about AI automating supply chain decisions. A CFO sees a demo of a chatbot analyzing shipping data. And the next time their carrier contract comes up for renewal, they think, “Why would I hire someone for this? AI can handle it.”
Sometimes they’re right. Often, they’re not, at least not yet. But the outcome is the same either way: they don’t call. And in my experience, the customer is worse off for it. Not because the AI gave them bad information, but because it gave them incomplete information wrapped in enough confidence to feel like the whole picture.
That’s the danger. AI doesn’t inherently know how surcharges interact with base rates. It doesn’t know how tiers get calculated and when they change. It doesn’t know when, and how much, cost to add for fuel. It must be taught these things through stringent iteration that can only really be gleaned through years of experience. Put it this way: if you really can do it yourself, then you really can do it with AI. But I’ve met a lot of Little Engines That Couldn’t.
AI sounds like it knows. And that’s enough to change a buying decision.
What We Actually Found
Here’s where I have to be honest about what we’ve seen on our side. Not in theory, but in practice, testing AI against the real data we have on our own customers.
It’s not always right. And rarely is it right enough, at least not without significant expert intervention.
I’m a parcel expert. I’ve negotiated 500+ carrier contracts. I know how these agreements work, how carriers price, how surcharges compound in ways that aren’t intuitive. And what we’ve found building AI into our workflow is that we spend more time now than we did before on things like analysis and systems QA. Not less.
Make no mistake: the outcomes are better. Meaningfully better. But that’s because we now have the additional bandwidth that AI provides. AI didn’t replace the work. It expanded what’s possible within the same hours. The analysis is deeper. The benchmarking is more comprehensive. The negotiation strategies are more precisely calibrated. All because AI handles the volume while I handle the judgment.
Researchers at UC Berkeley published a study in Harvard Business Review this year that validated exactly what we experienced. They studied how generative AI changed work habits at a 200-person technology company over eight months. AI didn’t reduce work. It intensified it. Employees worked at a faster pace, took on a broader scope of tasks, and extended work into more hours of the day. As one engineer in the study summarized: “You had thought that maybe, oh, because you could be more productive with AI, then you save some time, you can work less. But then really, you don’t work less. You just work the same amount or even more.”
That’s our experience exactly. The promise might be “easier.” The reality would be “better.” And better is rarely easier.
The Noise Problem
There’s another dynamic at play that makes all of this harder for the companies I work with: noise.
Customers now have an array of people like me knocking even harder on their doors. This perpetuates a problem I’ve personally seen decision-makers face more of over my 17-year career: too many people all saying essentially the same things.
I’ve had one-on-one conversations with executives at peer companies about AI. In one case, we openly discussed its limitations and how, as the exec put it, “it’s not really part of their business.” Less than a week later, I saw that same person was hosting a trade show breakout session about — you guessed it — AI.
I’m not calling anyone out. I get it. The pressure to have an AI story is enormous, regardless of whether you actually have one. But it goes to show how hard it is for decision-makers to sort through the abundance of additional information companies now have access to. Everyone’s saying the same things, using the same buzzwords, making the same promises. The signal-to-noise ratio has never been worse, and AI itself is making it worse by making it trivially easy to produce more noise.
Why Most Companies Aren’t Ready
A recent HBR piece, “To Scale AI Agents Successfully, Think of Them Like Team Members,” crystallized something I’ve been trying to articulate for a while. The authors found that “although many agents today are ready to act, companies are rarely ready to let them.” Their recommendation: stop treating AI as turnkey software that simply needs to be installed. Instead, treat it like a new kind of workforce, one that needs a defined role, a limited scope of authority, trusted sources of truth, and clear escalation rules.
They identified four frictions that consistently slow or derail AI deployment. Every one of them maps directly to what we’ve experienced testing AI against our own customer data.
The context problem. As the authors put it: “Enterprise data is fragmented across systems, duplicated across teams, and often contradictory. People handle this ambiguity by using judgment and experience that AI agents don’t yet have.” In parcel shipping, the “truth” is scattered across carrier agreements (~200-page PDFs), published rate guides, surcharge schedules that change on different cadences, and fuel surcharge indexes that update weekly. A generic AI model has none of this. The difference between “AI-powered” and “AI that works” is almost entirely about the data foundation underneath it.
The control problem. AI systems are probabilistic. The same question can produce different answers on different days. That’s fine when the output is a draft email. It’s a serious problem when the output is a carrier contract strategy. We’ve seen AI confidently report a 72% Ground discount when the actual contracted rate is 68%. Four points might sound small. On a $5 million annual spend, that’s a $200,000 gap between what the AI said and what the carrier will actually honor. You negotiate based on that number, and you’ve lost credibility before you’ve started.
The accountability problem. When AI generates a negotiation strategy and it backfires, who owns that? AI vendors won’t. The chatbot certainly won’t. And the shipper is left holding a counter-offer their carrier rep just rejected because the ask was unrealistic. It might have been “easier” to generate the counter-offer. But the outcome would be worse, because nobody with actual negotiation experience validated whether the ask was realistic. Whether the carrier would respond to that framing. Whether the sequence of concessions was designed to build leverage or accidentally surrender it.
The identity problem. Without a clearly defined role and scope, AI agents do what they’re designed to do: try to be helpful in every direction at once. In my world, that means an AI that tries to be analyst, strategist, copywriter, and negotiator simultaneously, without knowing which hat to wear, when to escalate, or where its expertise ends.
The gap between “AI can do this” and “AI does this well enough to trust with your carrier spend” is where companies like mine either die or differentiate.
Packaging, Not Augmentation
When we started building AI into ShipMint, I made a deliberate choice. The obvious move was augmentation: use AI to make our existing processes faster, cheaper, leaner. Replace analyst hours. Automate reports. The inward-facing efficiency play.
We went a different direction. As Accenture CEO Julie Sweet put it in a recent CNBC interview, “You can’t cut your way to growth.” The biggest focus with AI across the enterprise is still efficiency, but the real value, the compounding value, is in what it lets you build for your customers that you couldn’t build before.
We asked: what do our customers actually need, and how can AI help us deliver that in a cleaner package?
Augmentation is about us: our margins, our headcount, our speed. Packaging is about them: what the customer gets, how they experience it, whether the product solves their problem without creating new ones.
A mid-market shipper spending $2-10 million a year on parcel doesn’t need “AI-powered analytics.” They need to know whether their carrier discount is competitive. They need something that can read their agreement, compare it to what the market actually looks like, and tell them specifically where they’re leaving money on the table.
They need that to happen in minutes, not weeks. They need the output to be actionable. A strategy they can take to their carrier rep, not a dashboard they have to interpret. And they need it packaged in a way that doesn’t require them to hire a parcel optimization team to make sense of it.
That’s what we’re building.
What This Means Going Forward
I’ll say something that might be unpopular: I think most companies in my space are going to get this wrong.
They’ll bolt AI onto their existing offerings, add a chatbot to their platform, put “AI-powered” on the homepage, and call it innovation. Some of them will use it to cut headcount and protect margins. A few of them will build genuinely useful things. And a few more will host trade show breakout sessions about AI capabilities they privately admit they don’t have.
The broader shift here is real. A third HBR article published this month, “How AI Is Threatening Platforms’ Revenue Streams,” documents how AI agents are disintermediating established platforms by making decisions that humans used to make, bypassing the interfaces those platforms were built on. The authors describe a three-stage response: defend, adapt, reinvent. The companies that wait too long to adapt, they argue, “will not just lose competitiveness. They’ll lose the very foundation and economic logic upon which their business models rest.”
That applies to my industry too. The parcel optimization consultancies that treat AI as someone else’s problem, or worse, as a marketing talking point they don’t actually believe in, are going to wake up one morning and find that their customers already left. Not because AI replaced them, but because someone else figured out how to package AI into a product that made the old model feel unnecessary.
The ones that win — and I believe this — will be the ones that understand their customers deeply enough to package AI around the problem. Not around the process. Around the problem.
So, What Does It All Mean?
Here’s my $0.02.
AI is the great equalizer of our time. Large companies that long held the advantage of massive workforces, offshore developers, and enterprise-level infrastructure are now facing the fear of too much bloat and overhead. The very scale that made them dominant is becoming the thing that slows them down.
And in the spirit of the contrarian investment advice that Buffett, a man who’s seen a few market cycles, bestowed on shareholders and that bore his famous fear and greed metaphor: leaner companies that found a niche and stayed there are well-positioned to take off like a rocket.
That’s where I sit. Not because I have all the answers. I clearly don’t. But because I’ve spent 17 years in one niche, solving one set of problems, and I know those problems well enough to build AI that actually solves them rather than just talks about solving them.
Despite all of us who opine, only time will tell who wins.
All I know is spaceships don’t come equipped with rearview mirrors.
