I Was Half Right

Howdy Folks,

So. That went well.

In February I told you I was "coming back up for air" and planned to start writing this newsletter again. Then I wrote exactly two editions and disappeared for five months.

In my defense, the business had other plans. But there's no version of "I'll be posting weekly" that survives six months of silence, so let's just call it what it is: I whiffed. We're fixing it, and I'll get to what that looks like at the bottom.

First, though, I owe you a scoreboard.

Because in that February edition, Sauron's Gaze, I made a genuinely bold call about AI and freight. Bold calls are cheap when nobody checks them later. So I'm checking it.

Short version: I was half right. The half I got wrong is more interesting than the half I got right.

1. What I actually said in February

For those who missed it, here's the setup.

On February 12 and 13, a micro-cap company called Algorhythm Holdings (ticker RIME) put out a couple of white papers about its SemiCab platform, claiming 70%+ reductions in empty miles and customers scaling freight volume 300–400% without adding ops headcount. Mostly from deployments in India. No third-party audit.

The market lost its mind.

Billions in market value, gone in an afternoon, because a former karaoke-machine company published a PDF.

My take at the time was that this was a canary in a coal mine. I argued the big brokers and shippers had set the speed limit on tech adoption in freight for thirty years, but that AI would be different. The pace would get set by the AI labs (OpenAI, Anthropic, Google, Meta, xAI) rather than by anyone inside our industry. Sauron's eye hasn't turned toward freight yet, I said, but when it does, buckle up.

I still believe that. But six months of tape says I had the mechanism wrong.

2. The scoreboard

Here's what actually happened after the panic.

The brokers recovered, and then the narrative flipped completely.

C.H. Robinson put out a statement the same day pointing at its "Lean AI" program. The same day, which for CHRW is practically a scream. Since 2022 they've put up 40%+ productivity gains, automated millions of shipment tasks, and pulled down cost-to-serve.

Six months later, Evercore ISI named CHRW one of a short list of stocks sturdy enough to win while the rest of the market frets about AI disruption. Their reasoning: Robinson took market share and expanded margin through the longest freight downturn in history, specifically by applying AI.

Read that again. In February, C.H. Robinson was the poster child for what AI was going to kill. By summer, C.H. Robinson was the poster child for what AI rewards.

And the disruptor? Algorhythm is still standing and actually executing better than the cynics expected: roughly $12M ARR at mid-year, and a $9.5M capital infusion in late February bringing total raised to about $19.5M. That's real progress for a company that size. It is also not a threat to a $17 billion broker.

So the trade was wrong in both directions. The incumbent wasn't dead. The disruptor wasn't a dragon.

3. Where I was wrong

Here's the part I got backwards, and I think it matters more than the stock moves.

I framed this as labs versus incumbents, as though the disruption would arrive from outside and wash over the industry. What the last six months actually showed is that the winners are the incumbents who metabolized AI fastest. Not the ones who got replaced by it.

The advantage in freight isn't the model. Anyone can rent the model. Everybody is renting the same three models. The advantage is proprietary data plus the operational willingness to change how the work gets done. And on that scoreboard, a 120-year-old broker with decades of lane history beats a startup with a compelling PDF.

I said "AI Agents will simply bypass the old integration layers entirely." I'll partially walk that back. Agents are very good at bypassing the UI layer. They are still, as of today, quite bad at bypassing the trust layer. That's the part where somebody's $80,000 of freight is on the line and a human wants a name to call. That layer is turning out to be stickier than I gave it credit for.

The gaze is real. But it is not falling on us from the sky. It is being pointed, deliberately, by operators who decided to point it.

4. Where I was right

I'll take my points where I can get them.

I said the industry was predicting AI impact in 5–7 years and that it would land in 1–2. Six months in, that looks correct. Not because a lab shipped a freight product, but because the tooling got cheap and good enough that a mid-sized broker can now build things that used to require a dev team and a budget cycle. I'll show you exactly what we've built at Rocket in a couple of weeks. Some of it is genuinely silly how easy it was.

I also said the market senses the change faster than the industry admits. A one-day, multi-billion-dollar repricing on the strength of an unaudited white paper is not a market that feels calm about this question.

5. Meanwhile, the actual disruption nobody was watching

While the entire industry stared at brokerage stocks, something genuinely strange happened in LTL pricing. And almost nobody wrote about it.

Carriers decoupled price from volume. And they did not all do it the same way.

• Old Dominion grew Q2 revenue 10.4% to $1.554 billion while shipping less freight. Tons per day down 4.1%. Shipments per day down 5.7%. Revenue per hundredweight hit $37.84, up 15.2%.
• Saia went the other direction entirely. Revenue up 17.1% to $956.5 million, shipments per workday up 4.4%, tonnage per workday up 8.4%, operating income up 26% to $125 million.
• ArcBest chased volume with a roughly 4% base rate cut, didn't like the result, and reversed with a 5.9% GRI effective June 22.

Two of the biggest names in LTL ran opposite playbooks inside the same ninety days. One sold price and gave up freight. The other bought freight and gave up price.

Now strip the fuel surcharge out, which is where this gets expensive for you:

• Old Dominion's revenue per hundredweight excluding fuel was $29.71, up 5.5%. So of that headline 15.2%, roughly two thirds is fuel surcharge mechanics rather than base rate.
• Saia's revenue per hundredweight excluding fuel actually fell 2.2%.
• ArcBest's ex-fuel yield fell in the low single digits too, which is precisely why that June GRI happened.

Your carrier's investor deck says pricing discipline. Your invoice says something more complicated, and a large share of what moved is a table almost nobody redlines.

So two questions worth asking before you sign anything in Q4. First: how much of my increase is base rate and how much is the fuel table? Second: which playbook is my carrier actually running? Because Old Dominion closed the quarter at a 70.1 operating ratio and Saia at 86.9. One of those carriers keeps thirty cents of every dollar you hand it. The other keeps thirteen. They are not going to negotiate the same way, and they are not under the same pressure to keep you.

This is the thing about the AI panic that annoys me most. We spent February arguing about whether software would eat the broker, and meanwhile the biggest carrier in LTL quietly ran a masterclass in yield management while its closest competitor ran the exact opposite play. One of those two stories moved your P&L this year, and it wasn't the one about the karaoke company.

6. And the one that hits home delivery hardest

Same pattern in big and bulky, and this one isn't about technology at all.

Armstrong & Associates sizes residential delivery of oversized freight (furniture, appliances, mattresses, grills, saunas, all of it) at $10.63 billion in 2025, reaching $12.34 billion by 2027. That sounds fine until you look at the growth rate: 5.1% a year from here, against 10.6% annually from 2017 through 2025.

The category didn't get disrupted. It got housing-marketed.

Home turnover hit a 30-year low. Only 28 homes per 1,000 changed hands, down 38% from 44 per 1,000 in 2021. People sitting on sub-5% mortgages are not moving into 6–7% mortgages, and people who don't move don't buy sectionals.

Margins have been leaking the entire time. 28.9% in 2022, 28.1% in 2023, 27.9% in 2024, 27.5% in 2025. Not a cliff. A slow bleed, four years running.

Now look at what that does to an actual P&L. J.B. Hunt breaks Final Mile out as a reporting segment, so this is straight from the filing rather than anyone's estimate:

• Revenue down 9.5% to $823.6 million
• Stops down 11.3%, from 4.32 million to 3.83 million
• Operating income down 54.9%, from $60.1 million to $27.1 million

Divide the last two and you get the number the release doesn't print. J.B. Hunt earned $13.91 of operating profit per stop in 2024 and $7.07 in 2025. Halved in a single year, while revenue per stop actually went up 1.9%.

They charged a little more per delivery and kept less than half as much of it.

Meanwhile Armstrong has Ryder's E-commerce and Last Mile business up 9.1% to $983 million, and RXO Last Mile at roughly $1.2 billion. Two big providers in the same category, in the same year, nearly nineteen points of growth apart.

The obvious read is that one of them is simply better at this than the other. I am not convinced that's what happened, and the answer matters enormously if you're picking a carrier right now. That's next week's issue and I'm not going to pre-judge it here.

What I will commit to today is the margin line, because it shows up everywhere in this data. When the category stops growing, a bad route stops being a rounding error. Damage rates, delivery windows that actually hold, and a returns process that doesn't torch the customer relationship are no longer service differentiators. They are the distance between $13.91 and $7.07.

I've written before about how LTL curbside delivery is broken for big and bulky. In a 10% growth market you can paper over that with volume. In a 5% market you can't.

7. I could still be wrong, because...

Standing feature. Here's where this take falls apart:

• Six months is a terrible sample size. I'm grading a five-year prediction on two quarters of tape and calling it a scoreboard. That's a little generous to myself.
• CHRW's "Lean AI" story is partly a narrative asset. Productivity gains of 40% since 2022 sound enormous, but that baseline is a company staffed for the top of a historic freight boom. Some of that is AI. Some of that is headcount reduction with better branding.
• The ex-fuel numbers still carry mix. Old Dominion's 5.5% is real base-rate movement, but weight per shipment and average length of haul moved too and I have not backed those out. Same caveat on Saia's −2.2%. Directionally I trust these. To the decimal I don't.
• I may be talking my own book on the home delivery thesis. "Margin is where execution shows up" is a very convenient conclusion for a guy who runs a company that sells execution in that exact category. Discount accordingly.
• J.B. Hunt and Ryder are not an apples-to-apples comparison. J.B. Hunt reports Final Mile as its own segment, so those numbers come from a filing. Ryder's last mile sits inside a larger segment and Armstrong's $983 million is a research estimate. I am comparing a measured number to an estimated one, and I know it.
• And the big one: the labs still haven't actually turned toward freight. My original call was about what happens when they do. Nothing in the last six months tested that. It only tested what happens while we wait.

What this newsletter is now

Okay. Housekeeping.

Weekly. Actually weekly this time. And rather than whatever caught my attention that week, it rotates on a four-week cycle so you know what you're getting:

1. Home delivery, big & bulky. The industry, not my company. Economics, capacity, damage, the residential problem nobody has solved.
2. AI at Rocket, in the open. What we actually built, what it cost, what broke, and how you'd replicate it. Real prompts, real tools, real failures.
3. Thought leadership / the argument. Where I take a position on something in LTL or home delivery and defend it.
4. Carrier earnings & the market. The quarterly numbers, and in non-earnings weeks, the pricing and capacity data underneath them.

Then it repeats. If a quarter lands or something actually blows up, I'll break the rotation, but that's the shape.

Next week: big and bulky, and specifically that 9.5% versus 9.1% spread between two final-mile providers in the same market. I've got opinions about which of those numbers is skill and which is portfolio.

Thanks for still being subscribed to a newsletter that went dark for five months. That's a better vote of confidence than I earned.

See ya next week, and this time I mean it.

Gabe

What did I get wrong above? Reply and tell me. I'll run the best correction in a future edition, credited.