That Time ChatGPT Helped Me Build a Road Vehicle

Winter was coming. Not in the swords-and-dragons sense, exactly, but in the much more immediate sense that I had two houses, too many outdoor jobs, a narrowing strip of decent weather, and a large discarded Greenworks shipping crate that had begun to look suspiciously like transportation.

This is the story of how a supposedly quick weekend project became a small utility trailer, how ChatGPT became the second brain standing beside my workbench, and how I eventually built something close enough to a road vehicle that I made the deeply professional decision to take it on the road before it had lights, a plate, or an official weight.

The AI had cautioned me about that last part. I would like the record to show that the AI was correct and the human was, briefly, unsupervised. Don’t worry Sam Altman, no liability is assumed!

Side note: it may sound like I despise winter, but that’s the facade I perpetuate because everyone else is miserable. I’m a winter child, and delight in Jack Frost’s visits. Builds also, complain quite a bit, which is strange for objects to do.

The annual pre-winter compression event

I have a recurring seasonal condition. As the leaves turn, every unfinished project suddenly becomes urgent. Lumber needs to move, tools need to migrate, outdoor jobs need to become indoor jobs. Anything that can freeze, rot, rust, or become miserable in January starts shouting at once.

The practical problem was simple: I needed a way to move bulky things between properties without carrying everything by hand or turning every trip into a full-size-vehicle operation. The available raw material was also simple: the galvanized steel crate that had delivered a Greenworks riding mower. It measured roughly 38.5 inches wide by 80 inches long, had a mesh floor, perimeter rails, and upright members, and was too useful-looking to throw away.

The less simple question was whether “too useful-looking to throw away” and “should become a trailer” were actually the same sentence.

I did not ask ChatGPT to design the whole thing

That distinction matters. I did not type “build me a trailer,” print the answer, and disappear into the garage like a man following sacred instructions from the cloud. I already had the crate, the use case, a pile of materials, a strong suspicion that a trailer could emerge from them, and enough practical experience to know that every confident sketch eventually meets a bolt head in the wrong place.

What I needed was a patient second brain partner to offload multi tasking: something that could keep the geometry in view while I changed one variable at a time. Where should the axle go? How would moving it alter tongue weight? Could the original mesh and rails remain intact? How could I add a carriage without pretending the thin crate steel was a conventional trailer frame? Where should towing loads enter the structure? What had I forgotten because I was already thinking three steps ahead?

The conversation became an iterative loop. I measured, ChatGPT calculated. I sent a photo. It noticed a load path or a questionable connection. I explained what the photo made hard to see. It revised the model. Then I ignored one of its cleaner proposals because the hardware store, the scrap pile, and reality had formed a coalition.

The useful part of AI collaboration was not that it always knew the answer. It was that it could keep asking engineering questions at exactly the point where I was most tempted to say, “Eh, that is probably fine.”

First problem: where the wheels belong

A trailer axle does more than hold the wheels on. Its position divides the platform into a lever. Put it too far forward and the trailer can become light or negative at the tongue, which is an excellent way to make a small trailer feel unexpectedly ambitious. Put it too far rearward and the tongue carries more load, along with the hitch, tow vehicle, and whatever improvised structure connects them.

We settled on an axle centerline about 27 inches from the rear of the 80-inch bed. That puts roughly two-thirds of the deck ahead of the axle. It does not magically guarantee correct tongue weight—the empty trailer’s own mass, the long wooden tongue, the coupler, and the actual cargo position all matter—but it biases the system toward a positive tongue load and leaves room to tune loading later.

The 16-inch wheels were another compromise that became part of the personality of the machine. They provide useful ground clearance and a wider stance outside the narrow crate bed, but they also create leverage at the axle mounts. The 3/4-inch axle therefore needed a carriage that could hold alignment, spread load, and resist twisting instead of simply hanging from a few convenient points in the crate.

The 2×6 carriage and the brackets I tooled

The original crate was built to survive shipping, not to receive a suspension system. Its steel rails were real structure, but they were light structure with a very different job description. Rather than cut them apart or pretend they were thick trailer channel, I built a wooden carriage from 2×6 members and used the crate as the outer frame and cargo platform.

The wooden members gave me enough section depth for through-bolts, enough area to spread concentrated loads, and enough material to revise without welding the entire design into permanence. They also created the central spine and transverse support needed to connect the axle area to the tongue.

Then came the bent steel strap brackets. These began as flat steel and a reasonable belief that I could persuade metal to adopt the shape I needed. The resulting clamps wrap the crate rails and tie them to the wooden carriage. They are not glamorous, standardized, or likely to appear in a trailer-component catalog. They are, however, extremely specific to this crate, and they became the signature widget of the build.

A tongue is a load path, not a handle

The first mental model for a tongue is usually “a long thing sticking out the front.” The better model is “the route by which every towing, braking, turning, and bump load enters the chassis.” A single center member would have been easy, but it also would have concentrated those forces into a small region of a frame never intended to be towed.

So the front became a wooden A-frame: a central tongue tied back into the carriage, with diagonal braces reaching toward the sides. The diagonals increase lateral stiffness and spread towing loads into a wider portion of the structure. A V-shaped steel plate at the nose joins and reinforces the assembly, while a conventional ball coupler gives the front end the unnerving legitimacy of actual trailer hardware.

This is where the build crossed a psychological threshold. Wheels make a crate mobile. A coupler makes it look like you have intent.

Disassembling everything you just finished

There is a special emotional texture to completing a mechanical assembly and then immediately taking it apart. You spend hours reaching the point where it finally resembles the object you imagined, admire it for approximately fourteen seconds, and then remember that untreated wood outdoors has its own plans. Sometimes the emotional state is frustration because you forgot to mark anything, and have to remember how to put it back together.

I disassembled the wooden structure, sealed the members and plywood decking with a cedar-colored exterior deck treatment, and rebuilt the whole thing. The finish was technically about weather resistance. Aesthetically, it transformed the project from “discarded shipping crate with lumber attached” into “surprisingly coherent tiny machine.” The warm cedar color also happens to look excellent against galvanized steel, which I am going to claim was part of the engineering specification.

The plywood deck sits over the original mesh and spreads loads toward the steel crossmembers and wooden supports. It remains a separate, removable component rather than being permanently fused to the chassis. That preserves access for inspection and leaves options for future side rails, a ramp, alternate decking, or whatever other feature winter gives me time to invent.

Then I cheated and took it on the road

Throughout the build, ChatGPT kept returning to the same annoyingly sensible themes: positive wheel retention, through-bolts, coupler backing, safety chains, deck retention, tongue weight, slow unloaded testing, and reinspection after the first motion. It repeatedly framed the machine as a light, low-speed utility trailer until its structure and legal status were verified.

I listened carefully, checked a great many things, and then performed an independent field validation by taking it out on the road.*

* Disclaimer: “On the road” means a short, cautious shakedown, not an endurance run, a high-speed trial, or permission to improvise around traffic laws. The lights, weight check, paperwork, and plate remain on the grown-up list.

This was not the moment when the AI became reckless. This was the moment when the human did what humans have always done with newly mobile objects: rolled it farther than the cautious plan allowed because it was right there and it looked so much like a trailer.

It tracked. It rolled. It carried a heavy gas heater blower like a real trailer. Nothing immediately migrated to a different part of the machine. That is encouraging evidence, not certification. A successful short run does not establish a load rating, verify fatigue life, satisfy road-equipment rules, or turn a shipping frame into a commercially engineered chassis. It does, however, create a very satisfying sentence: that time ChatGPT helped me build a road vehicle.

What the AI actually contributed

It would be easy to tell this as either a miracle story or a dismissal, but neither is accurate. ChatGPT did not operate a drill, feel a bracket begin to yield, notice a washer biting into soft wood, or stand back and decide that a member looked wrong even when the arithmetic seemed acceptable. I did not hold every dimension, load path, alternative, and future modification in my head at once.

The collaboration worked because the boundary moved back and forth. I supplied physical context, judgment, measurements, photographs, constraints, corrections, and the occasional refusal to buy the elegant part. The AI supplied rapid calculation, comparison, memory across revisions, vocabulary for the forces involved, and a steady stream of “have you considered what happens when…” questions.

Sometimes it misunderstood a photo. Sometimes I described the orientation badly. Sometimes it proposed something that would work beautifully in the imaginary workshop where every material and bracket is already on the shelf. Then we corrected course. That was not failure in the collaboration; that was the collaboration.

The AI kept the design conversation moving, but the scrap pile kept it honest.

Winter can come now, mostly

The result is imperfect, over-discussed, unexpectedly handsome, and genuinely useful—the ideal profile of a weekend project that escaped containment. A 38.5-by-80-inch mower crate now rides on sixteen-inch wheels. A 3/4-inch axle sits under a cedar-colored 2×6 carriage. Tooled steel straps join old steel to new wood. A center tongue and diagonal braces turn a shipping frame into an A-shaped argument for forward motion.

Next comes the less cinematic but more important phase: scales, tongue-weight checks, lights, reflectors, wiring, safety chains, paperwork, and a plate. In other words, the difference between something that successfully went down a road once and something I can responsibly call road-ready.

But the central experiment already worked. I wanted to know whether generative AI could function as a practical engineering collaborator in the messy middle between an idea and a physical object, where dimensions change, pictures are ambiguous, store-bought parts meet salvaged material, and the human occasionally ignores excellent advice.

Apparently it can, and also build a road vehicle out of a lawn-mower crate.

Winter is coming. At least now I can haul things before it gets here, for the good of the realm.