This isn’t going where you think it is…..

AI is genuinely great. It lets people like me ship real things that used to require teams. I’ve lived that shift. But here’s the part we keep glossing over: simple is often better…… And left to its own devices, AI will almost never choose simple on its own.

I’ve been thinking about this while spending time under modern trucks.

The rats nest under the hood

Pop the hood on a current F-150 Platinum PowerBoost – which I have, or a comparable GM truck even just a 5.3L V8, and you’re looking at a high-power, very complex V6/V8 that is a packaging and repair nightmare. Twin turbos, hybrid components, layers of sensors, modules, coolant lines, wiring harnesses that snake everywhere, and enough emissions hardware to make a mechanic reach for the diagnostic tool before the wrench. The engine bay is a rats nest. When something fails — and things do fail — the labour to get at a relatively inexpensive part can easily run into the thousands.

Ford Broncos are the same story. High-output turbo engines packed with electronics. The complexity is driven largely by emissions standards and efficiency targets, plus the industry habit of adding every electrical doo-dad that marketing can sell. Many new vehicles (Looking at you Rivian!!) Air suspension is a perfect example. Plenty of owners who keep these vehicles long-term eventually rip it out because it becomes a reliability and cost problem. The truck often complains electrically when you do. We’re still putting systems in vehicles that a large percentage of serious owners will eventually delete. I’ve been looking at the Rivian R1T, R1S for years – but with air suspension on board – it’s a hard no, I just won’t do it.

Elon Musk has been consistent on this for years. The core of his engineering algorithm is straightforward: question the requirements hard, then delete every part or process you can. The best part is no part. If you finish a deletion pass and never have to put anything back, you didn’t delete enough — the rule of thumb is that you should be restoring roughly 10 % of what you removed. Only after that do you simplify, accelerate, and automate. He has proven the approach at SpaceX and Tesla. Complexity is the enemy of reliability, cost, and serviceability.

EVs get this closer to right…. Normally.

EVs start with a structural advantage: far fewer moving parts. Even a high-tech Tesla is simpler to work on than a comparable modern internal-combustion vehicle with its turbocharged, hybridized, sensor-laden powertrain. The Cybertruck pushed the idea further with deliberate simplification in structure and systems. With 48V and a simplified network on board, it just kept getting simpler – this was deliberate.

EV makers are not getting it perfect, no automotive manufacturer does, but as an example, putting a brake line that is uncoated, prone to rust and will 100% require a replacement at some point in an inaccessible spot – is bad design. In the early Tesla Model Y – Tesla routed this brake line above the battery pack, and another – over the rear drive motor. This means to replace those 2 lines requires 16 hours of labour to remove the pack and drive motor – to replace a $50 brake line….. and we were doing so well…

Then there’s Slate. That little EV truck is cool precisely because it starts simple. Minimal features from the factory, focus on the basics, customizable after the fact. I like the philosophy. Please give us an AWD option and a modestly larger battery pack as choices. Simple is the goal, but drivability and real-world usability still matter. You can keep the core simple while offering the capability people actually need. Being able to replace parts myself, no real stereo because my phone is enough, crank windows – all good concepts. Telling customers, it comes in one colour – and eliminating that. There’s a reason the “Scion” idea of 1000 options didn’t work here.

There has to be a balance.

The 2026 Tacoma reminder

I recently spent a few hours under a brand-new 2026 Tacoma with the 2.4 L turbo four-cylinder. What struck me wasn’t that it was primitive — it still has modern emissions equipment, direct and port injection, a turbo, the usual sensors. What struck me was access. I could see how to get the fuel injectors out. The alternator was reachable. The turbo itself was accessible with hand tools. I could “get to stuff” without inventing a thousand-dollar labour story to replace a thirty-dollar part. Even with today’s regulatory load, Toyota managed to keep the layout serviceable. That is a design choice. What we don’t want, is planned obsolescence to be a design choice – but one look at automotive youtube, specifically the likes of RichRebuilds, or CarMods and we see sub 10 year vehicles that are “Electrically totalled”. If we are concerned about sustainability maybe this, more so than exhaust should be a target.

Contrast that with the old carbureted V8 in an F-100 pickup. Crude by modern standards, but you could understand it, work on it, and keep it running with basic tools and a modest parts budget. Now look at a modern luxury V8 sedan. Layers of complexity, electronic everything, packaging that prioritizes packaging density and feature count over the ability of a competent owner or independent shop to keep it alive. The disaster isn’t power or refinement. The disaster is that the vehicle becomes a black box that only the dealer can economically touch once the warranty is gone – and repairs that are so expensive that replacing a brake line totals a 5 year old vehicle (Yes that happened to me).

Simplicity as a design constraint – bring it back…

Simplicity must be treated as a hard design constraint in modern engineering. It can also be a constraint on AI design and AI-assisted development — but only if you force it. AI will not choose simplicity on its own.

Left alone, a strong model behaves like a university graduate who is an idiot savant: enormous knowledge, pattern-matching power, zero lived experience of the long-term cost of complexity. It will happily add another layer, another abstraction, another dependency, another configuration option because it can. It has no scars from debugging a rats-nest wiring harness at 11 p.m. or explaining a $4,000 labour bill for a $40 sensor.

So what can you actually do?

In your bootstrap prompts, in your CLAUDE.md, in your system instructions, in whatever scaffolding you use, put rules that treat simplification as non-negotiable:

• Simplification is paramount.
• Do not over-complicate.
• Everything added to this project must be justified by real workload and measured against the bloat it introduces.
• Prefer the simplest solution that fully solves the stated problem.
• Before adding any new component, module, dependency, or abstraction, explicitly justify why it is required and what breaks if it is removed.

Force the model to argue for every addition. Make it prove the part is necessary the same way Musk forces engineers to prove a requirement. If you don’t, you will get impressive-looking systems that are fragile, hard to maintain, and expensive to evolve — the software equivalent of the PowerBoost engine bay.

AI is an extraordinary tool. It is not a substitute for taste, restraint, or the hard-won knowledge that comes from owning the long-term consequences of design decisions. Guide it hard toward simplicity or it will cheerfully help you build the next generation of beautiful, unmaintainable complexity.

In the meantime, I still want something closer to the mechanical honesty of that old carbureted F-100, not the elegant disaster that a modern high-end powertrain has become. Simple isn’t nostalgia. It’s survival.