I don't think it's fair comparison.

When using table saw, you still need to make a lot of decision: The depth of cut, the angle, the length. If you don't make them, your project will not progress.

With AI, you can stop decide, tell the chat bot to do whatever they want and make any decision they like. Your ability to analyze problem and make decision is not practiced anymore and begin corroding.

I'm very capable of getting sloppy cuts to fit into a project, whether I made them by hand or with power tools, or whether I paid someone else to make those cuts for me using whatever tools they use.

I mean: In cabinetry, for instance, that's the kind of slop that dado joints help to hide and nobody ever knows about it after it's glued up.

And I'm not Norm Abram over here: I can can cheat even harder than that. I can get a sloppy six-sided wooden box to look nice and square with neat joints after basic assembly is done, using just a few minutes of time and a router with an appropriate bit.

It won't pass muster in even the most simple of metrological examinations, but I don't have to care about that. (And my work doesn't detract from that of Norm, either.)

First, you know it's sloppy and can be improved and you know what to improve. An AI source code can look very high quality, until you finally realize it's not after diving deeper.

Second, you make decision to be sloppy. That means you can also make decision to do it right. With AI, no matter how much "make no mistake" you scream toward it, no guarantee whatever it produces is high quality.

No. I know it's sloppy, and I don't care to improve it. Some of the sins that the dado hides are truly hidden and can't be discovered without destructive disassembly.

I made a decision to do it fast-er instead of perfect-er. I understand and embrace the results of this decision.

>I know it's sloppy

That's exactly my point here, you know it's sloppy. To know it's sloppy you must know what a good joint look like. Many vibecode projects the author even can't tell if their code is good or sloppy. Is it important, the ability to judge and decide something, depends on each individual, personally it's very important to me.

No. It's sloppy, and it looks identical to a not-sloppy actually-snug joint. I think I've mentioned this a few times so far. (Many readers of comments can't tell what the comment in front of them actually spells out, I guess, no matter how plainly or often it gets repeated.) :)

---

Code. Despite this talk about woodworking (which I actually can accomplish with precision using a variety of tools when I elect to do that), I really can't write code for shit. I've been working on that for several decades and it's still just shit.

And I just don't care anymore.

I can feed stuff into the bot at a high level. And I can read code well-enough after all these decades of failure, so I can be hyper-critical of the results and produce precise demands for corrections. I can get resolutely-tight performance from it when it's important.

I once spent a lot of time working with the bot on RP2040 PIO assembly, where I was working with continuous high-speed signals. The RP2040 PIO has only a single-digit instruction count, a few registers, and just 32 instructions worth of memory. Individual clock cycles need counted, and slop doesn't work there at all. I accomplished what I needed to accomplish with perfection (as if it could ever be any other way). I'm pleased with the result, and I couldn't even have ever gotten started on that without the bot.

Some other times, I think that it's OK that it is slow or wasteful as long as the inputs eventually produce the right outputs. Like an intern who, when given a task to make a thing that implements a process comes back with something that's messy-but-working. It's not ideal, but it can be good enough some of the time.

I'm not fast at producing code with the bot and that doesn't matter to me: I'm getting things done (finished!) that I previously spent decades beating myself up about being unable to do. There are hardware+software things that are rolling right now and behaving themselves seemingly-perfectly that I wrote no code for at all; all I did was propose ideas, review the code where useful (because sometimes it's all fucked up), then discuss and implement changes.

Anyone can think of me as being stupider, somehow, than I used to be because of the addition of the bot into my life -- but I implore them to address that concern to me directly. I'm right here.

I agree that it's not completely a fair comparison. But here's why I think that:

With AI, you can offload almost everything. With the saw, all I'm offloading is the cutting.

I don't actually have to make any decisions when doing woodworking. I can just buy some plans and follow them. The only time I need to make decisions is when I make something unique.

TBF, that's almost all the time for me.

But also TBF, it's the same with AI.

Sure, I can let it make all the decisions, but then I'm not going to end up with something uniquely mine. It's going to keep pulling back towards its training. To get what I want, I need to keep course-correcting the AI while building.

Some people are going to be fine with that, and will let is just make whatever it wants, and call it good enough. That's just like buying the plans and following them blindly.

The difference between tablesaws and AI is, like I said above, the range of things it can do. It's like having all the saws I listed above, plus electric screw drivers, drills, laser levels, and the whole rest of the toolkit, all in 1 box. You even get a chatbot that will tell all those things what to do to follow the plans.

Making good plans has always been one of the hardest parts of both woodworking and software development.

When you have half-arsed plans (in either discipline) you need to fix things up as you go, and that hasn't changed with AI.

AI removed a lot of the other hard things, but left the hard planning step, and the hard "hide the screwups" steps.

I think that's why many high-regard people in tech start using LLM(Linus for example).

LLM let you focus on higher level, but LLM, to some degree, can also do higher level. Ask it "Design the architect of this software" and it will be happy to comply.

I've followed people using AI for a while now. At the beginning they are like you, write me a function that does this. They review every line of code. Then later they ask the LLM to write them a file, write them a module, etc. With the amount of code the LLM output, there is no practical way they review all of the code line by line. It's a slippery slope that require strict discipline and many people lack of that.