Post

Quitting programming as a career right now because of LLMs would be like quitting carpentry as a career thanks to the invention of the table saw.

162 reposts937 likes

So are we automating away the boring stuff (sawing) or all the production stuff (cutting all the pieces to size)? In which case are we left with a) designing the product & automation process (fun), loading and unloading the machine (less fun) and assembling the parts (for non technical people to do)

For me it feels like we’re at the point where both future are possible. However it feels as though one is desirable for craftspeople and the other is desirable for industrialists. And we sort of know who usually wins that battle.

An interesting thing about that is usually industrialists win because only they have the capital to build the factories Thanks in part to open weights model, building an "AI coding factory" is affordable to individuals It's rare to see a disruptive technology like this so widely available

I think you’re in a local minimum here; tools get better and people get better at using them. Eventually these tools will be as trustworthy as a compiler, and it would make about as much sense to manually program as it does to manually compile.

I can see how deterministic tools like compilers and table saws can be evaluated and adopted with confidence that they are a net positive. But blackboxed probabilistic tools like LLMs seem to defy rational evaluation and require "vibes" instead. So I'm not sure those are apt comparisons.

Non-AI tools (templates, IDEs, etc) should have already addressed these issues though, at least to a degree. The perception that rote repetitive work is *not* valuable is something that the AI zeitgeist is pushing. That's not a bad thing, except to the people who want to just do rote things.

Your right about the pre-LLM tools, and it's only the urgency of the questions surrounding the software developer's relevance that has changed. It's inevitable that people with a loyalty to the technical arcana of passing era will be disappointed.

I feel like I keep having to say this, but you don’t know that’s true at all! It could be like quitting agriculture as a career thanks to the invention of tractors and combine harvesters instead!

I’d agree with all these statements: * increase in AI usage is inevitable * we can’t put it back in a box * people who want to thrive in the future should be exploring it * countries should be engaging with it because if they don’t they’ll be left behind

But I feel like you go a lot further into statements which *aren’t* supportable: * we should make training on other people’s IP legal at the potential cost of many people’s livelihoods because we will otherwise miss sunlit uplands

* a massive rise in AI usage will definitely not have negative consequences to jobs in software engineering, design, law, call centers, illustration, music production and other knowledge professions You *might* be right in both of these, but you don’t have *evidence* to support those statements.

I agree with your second take. Regarding AI coding assistants, the best code is actually free and open source, so they're not trained on legal IP, per se. Whether this is the Industrial Revolution or the agricultural revolution I don't know.

The practice of software development isn’t about writing code, it’s about decomposing problems and figuring out what capabilities users need, in the context of what is technically possible. In the presence of sufficiently expressive programming languages, this is much clearer.

The assistance of AI to the programming process is effectively that you have a more expressive language in which to describe your objectives, and that language (English) gets “transpiled” into a machine language (a high-level programming language) in preparation for execution.

I base my opinion here on my own experience exploring this technology, which has convinced me that you need a *huge* amount of knowledge, skill and experience to use these tools effectively That's why I like the comparison to dangerous power tools

(I might only quibble here that carpentry as a skilled profession has, I’m guessing, 100% decreased as a percentage of the population since the invention of the power saw)

Overheard in a coffee shop this morning, android dev refused to use any AI. Like, good luck? Apply for all those jobs looking for programmers who are slower than everyone else, or maybe a job that does pull requests on post its?

People didn't quit carpentry when power tools were invented, they just continued to refine their craft. Today the stuff they build fetches a premium because people recognize it as superior to mass produced stuff built by machine.

I restored the Edwardian sash windows in my son's bedroom last year, and I did regret what power tools have done for the carpenter's skill. Efficiencies of some kind, I grant you. But the work was miraculous - you could not get a cigarette paper between some of the joins.

Yeah, I would absolutely quit a career in carpentry if my boss fired all my co-workers and replaced them with a table saw that periodically leaps free and cuts people's fingers off without warning

LLMs strike me, strictly from a development perspective as a pretty shit tooling for anyone who has written plenty of their own tools across multiple domains. But, lots of devs never bothered or bother to learn how to write tooling.

Let's not confuse tools with modules. Different paradigm, and a tool never exists inside the end product. A precast concrete form isn't a tool, the machine that makes whatever precast form you want with dimensional inputs is. It's an interesting set, but low stakes tell us nothing about LLMs.

High stakes r what a bunch of us play for, and right now its high stakes game is trash. And it's competing against decades of tool history that already exist and re-exist to chunk out entire deployable/functional applications for hot customization to exact business requirements in a button press.

Given the non-deterministic nature of LLMs, maybe sheep dogs are a better analogy than a table saw! bsky.app/profile/simo...

Simon Willison@simonwillison.net

That's entirely true: they are non-deterministic, which means you have to learn how to use them in a way that takes that into account Sheepdogs are non-deterministic too, they're still invaluable tools for someone who learns how to work with them

I've taken to talking about LLM as chef's knives -- very effective but you might take a finger off Tho sheep dog as a metaphor is pretty apt... border collies are super bright, super energetic, and occasionally have a spontaneous mental breakdown and aggressively go for someone

Love this observation: that table saw analogy really is an ink blot test for people's opinions of LLMs bsky.app/profile/yayp...

yay pineapple@yaypineapple.bsky.social

I went to your profile and read and re-read that table saw comment, and one can come away with several different interpretations! It's kind of like an ink blot test for one's thoughts about LLMs

Using an LLM is like having an adversarial agent data mining my research methodology and encumbering me with irrelevant distractions. It's an ADHD sheep dog that forgot to take its Ritalin and is being distracted by an orgy of squirrels and bouncing balls.

Agree. Are they perfect? No. Neither am I. Are they useful? For sure! Am I better with one in my tool kit? Absolutely. Are they overhyped? Hell yeah! Sadly, so is everything new in tech. I just don’t get the hate. Lean in. Learn how to use them. Get better. Lather rinse repeat. #kthxbye

I spent more than an hour strategizing and designing with Claude today. It was great and reminded me of having a couple of young programmers and a whiteboard. I imagined asking them how long to implement. Claude did it in ten minutes. I spent another half hour refining. Orders of magnitude for sure.

I just spent a few nights writing up my approach to trying to introduce some form of determinism to LLM-assisted coding workflows. Not a one size fits all of course. Would be interested in your thoughts. mariozechner.at/posts/2025-0...

mariozechner.atTaming agentic engineering - Prompts are code, .json/.md files are stateTreating LLMs as shitty general purpose computers we program with natural language. Because throwing shit at the wall wasn't working anymore.

Honestly, 'cheap homemade CNC machine' fits the bill quite well, but the number of people who might get that reference is vanishingly small. I suspect sheepdogs are much more reliable than cheap homemade CNC machines, which are, in turn, much more reliable than getting an LMM do do all your coding.

Which… doesn’t seem unreasonable. If you love making things by hand and see the oncoming trend toward cheap mechanically made goods that compete with your craft, it’s not unreasonable to start thinking about other possible careers.

At a code level: strongly agree At a societal level: given how the range of employers right now goes from everyday grifter to nation-state-level robber baron, quitting programming may be the most ethical alternative most laborers have.

The rejection lead to a beautiful arts movement. I think because it was treated as a creative constraint, rather than a reactionary impulse? Building vs quitting.

Fully agree. I think a lot of people misunderstand what a LLM is actually doing. I have yet to see any dent being made in the value of human creativity.

I'm really not. We're on the ground floor of a disruptive technology emerging into a capitalist system. It's no different than when cars came along, or airplanes, or CGI in films. If people like the product, then the paradigm will shift — the more they like the product, the faster it shifts.

I honestly feel bad for everyone impacted (my job as a dev is also in the crosshairs), but their skills aren't just vanishing, and their experience is gilded. Creativity is still the commodity. This just changes how we create, and we all just have to adapt and evolve as the new landscape is forged.

LLMs are completely lost in my world, which is full of institutional legacy code written mostly by other people over the last 15 years or so. The idea that DOGE was going to use an LLM to 'rewrite' all the software running the federal government wasn't just ridiculous. It was harmful. Dangerous.

Please share more about this point of view or cover it in your next blog. I’ve been thinking that now is the best time ever to get into programming because there’s so much potential to create, but again I’d be interested to hear what’s motivating your POV as I’m still really new to this space.

It's gonna be fun when everyone only knows how to talk to their agent, but knows nothing useful at all about actually writing code. I guess it'll keep us security nerds employed, so there's that

I didn't realise the table saw was so riddled with copyright theft, massive questions around ethics, had real impact on climate change, caused so many, many people to be exposed to the horrors of moderating before table saw could be used by public, and the absolute need for regulation 🤷

Pretty sure feeding and transporting a guy to saw boards by hand costs more to the environment than the electricity consumed by a table saw

But but... I thought the table saw was gonna do all the wood sourcing, the design, the planing/surfacing, the joining, assembly, the finishing and sealing. So remind me why we even needed carpenters any more?

I shared Armins video about agentic coding with our devs. They are all mid/end 20 years old. They said this video was too deep in it and they didn't get what he is talking about. I told them that I expected this and that they should start looking into it more in depth to not fall behind.

I hate the "fall behind" comment. It's complete BS and you're worrying them about something that isn't even close to the fundamentals. Any developer can learn AI agents in a few days. At this point there's no risk of falling behind.

I'd second this point. LLMs/coding tools allow catching up in record time. Also since the tools tend to get better and better, some older problems have already been solved, which required no longer useful patterns to work with LLMs.

They were in 'neglecting' phase. Trying one or two things and saying 'this is slop...'. Now they get that you need to learn how to do it right to make it useful and that this is also a skill.

The agentic coding stuff Armin is talking about is really at the cutting edge right now, it hardly existed six months ago Not surprising if it takes people quite a bit of work to catch up to that!

Down by ~2/3rds since 1900. It’s good though. The really skilled programmers will be in high demand. The ones who take 2 weeks to figure out why the program isn’t working will do something else. 20 years from now the average K-12 class will be 10 kids (or Trump jr will be ruling over N Korea east).

I hear senior devs complain that their job is now catching the errors made by junior people who trusted the LLM too much. But this may also be a version of a complaint senior devs always make

I think the situation is somewhere between knitting machines (agents like Claude Code) and table saws (more like autocomplete/chatbots). But it is golden time to become a programmer. These tools offer endless tutoring/help. They decipher errors to plain text. Discuss high level approaches. Etc.

Yeah, I listened to that podcast the other day where Sam Altman was talking about 'agentic power tools' where the agentic saw just kind of sits there, doesn't read your emails or connect to your bank account and waits for its human orders. Seems like a lot of folks here heard his other podcasts...

I think the cognitive decline thing is overblown, personally If you let LLMs think for you then sure, you won't learn as much - but you can also chose to use LLMs in a way that actively challenges you to learn more and take on more ambitious projects

Agreed. But to further the analogy it would be more like quitting carpentry as a career after the invention of a table saw that sometimes cuts straight lines and sometimes cuts crooked lines and sometimes just goes berserk and cuts your hand off. Even when given the exact same instructions. 🫠

I've gotten some useful summaries, snippets, and decent references from AI. It is like a slightly better Google in some cases (in other cases, worse). I worked w/a colleague a couple months back on a problem w/sorting derived, calc'd fields. We got 20 different answers in 20 prompt attempts. 🫠

None of the 20 answers provided or even led to a workable solution that covered all of our needs. We eventually had to request either a change of the specs or a change to use a SQL view vs. EF context. Our lead advised strongly against anything non-ORM. AI was less than useless in this case.