Kerrick’s Wager: on the Future of Manual Programming
Posted by KerrickLong@reddit | programming | View on Reddit | 48 comments
Posted by KerrickLong@reddit | programming | View on Reddit | 48 comments
gjosifov@reddit
or you can learn to read the hype cycle and make better judgments in the future if something is worth exploring
a good place to start is Computer Chronicles - News show from the 80s, 90s and early 2000s
https://www.youtube.com/@ComputerChroniclesYT/videos
Here I learn that the 80s were business application hype cycle - calendars, email clients, office suits etc
The hype was killed in 95 because Microsoft had Windows and Office
or that Machine Learning in the 80s was called Machine Reasoning
Learn to read the hype BS and you will have much easier time working with tech - no FOMO, no FUD
Cyclic404@reddit
The wrinkle with this is that, new gods don't come around all that often. Or perhaps they do, I'm a heathen, what do I know. With tech though, we have a nearly endless supply of hype. I for one, am glad I didn't put too much into blockchain for instance.
neilmoore@reddit
Can I interest you in my NFT? No? Well, good for you, and I hope others follow your example.
Cyclic404@reddit
Wait! Wait! I don't want to miss out. Tell me more about this, N... F... T... Hmmm.
In all fairness I doubt LLMs are going away, so worth atleast some investment.
neilmoore@reddit
No Fuggin Time (NFT) to do so. If you're in the US, you should direct your attention to our descent into authoritarianism. And, if you're not in the US, you should pressure your government to do something, whether that's tariffs or (even better) a land invasion.
Western_Bread6931@reddit
Whoa why do you want the US to be invaded. Like thats pretty loony in another direction
neilmoore@reddit
We need a new government sooner rather than later. Granted, I admit that Canada alone can't pull it off, but maybe a NATO coalition could
barmic1212@reddit
We don't ganbling with your nuclear weapons. You are the best way to change it, even without wait 4 years. The history of my country is an howto for it
neilmoore@reddit
Thanks! What country is yours? Based on the language of your comment history, I will assume either France or a former French colony.
Anyway, thank you for the hope! And I'm glad that France points its nukes in every direction, and not just towards the "enemies" of NATO.
barmic1212@reddit
Yes I'm French (check January 1793). We have a long story of friendship with USA but this type of friend who can say to you fuck when you make a wrong things.
neilmoore@reddit
Thank you! We definitely "make a wrong thing" right now, and I apologize for that. Please forgive us!
neilmoore@reddit
I'll just say (Tom Lehrer, "MLF Lullaby", from the early 1960s)): "A nuclear deterrent force, including our current friends, like France (audience laughter), and our traditional friends, like Germany (more laughter)."
neilmoore@reddit
Also, if the chaos clown does manage to launch nukes, I hope it's not at you. There's not much that I can do other than sacrifice myself :(
IDoCodingStuffs@reddit
I don't know, they have a promising track record
neilmoore@reddit
Please repeat!
RabbitDev@reddit
Hey, I'm willing to buy your NFT. Near Field Thermal Transistors are hot.
neilmoore@reddit
Negative Frempature Toafficient.
KerrickLong@reddit (OP)
It's all about risk assessment. Unlike some fools who lost everything by putting their life savings into blockchain, I'm only talking about learning one new skill at the cost of a few dozen hours and a few dozen dollars. If that skill turns out to be useless, I haven't lost more than a month's worth of nights and weekends and a couple nice dinners out.
pobbly@reddit
Your comparison to pascal's wager is apt and a useful way of thinking about it.
rzwitserloot@reddit
A few of the key problems with Pascal's Wager apply to Kerrick's Wager just the same:
Pascal's Wager's key logical error is that it ignores the fact that many religions exist. And Kerrick's Wager is logically a nice-sounding pile of bad reasoning for the exact same analogous reason: It fails to acknowledge that more than one career exists.
The job of 'managing a fleet of AI coding agents' is very different from 'write code'. And it won't be the only job for humans that exists in a world where AI coding agents exist and are decent. Who says that the 'manage a fleet' job is what you are good at in this world? Why not another amongst thousands of other career options? What made 'manage fleets of AI' so special?
Why not learn to bake artisanal cakes? The logic that you use to 'make the positive EV bet' also works here. It's also a positive EV bet to learn how to bake artisanal cakes according to its logic.
Sure, it's somewhat more likely that, given your background as a coder, you'd be faster at becoming a self-taught AI manager, and probably better at it too than an average human, vs the 'artisanal cake baker' career path. But then, if we must posit that AI based coding agents are good enough to significantly endanger your future career, I think it's rather weird to assume that the job of 'AI manager' is not significantly at risk of being taken over by AI. The increased odds you'd be a good AI manager are countered by the increased odds that job is irrelevant due to AI.
I think if you're really worried about swiftly being relegated to the status of 'horse' in a world where cars exist, then you should think about which job is highly unlikely to be AI-ified, and start spending your limited time self-teaching that one.
KerrickLong@reddit (OP)
I’ve thought about that too. I bought a farm and I’m one year away from finishing its organic transition. I’ve learned to care for and milk dairy cows, operate a tractor, and grow vegetables on the Elliott Coleman scale. This isn’t my only bet; unlike Pascal’s Wager, it’s non-exclusive.
neilmoore@reddit
Well, you're definitely more equipped than me to survive the collapse of Western civilization. But I hope that, at least, you'll drag me and my family into the "new world order" (perhaps as an "involuntary servant") rather than leaving us to starve.
KerrickLong@reddit (OP)
Umm… what? It’s just skills diversification in case knowledge work loses value. If civilization falls, I’d die.
blafunke@reddit
Before long everyone who jumped on the vibe bandwagon is going to come begging for the the most patient and capable "manual" programmers to debug the broken slop they let their AI infused editor barf out. If you stand that type of work you'll have a job. And there's always toilet cleaning (which is kind of the same thing...)
aookami@reddit
Man if you took like a couple hours at most with any moderately sized code based you would understand that llms are light years away of being able to operate on their own
suckfail@reddit
If you read the article he specifically says as a tool.
And as a tool they can be very useful. I use copilot everyday, usually with Claude 3.7 Thinking and it's quite good. But it's a tool, I have to wield it and there is definitely a skill in using it and prompting. Adapt or die.
One nitpick from the article: he lists prices to learn AI. Anyone can learn it for free. Google Flash 2.0 is free to use unlimited basically from AI studio. And VSC + copilot is free (with limits). Nobody needs cursor, at least not to learn.
aookami@reddit
yes, it is very good at thinking about things at class level, even package level
on system level it shits the bed consistently
KerrickLong@reddit (OP)
I'm familiar with Chat-Oriented Programming (CHOP) with Google Flash or ChatGPT, and with AI completions with Copilot. The thing that is new and unknown to me is the agentic stuff. The only free version I've seen is Aider with a local LLM.
suckfail@reddit
Sorry, but what the fuck is CHOP?
Nobody who actually uses AI calls it that. It's vibe coding.
KerrickLong@reddit (OP)
CHOP and vibe coding are two distinct workflows.
CHOP is a back-and-forth workflow where a human programmer asks a chatbot for some code, judges it, and (if it's good enough) integrates the code they get with the codebase or (if it's not good enough) writes a new prompt to get better code. "Coding via iterative prompt refinement."
Vibe Coding is a directed workflow where a human prompter asks an LLM that is capable of editing the filesystem and running CLI commands to do so. Looking at the code is optional, and the LLM integrates the code itself.
oadephon@reddit
Am I seriously underestimating the skill needed to use AI, or is there really just not that much to learn?
There are a few workflow things depending on Cursor or whatever you use, but the actual process doesn't have a lot of deep skill. It's just prompting, describing your code, and making it fit within whatever system Cursor uses.
And even if there was skill at that, you'd have to relearn it every time a new Cursor-like comes out. It's not learning a whole new paradigm, it's just learning a new IDE.
maxinstuff@reddit
There is a lot to the actual underlying technology - but that’s not what the majority of developers are playing with.
Most “AI” development happening today is rather thin wrappers around one of the big LLM vendors.
It’s progressing, but at the moment this is why there are a bajillion AI startups all doing similar things and hoping to get their exit before the hyper scalers rug-pull them (by just releasing their own version of whatever minor feature they’re selling).
techdaddykraken@reddit
It’s more system design that is the skill to learn. If you weren’t good at it before AI, you’re definitely going to have to be good at it to maximize the use of AU
Caraes_Naur@reddit
Here's the plot twist: the level of skill needed to use "AI" doesn't have a floor, it has a ceiling.
KerrickLong@reddit (OP)
That would be a cruel twist of fate
thetdotbearr@reddit
There's not that much to it, but I think there is a minor learning curve just in figuring out what exactly the models are good for, where they can be of actual use in your workflow and what their limits are.
huyvanbin@reddit
So the blog post itself seems to have little content aside from being a proxy for Steve Yegge’s fantastic prediction that there will be a huge market for his company’s products in the future and everyone should buy, buy, buy. And I don’t want to create a new thread for Yegge’s post so I will just write my response here: whatever happened to the Mythical Man Month?
Why does Yegge apparently think that AI just makes it obsolete? In other words, even if a junior developer can actually manage 100 AI agents productively, who says that will help the company achieve its goals any more than suddenly hiring 100 new human developers would, putting cost aside? Your average company’s ability to improve its products is simply not bound by the amount of code it can write, in fact that’s such an absurd strawman that I have trouble believing Yegge himself believes it.
It’s well known that (in, let’s say, the pre-AI days, if we grant the premise) a company of 5 could easily build an MVP product that would blow a legacy product with 100+ devs out of the water for some specific use case. The reason isn’t that they’re better devs, but that a legacy product has so many dependencies, chiefly on a business level, that the 100+ devs are basically sitting around doing nothing because every one-line code change has a massive potential set of repercussions.
Now imagine each of those 100 devs has access to 100 AI agents round the clock. What changes? If anything the problem gets worse, because the dev now can’t even make a case for his own code. He effectively becomes like his scared manager who has no way to know if the smallest code change will cause a massive shitstorm, because he didn’t write it himself.
blafunke@reddit
It's just called programming.
KerrickLong@reddit (OP)
I really hope we come out of the other end of this decade saying that. I do not want this to be like structured programming, high-level languages, object-oriented programming, version control, agile software development, web applications, enterprise open source adoption, test automation, continuous integration, continuous delivery, mobile apps, cloud computing, server-side JavaScript, or devops.
neilmoore@reddit
Deliberately naming something after yourself is definitely a red flag. I mean, it wasn't Abel who came up with the term "abelian group"; nor Gauss for the many many things that are called "Gaussian".
Also, using the capitalized "He"/"Him"/"His" when referring to works of Christian theology definitely implies some level of agreement, whether or not that's what you intended. But good on you for using Yeshua's "correct" pronouns, and I hope you'll do the same for everyone else.
vincentofearth@reddit
😂 lol, I never noticed that piece of irony
neilmoore@reddit
nice!
KerrickLong@reddit (OP)
I've been an atheist for decades.
neilmoore@reddit
Sure, me too (since the age of 8 or 9, back in the 80s). But then why did you capitalize Yeshua's pronouns? Out of respect for Christians as opposed to Muslims, Buddhists, CM Hindus, Jews, etc etc?
KerrickLong@reddit (OP)
I'm not sure who Lord Rama is, sorry. I'm an atheist due to not having any proof, not due to an extensive comparative religion study.
I capitalized the Abrahamic God's pronoun because that's what I see other people doing. A non-entity can't have a preferred pronoun, but a large community can.
neilmoore@reddit
Thanks for replying, and sorry to impute Judaeo-Christian-Muslim favouritism onto you!
KerrickLong@reddit (OP)
No worries. My use of Pascal's Wager probably primed you to assume I agreed with said wager.
neilmoore@reddit
Naah, it's clear that you don't agree with that, as should no-one (for example, the exact same argument could be used to prove that everyone should worship Satan). My objection was entirely to capitalising the pronouns.