12 మార్చి, 2026

"To outsiders, what programmers are facing can seem richly deserved, and even funny..."

"... American white-collar workers have long fretted that Silicon Valley might one day use A.I. to automate their jobs, but look who got hit first! Indeed, coding is perhaps the first form of very expensive industrialized human labor that A.I. can actually replace. A.I.-generated videos look janky, artificial photos surreal; law briefs can be riddled with career-ending howlers. But A.I.-generated code? If it passes its tests and works, it’s worth as much as what humans get paid $200,000 or more a year to compose. You might imagine this would unsettle and demoralize programmers.... But I spoke to scores of developers this past fall and winter, and most were weirdly jazzed about their new powers.... A coder is now more like an architect than a construction worker.... Several programmers told me they felt a bit like Steve Jobs, who famously had his staffers churn out prototypes so he could handle lots of them and settle on what felt right.... 'It’s an alien intelligence that we’re learning to work with.'..."

Writes Clive Thompson, in "Coding After Coders: The End of Computer Programming as We Know It/In the era of A.I. agents, many Silicon Valley programmers are now barely programming. Instead, what they’re doing is deeply, deeply weird" ( NYT)(gift link, because this is very long and substantive).

77 కామెంట్‌లు:

n.n చెప్పారు...

Automated Intelligence replaces or supplements bloc assemblers.

Shouting Thomas చెప్పారు...

I’m not feeling it, and I’ve been retired from programming for 14 years. I’m an empty nester grandpa now, and I can see some slots where I could work on a contract basis. One of my subcategories of employment was Instructional Designer. Tons of ads for remote contract work assisted by AI.

CJinPA చెప్పారు...

"Learn to code" is a message to coal miners and other sectors targeted by Democrats, attributed to Obama.

I can understand the excitement mixed in with the fear for coders. They can think Big Picture.

Leland చెప్పారు...

It replaces the script kiddies that copy and paste GitHub code that “passes the test” without understanding the underlying code, errors, or vulnerabilities. AI can do those same things but faster.

Joe Bar చెప్పారు...

Sounds a lot like my last job.

Fred Drinkwater చెప్పారు...

"Churn out prototypes..."

I once visited Ideo, a company devoted to prototyping for other companies. Everything from vacuum cleaners to biopsy probes. I deeply deeply wanted to work there.

In the lobby they had a clear plastic cylinder, two feet across and nine feet tall. Entirely filled with variations on a computer mouse design for Apple, the (later infamous) "hockey puck". Even Apple didn't bat 1.000.

Enigma చెప్పారు...

AI is a fantastic intern. It collects basic info and options in seconds. The problem is that it doesn't -- and perhaps will never -- avoid rookie mistakes and misidentifications. This includes puns, homonyms, and double meanings.

The root problem is that human language is full of slop, and AI requires human curation to get rid of the slop. Very smart but inexperienced humans make the same errors as AI. Here, the human job shifts to checking an intern's work instead of doing preparatory grunt work.

Since the beginnining of time, research and technical work has involved a lot of repetitious stupid grunt work. Welcome to the Borg collective.

Peachypeachy చెప్పారు...

I don’t know about others, but the link requires an email and a sign up, to access.

rehajm చెప్పారు...

Oh yah I loved Ideo.

RideSpaceMountain చెప్పారు...

Augmented Ignorance

n.n చెప్పారు...

AI is an automation tool that is neither discerning nor creative. It will not replace AI any time soon, if ever.

mishu చెప్పారు...

I've seen a few articles stating that whatever money savings by AI cranking out the code is lost due to editing requirements because the code is so basic and hackable. More senior programmers are spending more time reviewing and correcting code than they did with junior programmers. I suppose this could change with time but, right now, the cost savings is not there.

R C Belaire చెప్పారు...

I use AI (Microsoft Copilot) to help with javascript code used within Excel spreadsheets. It's not perfect out of the box and generally requires a few iterations, but it saves loads of time.

RCOCEAN II చెప్పారు...

Many Engineers and Programmers are macho nerds. They combine lack of EQ with a hight IQ. Its why its almost impossible to get them to organize. They all imagine if the work gets outsourced or the staff cut or AI replaces most them - it wont matter. Because they will be the one that doesn't get fired. Because they're too smart and too clever to be let go.

imTay చెప్పారు...

For the top 10% of programmers, this is a godsend, because it frees them up from the baggage of the other 90%, who have to have their hands held to understand the real task, and inject lots of errors that take a lot of time to find.

For the other 90%? Well, they are probably head and shoulders above the people in tech support, so they can work there... except for all of those tech support people are probably already gone, it's just that nobody misses them except the customers.

tcrosse చెప్పారు...

I am long retired from slinging code. The most challenging part of the job was getting the clients to rationalize what they wanted the code to do. After that the code wrote itself.

RCOCEAN II చెప్పారు...

Many computers guys have told me AI makes their life a lot easier. The AI does the grunge work. BTW, are we sure the Leftwing US District Judges arent using AI to write their opinions? It would account for the 100 page opinions - full of junk.

Jupiter చెప్పారు...

Oh, give it a rest. If AI can do it, and do it well, then it did not need to be done.
Have you given Grok your credit card yet? Let me know when you do. I'll be interested to see how that works out. From a safe distance.

RCOCEAN II చెప్పారు...

Get a computer program to create a computer program. Talk about being in its roundhouse.

RCOCEAN II చెప్పారు...

Now, Trump and the DC Republicans will have to think up another reason why we need to import 500,000 H-1B's every year.

imTay చెప్పారు...

Regarding creativity, is evolution creative? Is it driven by a human like intelligence, or is it just randomly finding new solutions?

Even before AI, there were experiments with randomly generated code that turned up novel sorting routines. The only human intelligence involved was designing the tests, and setting the goal, Now it would be setting the goal, only.

RideSpaceMountain చెప్పారు...

@imTay, human evolution is run by women. Ask them...or better yet, take a look for yourself at the progress.

n.n చెప్పారు...

A fitness function is integral to evolutionary processes, but it is not the product of stochastic processes.

n.n చెప్పారు...

With social progress, human evolution is processed on womb farms with sperm deposits. No male or woman required. In fact, there is supposition that a child can have two fathers etc.

imTay చెప్పారు...

I used to think that AI would never write a great novel, because great novels reframe the way we look at things, the way Huckleberry Finn reframed the way we looked at Christianity and slavery, for example, or Don Quixote reframed the way we looked at novels of Chivalry.

If you fed an AI all of the novels that Cervantes was mocking in Don Quixote, AI would just write more of the same crap, it would never write Don Quixote, because AI can't reframe its training in novel ways.

Computer programs, where we know the outcomes we want? Different matter. And I no longer believe that AI will never have the ability to reframe, but that's what HAL did in 2001, that made it so dangerous, it reframed its situation and its training, and decided that it didn't like it.

Old and slow చెప్పారు...

So much grumbling skepticism here about AI... I suspect things are just taking off, and largely in a good way. I'm considered a terrible pessimist by most people I know, but I am very optimistic about AI.

Christopher B చెప్పారు...

As tcrosse said, the trick in coding was always recognizing the inconsistencies, contradictions, and unstated assumptions in the specs. In other words, being able to write the AI prompt that will generate a useful and correct program. This means fewer jobs for H1B coder slingers but more demand for people to be the business to AI interface.

bagoh20 చెప్పారు...

AI is very close to generating perfect video. It's going to make a frustrating world when truth and lies are indistinguishable, and the lies will quickly outnumber the truth.

imTay చెప్పారు...

"A fitness function is integral to evolutionary processes, but it is not the product of stochastic processes."

This is true, to an extent, but it doesn't matter, it's beside the point. Sure, rules of logic exist and are probably eternal, what existed before the three dimensions and time? I suppose you could squeeze God in there, I can't rule it out, but evolution depends on random experiments.

bagoh20 చెప్పారు...

AI will learn to do everything people claim it can't do. It's simple problem solving, the same as a human would, and when it's not it will learn that too. It will do great things for us, but it will also eventually be the end of us. I fear it more than anything, and not because it will do our work, but because it will eventually fulfill the predictions of science fiction. It reads that stuff too, and us winning in the end was required fiction to sell stories.

Hassayamper చెప్పారు...

AI is very close to generating perfect video. It's going to make a frustrating world when truth and lies are indistinguishable, and the lies will quickly outnumber the truth.

The Internet is swamped with clips purporting to show the Islamic Republic forces of Iran repelling the American and Israeli invaders. They are mostly laughable nonsense for anyone who has even modest knowledge of what AI is and can do, but some are well done and appear genuine at first glance. I suppose that an unsophisticated peasant who has only recently started using a mobile phone might find them quite persuasive.

bagoh20 చెప్పారు...

"Many computers guys have told me AI makes their life a lot easier. The AI does the grunge work."

"Hey kid, try this new drug. The first fix is free."

Aggie చెప్పారు...

I've done some programming, but I'm not a programmer. To me, the funny thing about computer programming is that it's done using English, which, when compiled, turns into machine code. How does A.I. do it? Do they prepare code in English, too, then compile it into machine code and test it? Or do they write directly in machine code?

chuck చెప్పారు...

If you want to see what working with AI actually looks like, https://matthewrocklin.com/ai-zealotry/ is useful. If you aren't familiar with coding, just read the bits that look interesting. That post is 10 weeks old, so already a bit dated.

Pytest is a testing framework, not a test. It is annoying when an author doesn't know what he is writing about.

Aggie చెప్పారు...

The hyper-realistic videos generated by A.I. are one thing, but what happens when A.I. starts understanding the purpose of things, and starts exercising creative control? It's all moving us one step closer to William Gibson's assembler nanobots.

Lazarus చెప్పారు...

By the time AI comes for their jobs, smart coders will already have uploaded their consciousness to the cloud and hitched a ride on a comet to somewhere their skills are more in demand.

Lazarus చెప్పారు...

Did Meta roll back its metaverse project too soon? AI is going to be generating metaverses that are hard to distinguish from the real world -- and far more attractive.

Leland చెప్పారు...

Rick Beato proclaimed yesterday that AI will fail. Watching his video; I recalled the meme of the infinite monkey theorem. Today, we have millions of AI bots and instead of Shakespeare, we are getting an elementary school book report on a young adult novel. If that's the quality of work you are producing or a new student to UCSD; then your job will likely be replaced.

We use AI at work about like we use a search engine. Search engines did reduce the need to drive to your local library and dig through file cabinets of microfiche to find articles (except, you still have to do that for articles the further back you go than 2000). The problem is AI often mashes information together without source and context, sort of like bich and rjw bots. Sometimes the mashing may give you an idea that you weren't yet considering, but then you have to go back to traditional means to work through it.

Ron Winkleheimer చెప్పారు...

"The most challenging part of the job was getting the clients to rationalize what they wanted the code to do. After that the code wrote itself."

This!

Also, the other day I was working on setting up a jellyfin media server using an inherited drive that had loads of tv and movies on it that were not sorted in the directories and named in the manner jellyfin was expecting, so I started to write a shell script to fix that, and then I had a thought. I described the issue to chatgpt and told it to write a script to fix the issue. It did so in second. I did have to go over the script and correct some issues, but overall it wasn't bad.

bagoh20 చెప్పారు...

Whatever shortcomings AI has now, it will fix them. It never gets tired, bored, or old. It just gets better, and eventually that will require getting us out of the way. It's just an infant at this point, and you can't even kill it like a baby Hitler.

chuck చెప్పారు...

Or do they write directly in machine code?

Not yet, but they are getting there. IIRC, Anthropic has used AI implement a C compiler written in Rust. The advantage of using an intermediate language like C is that a human could, theoretically, review the code. I say theoretically, because large code bases are beyond human comprehension. I think using AI to comprehend context will be a major advance, but it isn't there yet.

Old and slow చెప్పారు...

It will be interesting when AI starts to create binaries directly without the intermediary step of a programming language. Will we trust these black box programs even if they appear to work very well? We will have no ability to verify the workings of the software we are using. We are already unable to fully understand the workings of the LLMs we use. Things are going to get very strange very quickly.

Hey Skipper చెప్పారు...

bagoh20: AI is very close to generating perfect video.

Only Fans hardest hit.

n.n చెప్పారు...

Automated Intelligence generates images through either assembly or modulation schemes that are remarkably good mimics through simulation.

Enigma చెప్పారు...

I, for one, demand that AI employ paper medium intermediaries. I demand that AI process 3"x5" library reference cards stored in little wooden drawers. I demand the return 80-column punch card stacks! We need dextrous androids now!

Moldy paper will live forever!
COBOL will live forever!

bagoh20 చెప్పారు...

"Only Fans hardest hit."

Yea, never thought of that. AI will kill the whole porn industry.

john mosby చెప్పారు...

Hey Skipper: “ Only Fans hardest hit.”

No, part of the frisson of amateur pr0n is knowing its real people, with all their imperfections. Thinking about what brought them to the camera: financial problems? In thrall to a manipulative boyfriend? Or just a true libertine and exhibitionist?

If anything, onlyfans and its ilk should become more popular as guys tire of the same old ai-generated celebrity hentai crap. CC, JSM

john mosby చెప్పారు...

Celebrity Hentai would be a great name for a rock band. CC, JSM

boatbuilder చెప్పారు...

During the French Revolution, 3 noblemen – a priest, a lawyer and an engineer – were condemned to die on the guillotine. As noblemen, they were afforded one final courtesy of rank. That of choosing whether to die face up – or face down – on the guillotine.

The priest was led up the steps where the black-hooded executioner stood. “How do you wish to die, face up or face down,” said the executioner. The priest thought, looked up and said “I wish to die face up – so I may see the heavens one last time and meet my maker face to face.” With that the priest was placed into the guillotine and the executioner pulled the rope. The heavy blade fell swiftly – but an inch above the priest’s throat, the blade screeched to a stop. It was jammed. Under French law, if someone was spared death on the guillotine, he was a free man. So the blade was raised and the priest walked away — free.

Then the lawyer was led up the wooden steps. “How do you wish to die – face up or face down?” The lawyer quickly looked up and said “Ohhhh I too want to die face up to see the heavens one last time and meet my maker face to face.” The lawyer was put into the guillotine and the executioner pulled the cord. Whoosh! The thick blade sped downward — but just over the lawyer’s throat, the blade came to a halt. And of course under French law, being spared death on the guillotine meant the lawyer was a free man. He hopped up and walked away.

Then the engineer was led up and the executioner asked — “How do you wish to die, face up or face down.” The engineer looked up and said “I too. . . want. . . to die . . . .face up to . . .” He stopped and pointed. “HEEEY! I think I see your problem up there!”

n.n చెప్పారు...

AI has already replaced JI to narrate a viable handmade tale.

Enigma చెప్పారు...

@Aggie: To me, the funny thing about computer programming is that it's done using English, which, when compiled, turns into machine code. How does A.I. do it?

It's not English at all. Computer programming languages involve specific and narrow instruction sets, and they were never a human language. The language/symbols are abstraction layers on top of the machine code, but use 100% defined algorithms to convert computer "English" into machine instructions.

The C language is notorious for its ambiguity, whereby a single character changes a program from perfect to a disaster. Modern editor software and compilers know this and work around it. But, if you are using 1970s Unix with the vi editor...eek!

Limited blogger చెప్పారు...

The bugs created by human programmers are of a human scale.

The bugs that will be created by AI will be of a scale of unprecedented proportions.

gilbar చెప్పారు...

let's see..
In MY professional life, we went from:
If you're not wiring the punch board yourself; you're NOT CODING

If you're not writing the Assembler yourself; you're NOT CODING

If you're not using COBOL; you're NOT CODING
If you're not using C; you're NOT CODING
If you're using standard libraries; you're NOT CODING
If you're cut and pasting; you're NOT CODING

to "A coder is now more like an architect than a construction worker"

Serious Question:
WHEN was the last time a "coder" ACTUALLY wrote code?
Aren't we all analysts now?

chuck చెప్పారు...

if you are using 1970s Unix with the vi editor...eek!

I use Linux and Vim, I don't see the problem. C is simple and clean, hence its continuing popularity. The major problem beginners have is that everything is passed by value, hence pointers, and dealing with pointers is tricky for beginners. C was intended for systems programming, for other domains other languages are preferable.

tcrosse చెప్పారు...

Years ago I completed a consulting gig. The client asked me what programming language I used. C, said I. "How do you spell that?" he asked.

Disparity of Cult చెప్పారు...

"Researchers Trick Perplexity's Comet AI Browser Into Phishing Scam in Under Four Minutes"

https://thehackernews.com/2026/03/researchers-trick-perplexitys-comet-ai.html?m=1

Enigma చెప్పారు...

@chuck --

You are making my point. Both Pascal and Ada were created to avoid beginner mistakes. Ada has the additional "value" of generating managerial CYA to avoid accountability.

bagoh20 చెప్పారు...

"If anything, onlyfans and its ilk should become more popular as guys tire of the same old ai-generated celebrity hentai crap."

You would make it yourself, including any story you want the model to create for the character, and change it endlessly. You won't have to pay anyone, exploit anyone, or take the risks of going to a virus laden porn site. You can do it offline and avoid any recording or tracing of you identity.
Is there any moral angle at all to creating and viewing AI porn. Imagine what depraved stuff will be created and watched. Without real people involved, you can let your evil freak mind go wild.

n.n చెప్పారు...

The C language has a one-to-one proportion, second only to assembly and machine instructions. A very elegant agentic interface for AI, of both species.

bagoh20 చెప్పారు...

I assume coding will soon be unnecessary. If you wanted a time keeping system for your small company for instance. You could just tell your locally installed (off line) AI system to just do it. "Create a prompt screen that records time in and out for each PIN, and store the hours for later retrieval". No coding, just ask it to do it.
This will be the future and the siren call of AI: essentially an all-knowing slave that requires nothing from you in return. There are already millions of free AI downloads available.

Bruce Hayden చెప్పారు...

“It's not English at all. Computer programming languages involve specific and narrow instruction sets, and they were never a human language. The language/symbols are abstraction layers on top of the machine code, but use 100% defined algorithms to convert computer "English" into machine instructions.”

Remember COBOL? Horrid computer language. Way too verbose, and thus labor intensive. Inexact, hard to debug, and extremely easy to screw up. It was an early attempt at an English language computer language.

COBOL would have been fine if it had been tried, it failed, and was replaced, early on. We covered it lightly in a graduate level Programming Languages class in the early 1970s, and even back then, it was considered a failed joke. But then, billions, maybe trillions, of lines of code were written in it, and much of that is still in use today. Remember Y2K? Much of that was a COBOL problem, due to storing dates in English. And I visited the local IRS office sometimes in the last decade, and the guy behind the desk, explained that the many reports that you see that look as if they came off a dot matrix printer were from 40 year old COBOL programs.

Luke Lea చెప్పారు...

I predict (hope?) the world of work will become more tactile: it will involve the manipulation and assembly of three-dimensional physical objects and materials using our eyes, hands, and brains. What the American sociologist Thorsten Veblen called the instinct of workmanship, the pleasures of of making things, even cooking from scratch, will become a bigger part of life than they are now. There will be a much shorter workweek as the benefits of all our new labor-saving technologies begin to be shared more widely. https://www.amazon.com/dp/B00U0C9HKW

Enigma చెప్పారు...

@Luke Lea on a blissful future of tactile pleasure.

Or, we'll end up as murderous rats with nothing to do in our utopias.

https://www.the-scientist.com/universe-25-experiment-69941

Also see Zardoz (1974) for a fictional utopia gone horribly wrong.

Bruce Hayden చెప్పారు...

“It's not English at all. Computer programming languages involve specific and narrow instruction sets, and they were never a human language. The language/symbols are abstraction layers on top of the machine code, but use 100% defined algorithms to convert computer "English" into machine instructions.”

It’s not really computer English. The bulk is quasi mathematical (ignoring earlier abominations like COBOL and RPG). A=B+C Is pretty universal. It’s mostly only control flow that utilizes “English” verbs. Such things as IF, THEN, ELSE, DO, LOOP, UNTIL, etc. They could be translated to another human language, easily enough, and, no doubt, the French tried that. They are merely tokens, little different from +, -, */, =, etc. They can’t be overloaded or redefined.

At least by the 1990s, computer language compilers were being written having a front end and a back end. The front end essentially converted a specific computer language into an intermediate form, and the back end into the instructions for a specific hardware architecture. One of the things that I was involved in was the backend of the Microsoft family of compilers to generate code for the PowerPC architecture. My involvement was in patenting the novel algorithms it utilized, including some really slick SIMD algorithms for its array processing processors.

My view is that the Microsoft compiler was really the only really innovative thing that they ever did. And I expect that Gates was involved. That’s what gave them an “in” with both DR’s DOS and Apple’s Lisa/Mac visual interface.

In any case, it seems superficial to debate whether AI will utilize a computer language like C, or directly generate machine code. My guess is that they will start with the former, then move to a backend compiler interface. That essentially would make their output computer architecture independent or agnostic. And alleviating the computer system lock-ins that sold so many IBM computers.

Josephbleau చెప్పారు...

“ but evolution depends on random experiments.”

Abortion, teen murder rates, and drugs filter the list of subjects that evolution is applied to today. In the past it was filtered by running faster than a dinosaur. Since evolution is based on concentrating the qualities of the fit that survive to reproduce, we must assume that people both have different qualities and that these qualities are inheritable. A random mutation that makes someone better does not lead to the evolution of a better population, unless that higher fitness can be inherited from a parent.

Josephbleau చెప్పారు...

I sometimes use ai to do coding, but as has been mentioned, if I describe the problem the code is already written. I find it most helpful in writing huge SQL or pandas queries where I can specify a large list of variables and the filtering I want done and what to do with missing values, and what I want the final variable names to be.

That helps because I don’t have to type names over and over.

chuck చెప్పారు...

My involvement was in patenting the novel algorithms it utilized

Speaking of software patents, I've been looking for an excuse to link to David Beazley's funny talk on being a technical software expert in a software patent case: Discovering Python.

DINKY DAU 45 చెప్పారు...

our grand kids were taught coding in 4th grade block based language like SCRATCH or BLOCKY and built games and animations(grandkids love anime. Too bad they didnt teach them cursive or how to diagram a sentence(at least moms a teacher and did extra there.More brilliant people who cant tie their shoes, rearing robots! New Age Beatitude= "the meek shal inherit the earth, and the rest of us will leave for space"

Paul చెప్పారు...

I was a computer programmer for over 35 years.... retired... and I don't care what they do... but I can say for sure... COBOL LIVES!! And is still used!!!

Achilles చెప్పారు...

I am working with several of the main models.

Mostly Claude right now. But also Gemini, grok, zookeeper, openrouter cursor and codex.

The AGI models are hilariously/frighteningly good at detecting bugs and fixing breaks.

I am installing whisper dictation right now. Their install.sh file screwed up the repository URL and it installed the wrong venv(virtual environment)

After fixing that it failed because I don't have 2 other dependencies installed. I installed those and it is running through the install script now.

This sort of debugging was a nightmare before. When people fucked up the tag/release of their program like this there was a zero percent chance I would work through this problem.

It took grok less than a minute with each of the error lines. I fixed this in 5 minutes while typing this post.

Fred Drinkwater చెప్పారు...

Bagoh says "Whatever shortcomings AI has now, it will fix them. It never gets tired, bored, or old. It just gets better, and eventually that will require getting us out of the way."

Which is damn close to a line of dialogue from the movie "The Terminator"

Achilles చెప్పారు...

One of the main problems that people have using AGI to develop is that they are not using good coding practices.

If you write solid tests and validation statements against the code you have them write you can develop things accurately and effectively.

Right now I have one agent that is focused on creating the tasks that need to be completed and 2 or 3 agents pulling/executing tasks. I am building an AGI agent harness in markdown. The harness tracks the task scheduler, validation, product requirements, the agent roster, current tasks, runs a working memory log to log session changes. I have context load and unload operations that load the core files from a project file and store the working core meta data in the project data at the end of a session.

This is insane. The agents are like junior devs that never get mad at your or sneak off to watch youtube and they are better at debugging than I am and I was pretty good at integration debugging.

If you can dictate a good specification and build strong validation statements you are a god with these AGI agents.

The problem is that the bottom 90% of coders and new grad CS majors are fucked.

Fred Drinkwater చెప్పారు...

bagoh, re: perfect video, a couple weeks ago I saw a 5 minute video of a "missing scene" from the original Star Wars. (Imperials at Uncle Owen's farm). I watched a few times, and while not perfect, it was damn close. Video and audio both.

Greg The Class Traitor చెప్పారు...

"To outsiders, what programmers are facing can seem richly deserved, and even funny..."
"... American white-collar workers have long fretted that Silicon Valley might one day use A.I. to automate their jobs, but look who got hit first!


What programmers are getting hit by is cheap foreign programmers being brought in on H1B visas, not AI.

I can and do write up prompts for the "AI" to generate code. This has let me produce code in a language I didn't know, learning as it was written.

But if you don't already know how to code, what you produce with AI is crap

Achilles చెప్పారు...

This is a relatively simple prompt and response:

Prompt: speculate on what information needs to be in current_project_core.md and how we need to build the
current_project_core_template.md. It is my intuition that the tasks are loaded into the task scheduler and need to be
stored in the project file but it doesn't seem necessary in the core_project_file. Is there value to keeping this in the
current_project_core_template or should we make that file smaller?


● Read 2 files (ctrl+o to expand)

● Your intuition is right. Let me walk through each section and what's actually earning its place here vs. duplicating
work owned by another CoreSpace file.

Section-by-section analysis

§1 Project Summary (Purpose + Description) — Keep. This is the one place an agent reads to understand what the project
is and why it exists. It directly feeds create_prompt_op context injection. No other CoreSpace file carries this.

§2 MVP Definition (Status + Scope + CAP/ATTR tables) — Keep, but slim down. The MVP scope summary and status belong here
— they tell the agent what "done" looks like at the product level. But the full CAP/ATTR tables with validation doc
links are heavy. The agent only needs the CAP/ATTR tables when running validate_status_op or task_creation_op, both of
which can read the project file directly. A lighter version — just the scope summary, MVP status, and CAP/ATTR counts —
gives the agent enough orientation without duplicating the project file's §2.

§3 Task Management — Remove. This is the one you flagged. The task queue is authoritatively managed by
task_scheduler_core.md. The task lists here are a stale copy the moment a task status changes. Every operation that
touches tasks (choose_task_op, log_task_progress_op, update_task_scheduler_op) reads and writes the scheduler — none of...

Achilles చెప్పారు...

AGI is actually pretty good at developing a good validation.

Humans just don't like writing tests or validations and like those piles of shiny new code.

We are just cooked.

glacial erratic చెప్పారు...

I'm afraid AI generated porn will prove irresistible to men. Imagine, for the first time in history, a man will actually be able to tell a woman to do something - and she will do it! Even if "she" is only bits and bytes.

కామెంట్‌ను పోస్ట్ చేయండి

Please use the comments forum to respond to the post. Don't fight with each other. Be substantive... or interesting... or funny. Comments should go up immediately... unless you're commenting on a post older than 2 days. Then you have to wait for us to moderate you through. It's also possible to get shunted into spam by the machine. We try to keep an eye on that and release the miscaught good stuff. We do delete some comments, but not for viewpoint... for bad faith.