IT/Software career thread: Invert binary trees for dollars.

TJT

TJT

Mr. Poopybutthole
<Gold Donor>
47,106
129,407
ShakyJake said:
I’m not saying juniors + AI will automatically outperform great seniors. I’m saying we shouldn’t romanticize the existing senior talent as flawless, because they aren’t.
Click to expand...
The value you provide as a staff/principal engineer is rarely the code you write. It's understanding design patterns and various technologies. When does it make sense to do X instead of Y when both have the same immediate outcome but different longterm implications. You learn this kind of expertise almost explicitly through years of interactions across dozens of projects, use cases, business models, coding frameworks, technologies, and so on.

This will always be where AI fails and where juniors with AI will make AI fail spectacularly. As constraining the AI to your specific use case with the long term implications in mind is just not something a 27 year old engineer will do in most cases. Because they had no idea of the downstream effect of using Y when X actually makes it easier for downstream dependencies for various reasons. But both achieved the same immediate output.

Most of us have spent years dealing with this exact kind of technical debt. All organizations create this and all orgs have it but the difference becomes when you spend 50% of your engineering capacity trying to keep day to day functionality due to tech debt or 10%. That kind of thing.
 
  • 2Like
Reactions: 1 users
ShakyJake

ShakyJake

<Donor>
8,675
21,397
So, some asshole exhausted our credits, and now we're having to code like cavemen for the rest of the month, I guess. Is anyone else having this problem?

As developers, we have zero visibility into how many tokens we're consuming, what the limit is, or if we're approaching it. Not only that, but the credits are grouped at the department level, so a single dev could be hoovering up all of them (which, admittedly, could've been me).
 
  • 3Worf
Reactions: 2 users
Kuro

Kuro

Karazhan Raider
10,710
34,485
ShakyJake said:
So, some asshole exhausted our credits, and now we're having to code like cavemen for the rest of the month, I guess. Is anyone else having this problem?

As developers, we have zero visibility into how many tokens we're consuming, what the limit is, or if we're approaching it. Not only that, but the credits are grouped at the department level, so a single dev could be hoovering up all of them (which, admittedly, could've been me).
Click to expand...
Sounds like you need a 24 year old female college grad with a jira board scheduling twice daily meetings.

To track token usage, but also to reduce productivity by trapping you in endless meetings so that you use fewer tokens
 
  • 2Worf
Reactions: 1 users
Khane

Khane

Got something right about marriage
21,976
16,004
Kuro said:
Sounds like you need a 24 year old female college grad with a jira board scheduling twice daily meetings.

To track token usage, but also to reduce productivity by trapping you in endless meetings so that you use fewer tokens
Click to expand...

Jokes on you bud, she's just using tokens to automate all of that.

Checkmate.

This is actually happening faster than I expected. You know, the whole AI drug dealer dilemma. It was always gonna happen, I just figured it would be after some of the bigger players failed and left the space and the kings were crowned and had their captive audience
 
  • 2Solidarity
  • 1Worf
Reactions: 2 users
Noodleface

Noodleface

A Mod Real Quick
39,728
18,395
ShakyJake said:
So, some asshole exhausted our credits, and now we're having to code like cavemen for the rest of the month, I guess. Is anyone else having this problem?

As developers, we have zero visibility into how many tokens we're consuming, what the limit is, or if we're approaching it. Not only that, but the credits are grouped at the department level, so a single dev could be hoovering up all of them (which, admittedly, could've been me).
Click to expand...
Our org is given a pool of credits for the month and it is then doled out to users based on what role they're in. We are also able to request increases where they take from those not using as many.

Not having a guard rail is ridiculous lol.
 
Khane

Khane

Got something right about marriage
21,976
16,004
Most companies are/were forcing their entire org to use AI for EVERYTHING. So how's it gonna go when they finally wake up to what that actually costs and try to put a governor on it?

"Why are you guys using so many tokens?!"
"Because you literally threatened my job if I didn't"
 
Last edited:
  • 1Truth!
  • 1Like
Reactions: 1 users
ShakyJake

ShakyJake

<Donor>
8,675
21,397
Khane said:
Most companies are/were forcing their entire org to use AI for EVERYTHING. So how's it gonna go when they finally wake up to what that actually costs and try to put a governor on it?

"Why are you guys using so many token?!"
"Because you literally threatened my job if I didn't"
Click to expand...
For our team, most individuals were unaware of AI coding agents until quite recently. There were no mandates requiring anyone to use them. Instead, only a small number of people expressed curiosity and experimented with the IDE plugin. Then, a few discovered the CLI and enthusiastically began creating skills and other tools. And now here we are.
 
TJT

TJT

Mr. Poopybutthole
<Gold Donor>
47,106
129,407
ShakyJake said:
So, some asshole exhausted our credits, and now we're having to code like cavemen for the rest of the month, I guess. Is anyone else having this problem?

As developers, we have zero visibility into how many tokens we're consuming, what the limit is, or if we're approaching it. Not only that, but the credits are grouped at the department level, so a single dev could be hoovering up all of them (which, admittedly, could've been me).
Click to expand...
This will be THE one and only problem in the coming years.

Welcome to the Token Economy broski!
 
moonarchia

moonarchia

The Scientific Shitlord
<Bronze Donator>
31,221
61,572
TJT said:
This will be THE one and only problem in the coming years.

Welcome to the Token Economy broski!
Click to expand...
They have unironically created a KPI for coding ability and productivity. Now AI can start replacing management as well.
 
Phazael

Phazael

Confirmed Beta Shitlord, Fat Bastard
<Gold Donor>
16,452
37,137
Khane said:
Jokes on you bud, she's just using tokens to automate all of that.

Checkmate.

This is actually happening faster than I expected. You know, the whole AI drug dealer dilemma. It was always gonna happen, I just figured it would be after some of the bigger players failed and left the space and the kings were crowned and had their captive audience
Click to expand...
Money is drying up so they are accelerating the time table to get out before the bubble pops. Out company is going to escape this trap thanks to its sheer ineptitude and inability to adopt anything new in a timely manner, which is equally hilarious to me. And yea the MBA non technical fucktards gobbled up most of our tokens purely to have AI crank out their buzzword laden business plans that never materialize (and are probably most about getting more AI, LOL) and the actual dev ops and technical people got stiffed. I was actually in the process of automating a bunch of our workflows for end user service when we got told to hold off because there were higher priority people getting to burn them. Turned out it was a bunch of middle managers having AI do useless administrative tasks for them mostly, when we looked into the metrics.
 
  • 2Like
  • 1Worf
Reactions: 2 users
Noodleface

Noodleface

A Mod Real Quick
39,728
18,395
Khane said:
Most companies are/were forcing their entire org to use AI for EVERYTHING. So how's it gonna go when they finally wake up to what that actually costs and try to put a governor on it?

"Why are you guys using so many tokens?!"
"Because you literally threatened my job if I didn't"
Click to expand...
Already happened here. We had a meeting where they said "ok I know we said use it for everything but now we want you to use them smarter"
 
TJT

TJT

Mr. Poopybutthole
<Gold Donor>
47,106
129,407
Speaking of using them smarter I can been using Rules | Cursor Docs because these assholes explained that this was the correct way to use Cursor to the best and most efficient effect.

However, clear project level rules for things inflates your contexts and thus your tokens to unmanageable levels really fast. So we instead have to pivot to using skills which are just a lot different.
 
Control

Control

Golden Baronet of the Realm
5,746
15,995
Phazael said:
And yea the MBA non technical fucktards gobbled up most of our tokens purely to have AI crank out their buzzword laden business plans
Click to expand...
I've found that creating reasonable text content is far more token hungry than coding tasks.so the people using it for emails and plans/reports that no one will ever read just makes me lol. On my home projects at least, that might change if the codebases were massive, but still, generating text is the only thing that I can easily cap my limits doing.
 
Khane

Khane

Got something right about marriage
21,976
16,004
Control said:
I've found that creating reasonable text content is far more token hungry than coding tasks.so the people using it for emails and plans/reports that no one will ever read just makes me lol. On my home projects at least, that might change if the codebases were massive, but still, generating text is the only thing that I can easily cap my limits doing.
Click to expand...

What are you using that this is the case? It's the opposite with ChatGPT and Claude.

Claude Code GUZZLES tokens but Claude Cowork, and especially Claude Chat use very little in comparison. In fact, for most coding scenarios I reverted entirely away from Claude Code because I don't actually need its functionality in most cases. I just want to give it small, discrete tasks. I rewrote my entire resume and LinkedIn profile with Claude Chat and it used basically nothing as far as tokenization.
 
Control

Control

Golden Baronet of the Realm
5,746
15,995
Khane said:
What are you using that this is the case? It's the opposite with ChatGPT and Claude.

Claude Code GUZZLES tokens but Claude Cowork, and especially Claude Chat use very little in comparison. In fact, for most coding scenarios I reverted entirely away from Claude Code because I don't actually need its functionality in most cases. I just want to give it small, discrete tasks. I rewrote my entire resume and LinkedIn profile with Claude Chat and it used basically nothing as far as tokenization.
Click to expand...
claude code for both, but to be fair, to try and improve the writing, I'm using multiple revisions and audits (like 10+), never tested to check the token difference between the raw text generation and the fully revised versions though. of course, the coding has it's own audits and reviews too, but then the amount of code I can have it write ends up being limited by my ability to test/verify. But with text, I can just have it write infinite whatever. I realized yesterday that I basically hadn't touched it this week, and credits were resetting at the end of the day, so I basically had it write articles to burn tokens. Capped 3 different 5 hour windows no problem on max20 (which is apparently like 65%-70% of a weekly quota).
 
Khane

Khane

Got something right about marriage
21,976
16,004
Why use Claude code for that? And are you using Opus, Sonnet or Haiku? You'll use, very literally, like 1/100th the tokens for that kind of task using Claude Chat with Haiku
 
Control

Control

Golden Baronet of the Realm
5,746
15,995
Khane said:
Why use Claude code for that? And are you using Opus, Sonnet or Haiku? You'll use, very literally, like 1/100th the tokens for that kind of task using Claude Chat with Haiku
Click to expand...
Opus, well, third session yesterday was Fable. I probably should run some tests to see how the different models compare, but I was trying to optimize for quality/accuracy, not tokens. Using claude code because it's not just random text, it's filling in a content framework for a site based on notes, research, voice definitions, etc. (basically an ai slop site as an experiment to see how good it is). So I basically turn it loose on writing when I have tokens to burn at the end of a week.
 
ShakyJake

ShakyJake

<Donor>
8,675
21,397
That's the problem I have is that there's no good way to judge which model is appropriate for a task other than feels. So I just crank it up to the max.
 
Khane

Khane

Got something right about marriage
21,976
16,004
That is the grift. Claude Code (and all the different models on the Anthropic platform) are the most egregious out of all of them from what I've experienced first hand. And they can get away with that, because as of right now, Anthropic has the best models.

AI is packaged and sold as automation on steroids, as smarter than humans, as the end of us needing to think. And curiously, at the same time, it's not smart enough to context switch on its own or choose the correct model for the task you are presenting it with. Why exactly do we need different models anyway? Have you heard an actually reasonable explanation for that when the answer has to contend directly with the use cases for AI in the first place?

Really weird isn't it? It's like they understand human psychology, know people don't want to have to bother themselves with details and will let it run wild because its just easier. And they also understand, psychologically, people won't question the obvious dichotomy because its making their lives easier.
 
You must log in or register to reply here.