Comment

Epp@lemmus.org ⁨3⁩ ⁨days⁩ ago

The future is in agentic AI with a single developer for code review. Management tells the developer what they want, developer engineers the prompt, gives it to the AI agent that has complete access to the relevant projects and DB schema. It generates a change log, and the developer reviews it, asking for changes as needed.

Huge teams are about to be consolidated, with a huge inflow of software engineers into the unemployment bin, and entire downstream economies are going to collapse from the resulting unemployment of previously high paying careers. We need Universal Basic Income yesterday.

source

Sort:hotnew top

mrgoosmoos@lemmy.ca ⁨3⁩ ⁨days⁩ ago
constantly reviewing low quality work. kill me now.

what godawful boring job that would be. I’d have absolutely no motivation to do it well

source
- Epp@lemmus.org ⁨3⁩ ⁨days⁩ ago
  It’s only low quality until it isn’t. Have you used Gemin 3.1 Proi lately? Anthropic’s Opus 4.6?
  
  Everything looks like low-quality crap when you only use the free models from Microsoft and OpenAI. But I suspect you haven’t utilized the paid, professional models if you hold that opinion.
  
  source
sacredfire@programming.dev ⁨3⁩ ⁨days⁩ ago
The near future? How is this a sustainable business model for any business? You just need one developer and “agentic” AI to build anything, how do you differentiate yourself?

But before that problem, I don’t see the current tools anywhere near able to deliver on the hype. They are incredible and they have plenty of use cases, but for anything non trivial it feels like it’s more work fixing the errors they create than just doing it myself. I think I’d kill myself if I had to review and fix multiple agents worth of indecipherable code.

All that being said, everyone still might get laid off! It doesn’t have to be good to crash the market.

source
- InputZero@lemmy.world ⁨3⁩ ⁨days⁩ ago
  You differentiate yourself by being first. That’s partly why OpenAI and Sam Altmann are so fixated on bringing general AI to light. They know that if AGI is possible that the first one to reach it will see all the benefits. Second place gets nothing. Unfortunately it’s becoming more and more obvious that AGI is a dream and not actually possible.
  
  source
- Epp@lemmus.org ⁨3⁩ ⁨days⁩ ago
  What tools do you have experience with? Just the free ones? The crap from Microslop or OpenAI? I sincerely believe you’d change your opinion if you were using the professional products from companies that have created working models, but I imagine most people only have limited experience with those models, if any. Most use ChatGPT or Copilot and surmise it’s all inaccurate crap. They’d be right from that limited sample, but wrong about the market at large.
  
  source
Epp@lemmus.org ⁨3⁩ ⁨days⁩ ago
Sorry, missed this was in ShitPost. Please, allow me to revise: Dang clankers, Deytükurjerbs!

source
vane@lemmy.world ⁨3⁩ ⁨days⁩ ago
As always all those claims have very big gaps: relevant projects, db schema, management know what they want. Dude there are multi billion dollar companies that take hundreds of millions of dollars to figure out what management want and you think they will replace it with single prompt ? It’s like talking to 5 year old.

source
- Epp@lemmus.org ⁨3⁩ ⁨days⁩ ago
  That’s why you still need one skilled developer per project as the middle man, for requirement elicitation. You don’t have to believe me, no skin off my nose, but I’m with an organization that’s making it work exactly as described. Months of work done in weeks instead.
  
  source
  - vane@lemmy.world ⁨3⁩ ⁨days⁩ ago
    8 weeks are still months. If your code looks like your math then RIP, LLM junkie.
    
    source
    Epp@lemmus.org ⁨3⁩ ⁨days⁩ ago
    I don’t know why someone down voted you, but I also don’t know what you mean by my math. Which math would that be? I was using AI before LLMs ever even hit the scene.
    
    source
vala@lemmy.dbzer0.com ⁨3⁩ ⁨days⁩ ago
It’s somewhat naïve to think we need a human in the loop, but only a single human.

Sounds like your org mostly builds CRUD apps. Maybe this is how it will work in that space but I don’t see it happening in general.

source