• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
Technology

AI Code Tools Drive Up Maintenance Costs, Developers Debate Risks

Replies debate the claim and whether models now reduce technical debt faster.

ZE
GA
D4
8 Sources, 21d ago, first seen 21d ago

TLDR

A post from the zerohedge account states that maintenance costs for AI-generated code have jumped 300% in 18 months, code duplication is up 48%, and refactoring has dropped 60%. It attributes the figures to Checkmarx and warns of hidden technical debt in companies that adopted AI coding tools. Several replies push back: one developer says models have improved enough to cut cleanup time compared with 18 months ago; another reports using AI agents to refactor legacy code and improve documentation. A third calls the post directionally correct but wants the underlying Checkmarx data.

Combined views

338.8K

8 Sources, first seen 21d ago

4.6K likes270 comments848 saves514 reposts

Combined views

338.8K

8 Sources, first seen 21d ago

4.6K likes270 comments848 saves514 reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

8 Sources

@zerohedgeMaintenance costs for AI-generated code have jumped 300% in 18 months, code duplication is up 48% and refactoring has dropped 60%. The result could be a massive technical debt problem hiding inside companies that initially saw AI coding as a productivity win: Checkmarx
@ArgumentsGod@zerohedge I've thought of this. But. Models have also gotten massively better. It's easier now to clean up code than it was half a year ago, to say nothing of 18 months. Tech debt is probably going down in terms of time and money required to make it right again.
@ExposeDarkDeeds@zerohedge Somehow still better than offshore workers.
@mccryan@zerohedge It is the absolute best refactoring tool if you control it properly. An LLM's natural disposition (every single model I've seen) is to staple things on inelegantly. It's testament to its ability to track the complexity of state but it looks horrifying to us.
@blafasel42Interesting. Because it has never been easier to remove technical debt and improve code quality. My teams are doing this all the time: Running refafactoring- and modernization sweeps with AI Agents over human-produced legacy code to make it better. Improving documentation and test coverage in the process. My impression of the effects of Agentic coding is exaclty the opposite of what you are describing. It might be a question of how naive the approach to agentic coding is...
@AustenThis is 100% true. There’s way, way more technical debt now. The good news is the cost of dealing with technical debt is a fraction of what it used to be (per amount of technical debt.) The average company should (and will) have way, way more technical debt.
@bendee983I would like to see the actual data and evidence on this from @Checkmarx, but directionally, I find it to be correct. What I'm seeing is a lot of companies recklessly using AI coding tools to push code into production. In many cases, the people who are tasked with leading the software engineering efforts do not have proper experience in coding and are just burning AI tokens and using Claude Code Max plans to "let the agents cook." In instances where I had the chance to look at the code, I have to say it's messy. And the AI agents confuse themselves with their own convoluted code. We need more discipline in the field.
@kritikakodesAI is only good for founders and builders, But if you want a job, you still need to learn how to code.
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    8 Sources

    @zerohedgeMaintenance costs for AI-generated code have jumped 300% in 18 months, code duplication is up 48% and refactoring has dropped 60%. The result could be a massive technical debt problem hiding inside companies that initially saw AI coding as a productivity win: Checkmarx
    @ArgumentsGod@zerohedge I've thought of this. But. Models have also gotten massively better. It's easier now to clean up code than it was half a year ago, to say nothing of 18 months. Tech debt is probably going down in terms of time and money required to make it right again.
    @ExposeDarkDeeds@zerohedge Somehow still better than offshore workers.
    @mccryan@zerohedge It is the absolute best refactoring tool if you control it properly. An LLM's natural disposition (every single model I've seen) is to staple things on inelegantly. It's testament to its ability to track the complexity of state but it looks horrifying to us.
    @blafasel42Interesting. Because it has never been easier to remove technical debt and improve code quality. My teams are doing this all the time: Running refafactoring- and modernization sweeps with AI Agents over human-produced legacy code to make it better. Improving documentation and test coverage in the process. My impression of the effects of Agentic coding is exaclty the opposite of what you are describing. It might be a question of how naive the approach to agentic coding is...
    @AustenThis is 100% true. There’s way, way more technical debt now. The good news is the cost of dealing with technical debt is a fraction of what it used to be (per amount of technical debt.) The average company should (and will) have way, way more technical debt.
    @bendee983I would like to see the actual data and evidence on this from @Checkmarx, but directionally, I find it to be correct. What I'm seeing is a lot of companies recklessly using AI coding tools to push code into production. In many cases, the people who are tasked with leading the software engineering efforts do not have proper experience in coding and are just burning AI tokens and using Claude Code Max plans to "let the agents cook." In instances where I had the chance to look at the code, I have to say it's messy. And the AI agents confuse themselves with their own convoluted code. We need more discipline in the field.
    @kritikakodesAI is only good for founders and builders, But if you want a job, you still need to learn how to code.
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet