This talk was recorded at NDC London in London, England. #ndclondon  #ndcconferences #developer #softwaredeveloper    

Attend the next NDC conference near you: 
https://ndcconferences.com
https://ndclondon.com/

Subscribe to our YouTube channel and learn every day:   
/        @NDC 

Follow our Social Media!

https://www.facebook.com/ndcconferences
https://twitter.com/NDC_Conferences
https://www.instagram.com/ndc_conferences/

#ai #machinelearning 

AI pair programming feels magical, until you realize your coding partner has a habit of confidently repeating the same mistakes.

A small bug slips into a Copilot suggestion, you accept it, and later the AI uses that flawed code as context. Soon you’re trapped in an AI error loop: a cycle where errors don’t just persist, they multiply.

This talk explores how error loops manifest in AI-assisted development: circular bug fixes, brittle test generation that always passes, boilerplate repetition, and the trust bias that convinces developers to accept faulty suggestions. We’ll break down real-world examples of AI “help” that actually locked teams into cycles of bad code.

Most importantly, you’ll learn strategies to break the loop. From static analysis and diversified prompts to human-in-the-loop reviews and “reset the loop” techniques, we’ll look at practical ways to turn AI into a reliable coding partner, not a parrot of your mistakes.

Key Takeaways:

- Identify AI error loops in coding workflows and why they’re harder to spot than traditional bugs.

- Understand the risks of circular fixes, repetitive tests, and confidence bias in AI pair programming.

- Apply loop breakers like static analysis, prompt resets, and review culture.

- Reframe AI as a powerful but fallible junior dev: fast, useful, but in need of supervision.

NDC Conferences

A conference talk exploring the 'AI error loop' phenomenon where developers using AI coding assistants like GitHub Copilot or ChatGPT get trapped in cycles of bug fixes that introduce new bugs. The core problem is context drift: accepted AI suggestions become future context, causing models to optimize for consistency over correctness. Key concepts covered include context engineering vs. prompt engineering, context rot (performance degradation from too much irrelevant context), and three strategies to break the loop: resetting context frequently, anchoring the model to an external source of truth (e.g., a GitHub Copilot instructions file), and maintaining human judgment to verify AI outputs. The central warning is that AI models fail silently by agreeing rather than loudly by refusing.

Trapped in the AI Error Loop: Why Smart Systems Keep Making Dumb Mistakes - Omphile Matheolane