6 AI-assisted coding traps that kill productivity
AI-assisted coding often involves a delicate balancing act between the benefits of automatic code generation and the productivity lost in lengthy code reviews, or when the LLM goes in circles.
To develop a sense of when things are not moving in the right direction, I reflect on past coding sessions and take notes on the productivity traps I had fallen into, to avoid them the next time. This is my summary, so far:
Wasted Time
- When the LLM completes 90% of the task, and it falls into a rabbit hole, repeatedly going in circles without finding a fully working solution for the remaining 10%, again and again and again. It then becomes hard to resist the temptation to try another spin at the slot machine instead of accepting the loss.
- The time spent reviewing both valid and invalid code changes across multiple retries exceeds the time required to simply write the code manually.
Lower Quality
- When the LLM creates code duplication, dead code, unnecessary code, adds unnecessary dependencies, misses tests, or creates tests that give false positives.
- When the LLM makes random changes to the code, unrelated to the current task, that introduce bugs or break existing features.
- When the LLM fakes a successful task completion while changing the test to make it pass instead of fixing the broken implementation, or disabling a Lint/Compiler warning instead of improving the code, etc.
Missed Learning
- Missing out on the learning process when completely delegating the tasks to an LLM can create ‘knowledge debt’ over time. This can make developers less capable and productive and may lead to poorer decision-making and lower productivity.
________________
What productivity traps have you experienced and identified so far?
Useful Heuristics
To move from awareness to improvement, here is my list of heuristics (telling me “what may work when this … is happening”): AI-Assisted coding helpful heuristics
And here, near the bottom, AI-assisted coding: how to ensure real productivity gains? you find the paragraphs:
- Areas of big (productivity) gains
- Limitations (tasks with less chances of productivity gains)
IMPORTANT NOTE: This is a fast-evolving field; the main takeaway here is the questions to ask, how to codify and document the answers that are working well for you today, and the method behind all this.
Develop Technical Excellence that delivers.
See how we can help.
You, your team, your Tech.