My thoughts and favorite points of someone else’s writing from the web:
The role of developer skills in agentic coding
By: Birgitta Böckeler
I like Birgitta’s categorization of AI errors based on their radius of impact:
- Small impact: slowed down my speed of development and time to commit instead of speeding it up (compared to unassisted coding)
- Medium impact: create friction for the team flow in that iteration
- Large impact: negatively impact long-term maintainability of the code
One engineer with AI can create the tech debt of 50Section titled: One engineer with AI can create the tech debt of 50
- Brute-force fixes instead of root cause analysis
- Verbose and redundant tests
- Lack of reuse
One of my biggest fears of the long term fears with agenic AI is the amount of tech-debt that it can create in the blink of an eye. I’ve played with loveable.dev and they should rename it techdebt-as-a-service.ai.
What can you do to safeguard against AI missteps?Section titled: What can you do to safeguard against AI missteps?
-
Always carefully review AI-generated code. It’s very rare that I do NOT find something to fix or improve.
-
Stop AI coding sessions when you feel overwhelmed by what’s going on. Either revise your prompt and start a new session, or fall back to manual implementation.
-
Stay cautious of “good enough” solutions that were miraculously created in a very short amount of time, but introduce long-term maintenance costs.
