Switching AI Models When Facing Persistent Issues

Discussions
1

When you’re struggling to solve a problem with an Agent for a long time, do you switch to a different model, or do you stick with one particular favorite?

2

Depends on the model and task, there are likely other solutions to model switching as well.

  • Never let the model go on in circles. Identify what prompt might have caused this issue, then adjust the prompt or add more info before the and restore that code change version.
  • Often its caused by long context or info in context polluting / confusing AI, new chat likely helps.
  • If its just a context size issue switching to a model with larger context helps.
  • Updating rules with clearer process or simplifying steps helps often.
  • Well prepared tasks (in .md files) with detailed process steps (Find RIPER5, BMAD, Plan vs Act in Forum) can prevent such cases.
  • Etc.

Could you explain a bit more what kind of “struggling to solve a problem” this would be as the solutions depend a lot on the issue.

3

When you need the code to do something and it doesn’t do it. And neither you nor the Agent can quickly figure out what’s wrong.

4

Right, this is tricky but some of the solutions I posted above still could apply (specially context pollution trips up AI).

  • Mostly I try to prevent this by asking AI to prepare steps and review what needs to be changed, so there are less such cases.
  • If this still happens one of the following was so far often the cause
    • a) I misunderstood a good suggestion from AI and overrode its plan
    • b) AI missed a point I made to prevent such a case
    • c) It wasnt predictable that the issue would happen.
  • So if its case c) or something much harder to solve I use a different AI in Ask mode to research possible solutions or what alternative approaches exist.
  • Altarnatively I ask the same or other AI model in new chat to review the latest changes or a specific feature and suggest improvements to prevent the issues.

For me switching the AI model within the same chat is mostly not a good solution as my prompts are tailored for the AI that I am usually using and wont give as great results on other models.