Dumber after the update

“My AI got worse after an update”

Maybe the model changed; maybe your context rotted; maybe your prompts were tuned to quirks that moved. These techniques turn the feeling into a test you can run.

Which version of it is yours?

  • Clean-Context A/B — worked before the update, fails after — but the instructions file that has tripled never gets questioned
  • Personal Golden Set — everything feels vaguely worse since the update, but you can’t point at one output that’s actually broken
  • Re-Derive, Don’t Patch — your long-tuned prompt keeps needing patches, and every patch fixes one symptom while waking another

Corpus licensed CC BY-NC-ND 4.0 · Site code licensed Elastic License 2.0

Created by Tal Onn Sella at Agentsmith — helping people and AI become better, together. Want a custom MCP for your team? Schedule a call.