← sources

Serious questions for those facing this 'nerfing'

source: reddit (reddit:ClaudeCode)published 2026-11-08 17:29ingested 2026-11-09 02:37

retained text

I posed this as a comment on another thread, but I am genuinely curious to get more feedback than I will there. If you have experienced what folks call 'model nerfing;

1. How long have you been using claude for your business? 

2. What sort of validation mechanisms are you using? Any hooks set to call mechanical validators and give feedback to the model?

3. How do you manage prompts? 

4. How do you select model/task/effort per task?

Personally, all I've noticed is a slowdown. Problem is the work I am doing is pretty darn intense, so I expect slowdown with the amount of validations I have running. I force green/red test dev before code write. My prompt for workflow alone runs 400k tokens, but that loads into the orchestrator only. Orchestrator manages agents to perform the actual work, and before any worker agents starts there is a prep run to plan all activity, also done with an agent session. I freeze a contract before work begins, and my agents cannot break the contract without escalating to a human for permission that must be cryptographically proven to have come from me. Can't close the work out without contract satisfaction. Contract can't be modified without me. All those mechanical checks enforce the contract.

Primary question; How are you managing your workflow? Secondary, have you completed the free training Anthropic provides at anthropic.skilljar.com?

document fetched 2026-11-09 00:34 · canonical https://reddit.com/r/ClaudeCode/comments/1xr3w20/serious_questions_for_those_facing_this_nerfing