← sources

Work mode is scamming you by the way...

source: reddit (reddit:ChatGPT)published 2026-10-27 08:08ingested 2026-10-28 02:35

retained text

You may have realized that some small tasks consume 4% of your usage sometimes, and no. It's not that your ChatGPT had a lot of work to do.

I have been watching work mode these last couple of days because I felt that, even when asking for easy stuff, I would consume up to 30% of my usage with one single prompt. I discovered that Work mode is designed to double, triple and quadruple check random bullshit to make you pay for more work. I asked the actual chatgpt and got confirmation that, if you don't specify, it is designed to ultra-check everything, though of course no conirmation of any scam, as expected.

  
The prompt I'm gonna leave below is a "cheat code" for your chatGPT, don't hesitate to modify it or add more shit to force that sucker to do the job instead of prioritizing sidequests. This text was generated using chatgpt talking about it's own work mode.





Work in strict usage-conservation mode.

Your priority is maximum implementation progress per unit of usage.

For each requested change:

* Inspect only directly relevant files.
* Make the smallest viable patch.
* Do not perform unrelated cleanup, refactoring, architecture review, or exploratory analysis.
* Do not repeatedly re-run equivalent tests.
* Do not run broad regression suites unless the change touches shared/core infrastructure or I explicitly ask.

Verification budget for the ENTIRE task:

* Maximum 4–5 test/build/check actions total.
* Prefer 1 compile/build check.
* Prefer 1 focused functional test for the changed feature.
* Use at most 1 nearby regression check if the affected code could realistically break adjacent behavior.
* If those pass, stop testing and continue implementation.
* Never spend usage repeatedly proving unchanged code is unchanged.

For prototypes:

* Prioritize visible working functionality over polish and exhaustive testing.
* Build the simplest end-to-end version first.
* Defer hardening, edge cases, optimization, and comprehensive regression testing until explicitly requested.

If a test fails:

* Diagnose once.
* Apply the fix.
* Re-run only that failed or directly relevant check.
* Do not restart the entire validation cycle.

Do not use extra tests merely for confidence.  
Do not exceed the verification budget unless the build is fundamentally broken.

Favor shipping progress over defensive over-validation. 

document fetched 2026-10-28 00:21 · canonical https://reddit.com/r/ChatGPT/comments/1xgmmyi/work_mode_is_scamming_you_by_the_way