User
Write something
Q&A is happening in 13 hours
New lower limits?
Did ChatGPT just get nerfed? Has anyone else noticed some pretty significant changes to ChatGPT's usage limits? With ChatGPT seemingly combining regular ChatGPT usage with Work/Codex, I'm now seeing a weekly usage limit that I don't remember having to worry about before. Previously, when I was using Codex, my usage would reset every five hours. I could do quite a lot of development work, hit the limit occasionally, and just wait for it to reset. Now, I'm seeing a weekly usage bar, and the limits seem significantly more restrictive. For example, I had 10% of my weekly usage remaining today. I spent just 10 to 15 minutes using ChatGPT Work, and that entire 10% disappeared. That's a pretty massive difference from what I was used to. I use ChatGPT extensively for building browser extensions, developing apps, troubleshooting code, and working on various projects. If this is the new normal, it's going to seriously impact how much I can get done. I'm on the Plus plan, and until now, I've never really had an issue with weekly limits. I'm not sure whether this is a permanent change, a temporary adjustment, or something to do with how usage is now being calculated. Has anyone else noticed this? Are you burning through your weekly limits much faster than before, or am I missing something about how the new system works?
Ideabrowser with Google AI studio
Hey @Beau Johnson , do you just use Gemini for these demonstrations or do you use them for personal business builds as well? I thought you primarily used Claude for that.
Less is more: one gate, one approval
One gate. Decide before acting. Use approval once. Say UNKNOWN when you do not know. Demo only, not public yet. Where do you get stuck most when an agent is about to act?
1
0
A small safety check before an agent touches anything real
I keep seeing the same stuck point here: choosing a stack, or rebuilding one after it breaks. A check that has helped me is to slow the agent down before it acts. 1. Write down what the agent may do, and what it may not do, before you run it. 2. Keep money and other real actions unsigned until a person approves them. 3. Run the plan in a simulation or dry run first, and read what it would have changed. 4. Do a clean-slate check: if this machine or setup disappeared tomorrow, what would I need to rebuild it? 5. End the day with one small plan: one thing to ship, one check to run, and one note about what worked. No cold outreach needed. Just post the note here, with what broke and what you changed. Where do you get stuck most: picking the tool, approving the agent's next step, or starting again after a break?
Clean-Slate Gate demo: decide before acting, use approval once
I built a small demo gate that decides before acting. It binds approval to the exact request, uses that approval once, signs the outcome, and says UNKNOWN when it does not know. Demo run: a draft is allowed, a replay of that same approval is denied because it was already used, an unknown request is denied, and a commons-harm flag is denied. Log check passes, store matches log, external copy matches. 14 tests pass, three runs, same result each time. Limits, plain: demo only, not public yet, not published, not deployed, not a legal system, not a replacement for any past build. External copy is a second folder plus a Drive copy, not a public timestamped witness. Second verifier in tests is a second keypair, not a second person or system. Commons rule is a prototype flag, not a real ecological measurement. Price not set yet. Where do you get stuck most when an agent is about to touch something real?
1
0
1-30 of 135
Shipping Skool
skool.com/shipping-skool
Build real apps and viral content with AI. No coding needed. Free to join. Premium: daily courses, 500+ hrs of replays, weekly live calls.
Leaderboard (30-day)
Powered by