My ChatGPT subscription ends tomorrow, all I got were 3 resets that just extended the weeks. i never received my original weekly usage. Now weekly is at 0% with no credits left. @thsottiaux where is my regular usage? This reset game feels like not worth it at all.

Astra: stronger computer use and long multi-step work. Phased rollout, extra safety checks Fable 5.1: better coding and 75% cheaper cache reads for agents that remember context Gemini 3.8 Flash: fast coding workhorse, plus a gated Cyber variant Qwen and Muse Spark: good enough at a price that makes loyalty look expensive

New model tests that actually teaches something Giving it a work that we already do Cleaning a messy sheet Turning 3 docs into a client brief Find why a script fails Prepares for Monday from your calendar. If it cannot finish that, the launch does not matter to you yet.

GPT-6 Astra is now available to plus subscribers

I started asking AI for options not answers, then I chose. That tiny change is when it stopped feeling like a toy and started feeling like a teammate.

Don't automate because AI can do it Automate it because: Is it happens repeatedly, the inputs are available, done can be defined, mistakes can be detected, the result has measurable value. That filter eliminates a lot of impressive but useless AI projects.

When to use an agent vs a script: Use a SCRIPT when: ~ Same input - same output every time ~ Speed matters more than flexibility ~ You can write the logic in 10 lines Use an AGENT when: ~ The task is fuzzy (make this better) ~ You'd need 100 if-statements ~ You're okay with 90% accuracy Save this matrix

I Wrote a message to someone the same way I prompt now Short, One job, No extra context, That was the quiet shift. AI didn't just change my workflow. It changed how I talk. Not sure thats better. Just noticing it. 😄

Before connecting an AI agent to a tool, define its failure policy. Not just: When this happens ~ do that Also define: If data is missing ~ stop. If confidence is low ~ ask. If the tool fails ~ retry once. If the action is irreversible ~ require approval. Good automation is mostly designing what happens when things don't go perfectly.

The most expensive AI habit I see isn't the subscriptions It's the pile of tools you never actually use. The story we tell ourselves is more apps = more leverage So we adding new one every week and call it progress What really happens, we spend the week logging in, moving files, and explaining the same job to five different chats. Still ships nothing A stack that really works is boring One model, One notes app, One task we already do every day That's enough to feel it If a tool does not kill a job we already hate, it's just decoration Be honest what are you still collecting that you don't open?

The prompt format that still wins: Job title, Hard constraint, Definition of done, you are a staff engineer, Do not invent APIs. Done means the tests pass. Role + guardrail + finish line beats a paragraph of vibes.

A simple rule for building AI agents Don't give the agent more tools Give it better state, After every meaningful action, store What happened What changed What's still missing What should happen next Agents often fail because they lose the thread, not because the model isn't smart enough.

Tightening the input from Day 1 habit compounds harder than any model upgrade.